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/init.h>
46 #include <linux/profile.h>
47 #include <linux/cpu.h>
48 #include <linux/security.h>
49 #include <linux/percpu.h>
50 #include <linux/rtc.h>
51 #include <linux/jiffies.h>
52 #include <linux/posix-timers.h>
53 #include <linux/irq.h>
54 #include <linux/delay.h>
55 #include <linux/irq_work.h>
56 #include <asm/trace.h>
57
58 #include <asm/io.h>
59 #include <asm/processor.h>
60 #include <asm/nvram.h>
61 #include <asm/cache.h>
62 #include <asm/machdep.h>
63 #include <asm/uaccess.h>
64 #include <asm/time.h>
65 #include <asm/prom.h>
66 #include <asm/irq.h>
67 #include <asm/div64.h>
68 #include <asm/smp.h>
69 #include <asm/vdso_datapage.h>
70 #include <asm/firmware.h>
71 #include <asm/cputime.h>
72
73 /* powerpc clocksource/clockevent code */
74
75 #include <linux/clockchips.h>
76 #include <linux/timekeeper_internal.h>
77
78 static cycle_t rtc_read(struct clocksource *);
79 static struct clocksource clocksource_rtc = {
80 .name = "rtc",
81 .rating = 400,
82 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
83 .mask = CLOCKSOURCE_MASK(64),
84 .read = rtc_read,
85 };
86
87 static cycle_t timebase_read(struct clocksource *);
88 static struct clocksource clocksource_timebase = {
89 .name = "timebase",
90 .rating = 400,
91 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
92 .mask = CLOCKSOURCE_MASK(64),
93 .read = timebase_read,
94 };
95
96 #define DECREMENTER_MAX 0x7fffffff
97
98 static int decrementer_set_next_event(unsigned long evt,
99 struct clock_event_device *dev);
100 static void decrementer_set_mode(enum clock_event_mode mode,
101 struct clock_event_device *dev);
102
103 struct clock_event_device decrementer_clockevent = {
104 .name = "decrementer",
105 .rating = 200,
106 .irq = 0,
107 .set_next_event = decrementer_set_next_event,
108 .set_mode = decrementer_set_mode,
109 .features = CLOCK_EVT_FEAT_ONESHOT,
110 };
111 EXPORT_SYMBOL(decrementer_clockevent);
112
113 DEFINE_PER_CPU(u64, decrementers_next_tb);
114 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
115
116 #define XSEC_PER_SEC (1024*1024)
117
118 #ifdef CONFIG_PPC64
119 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
120 #else
121 /* compute ((xsec << 12) * max) >> 32 */
122 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
123 #endif
124
125 unsigned long tb_ticks_per_jiffy;
126 unsigned long tb_ticks_per_usec = 100; /* sane default */
127 EXPORT_SYMBOL(tb_ticks_per_usec);
128 unsigned long tb_ticks_per_sec;
129 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
130
131 DEFINE_SPINLOCK(rtc_lock);
132 EXPORT_SYMBOL_GPL(rtc_lock);
133
134 static u64 tb_to_ns_scale __read_mostly;
135 static unsigned tb_to_ns_shift __read_mostly;
136 static u64 boot_tb __read_mostly;
137
138 extern struct timezone sys_tz;
139 static long timezone_offset;
140
141 unsigned long ppc_proc_freq;
142 EXPORT_SYMBOL_GPL(ppc_proc_freq);
143 unsigned long ppc_tb_freq;
144 EXPORT_SYMBOL_GPL(ppc_tb_freq);
145
146 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
147 /*
148 * Factors for converting from cputime_t (timebase ticks) to
149 * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
150 * These are all stored as 0.64 fixed-point binary fractions.
151 */
152 u64 __cputime_jiffies_factor;
153 EXPORT_SYMBOL(__cputime_jiffies_factor);
154 u64 __cputime_usec_factor;
155 EXPORT_SYMBOL(__cputime_usec_factor);
156 u64 __cputime_sec_factor;
157 EXPORT_SYMBOL(__cputime_sec_factor);
158 u64 __cputime_clockt_factor;
159 EXPORT_SYMBOL(__cputime_clockt_factor);
160 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
161 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
162
163 cputime_t cputime_one_jiffy;
164
165 void (*dtl_consumer)(struct dtl_entry *, u64);
166
calc_cputime_factors(void)167 static void calc_cputime_factors(void)
168 {
169 struct div_result res;
170
171 div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
172 __cputime_jiffies_factor = res.result_low;
173 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
174 __cputime_usec_factor = res.result_low;
175 div128_by_32(1, 0, tb_ticks_per_sec, &res);
176 __cputime_sec_factor = res.result_low;
177 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
178 __cputime_clockt_factor = res.result_low;
179 }
180
181 /*
182 * Read the SPURR on systems that have it, otherwise the PURR,
183 * or if that doesn't exist return the timebase value passed in.
184 */
read_spurr(u64 tb)185 static u64 read_spurr(u64 tb)
186 {
187 if (cpu_has_feature(CPU_FTR_SPURR))
188 return mfspr(SPRN_SPURR);
189 if (cpu_has_feature(CPU_FTR_PURR))
190 return mfspr(SPRN_PURR);
191 return tb;
192 }
193
194 #ifdef CONFIG_PPC_SPLPAR
195
196 /*
197 * Scan the dispatch trace log and count up the stolen time.
198 * Should be called with interrupts disabled.
199 */
scan_dispatch_log(u64 stop_tb)200 static u64 scan_dispatch_log(u64 stop_tb)
201 {
202 u64 i = local_paca->dtl_ridx;
203 struct dtl_entry *dtl = local_paca->dtl_curr;
204 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
205 struct lppaca *vpa = local_paca->lppaca_ptr;
206 u64 tb_delta;
207 u64 stolen = 0;
208 u64 dtb;
209
210 if (!dtl)
211 return 0;
212
213 if (i == vpa->dtl_idx)
214 return 0;
215 while (i < vpa->dtl_idx) {
216 if (dtl_consumer)
217 dtl_consumer(dtl, i);
218 dtb = dtl->timebase;
219 tb_delta = dtl->enqueue_to_dispatch_time +
220 dtl->ready_to_enqueue_time;
221 barrier();
222 if (i + N_DISPATCH_LOG < vpa->dtl_idx) {
223 /* buffer has overflowed */
224 i = vpa->dtl_idx - N_DISPATCH_LOG;
225 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
226 continue;
227 }
228 if (dtb > stop_tb)
229 break;
230 stolen += tb_delta;
231 ++i;
232 ++dtl;
233 if (dtl == dtl_end)
234 dtl = local_paca->dispatch_log;
235 }
236 local_paca->dtl_ridx = i;
237 local_paca->dtl_curr = dtl;
238 return stolen;
239 }
240
241 /*
242 * Accumulate stolen time by scanning the dispatch trace log.
243 * Called on entry from user mode.
244 */
accumulate_stolen_time(void)245 void accumulate_stolen_time(void)
246 {
247 u64 sst, ust;
248
249 u8 save_soft_enabled = local_paca->soft_enabled;
250
251 /* We are called early in the exception entry, before
252 * soft/hard_enabled are sync'ed to the expected state
253 * for the exception. We are hard disabled but the PACA
254 * needs to reflect that so various debug stuff doesn't
255 * complain
256 */
257 local_paca->soft_enabled = 0;
258
259 sst = scan_dispatch_log(local_paca->starttime_user);
260 ust = scan_dispatch_log(local_paca->starttime);
261 local_paca->system_time -= sst;
262 local_paca->user_time -= ust;
263 local_paca->stolen_time += ust + sst;
264
265 local_paca->soft_enabled = save_soft_enabled;
266 }
267
calculate_stolen_time(u64 stop_tb)268 static inline u64 calculate_stolen_time(u64 stop_tb)
269 {
270 u64 stolen = 0;
271
272 if (get_paca()->dtl_ridx != get_paca()->lppaca_ptr->dtl_idx) {
273 stolen = scan_dispatch_log(stop_tb);
274 get_paca()->system_time -= stolen;
275 }
276
277 stolen += get_paca()->stolen_time;
278 get_paca()->stolen_time = 0;
279 return stolen;
280 }
281
282 #else /* CONFIG_PPC_SPLPAR */
calculate_stolen_time(u64 stop_tb)283 static inline u64 calculate_stolen_time(u64 stop_tb)
284 {
285 return 0;
286 }
287
288 #endif /* CONFIG_PPC_SPLPAR */
289
290 /*
291 * Account time for a transition between system, hard irq
292 * or soft irq state.
293 */
vtime_delta(struct task_struct * tsk,u64 * sys_scaled,u64 * stolen)294 static u64 vtime_delta(struct task_struct *tsk,
295 u64 *sys_scaled, u64 *stolen)
296 {
297 u64 now, nowscaled, deltascaled;
298 u64 udelta, delta, user_scaled;
299
300 WARN_ON_ONCE(!irqs_disabled());
301
302 now = mftb();
303 nowscaled = read_spurr(now);
304 get_paca()->system_time += now - get_paca()->starttime;
305 get_paca()->starttime = now;
306 deltascaled = nowscaled - get_paca()->startspurr;
307 get_paca()->startspurr = nowscaled;
308
309 *stolen = calculate_stolen_time(now);
310
311 delta = get_paca()->system_time;
312 get_paca()->system_time = 0;
313 udelta = get_paca()->user_time - get_paca()->utime_sspurr;
314 get_paca()->utime_sspurr = get_paca()->user_time;
315
316 /*
317 * Because we don't read the SPURR on every kernel entry/exit,
318 * deltascaled includes both user and system SPURR ticks.
319 * Apportion these ticks to system SPURR ticks and user
320 * SPURR ticks in the same ratio as the system time (delta)
321 * and user time (udelta) values obtained from the timebase
322 * over the same interval. The system ticks get accounted here;
323 * the user ticks get saved up in paca->user_time_scaled to be
324 * used by account_process_tick.
325 */
326 *sys_scaled = delta;
327 user_scaled = udelta;
328 if (deltascaled != delta + udelta) {
329 if (udelta) {
330 *sys_scaled = deltascaled * delta / (delta + udelta);
331 user_scaled = deltascaled - *sys_scaled;
332 } else {
333 *sys_scaled = deltascaled;
334 }
335 }
336 get_paca()->user_time_scaled += user_scaled;
337
338 return delta;
339 }
340
vtime_account_system(struct task_struct * tsk)341 void vtime_account_system(struct task_struct *tsk)
342 {
343 u64 delta, sys_scaled, stolen;
344
345 delta = vtime_delta(tsk, &sys_scaled, &stolen);
346 account_system_time(tsk, 0, delta, sys_scaled);
347 if (stolen)
348 account_steal_time(stolen);
349 }
350 EXPORT_SYMBOL_GPL(vtime_account_system);
351
vtime_account_idle(struct task_struct * tsk)352 void vtime_account_idle(struct task_struct *tsk)
353 {
354 u64 delta, sys_scaled, stolen;
355
356 delta = vtime_delta(tsk, &sys_scaled, &stolen);
357 account_idle_time(delta + stolen);
358 }
359
360 /*
361 * Transfer the user time accumulated in the paca
362 * by the exception entry and exit code to the generic
363 * process user time records.
364 * Must be called with interrupts disabled.
365 * Assumes that vtime_account_system/idle() has been called
366 * recently (i.e. since the last entry from usermode) so that
367 * get_paca()->user_time_scaled is up to date.
368 */
vtime_account_user(struct task_struct * tsk)369 void vtime_account_user(struct task_struct *tsk)
370 {
371 cputime_t utime, utimescaled;
372
373 utime = get_paca()->user_time;
374 utimescaled = get_paca()->user_time_scaled;
375 get_paca()->user_time = 0;
376 get_paca()->user_time_scaled = 0;
377 get_paca()->utime_sspurr = 0;
378 account_user_time(tsk, utime, utimescaled);
379 }
380
381 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
382 #define calc_cputime_factors()
383 #endif
384
__delay(unsigned long loops)385 void __delay(unsigned long loops)
386 {
387 unsigned long start;
388 int diff;
389
390 if (__USE_RTC()) {
391 start = get_rtcl();
392 do {
393 /* the RTCL register wraps at 1000000000 */
394 diff = get_rtcl() - start;
395 if (diff < 0)
396 diff += 1000000000;
397 } while (diff < loops);
398 } else {
399 start = get_tbl();
400 while (get_tbl() - start < loops)
401 HMT_low();
402 HMT_medium();
403 }
404 }
405 EXPORT_SYMBOL(__delay);
406
udelay(unsigned long usecs)407 void udelay(unsigned long usecs)
408 {
409 __delay(tb_ticks_per_usec * usecs);
410 }
411 EXPORT_SYMBOL(udelay);
412
413 #ifdef CONFIG_SMP
profile_pc(struct pt_regs * regs)414 unsigned long profile_pc(struct pt_regs *regs)
415 {
416 unsigned long pc = instruction_pointer(regs);
417
418 if (in_lock_functions(pc))
419 return regs->link;
420
421 return pc;
422 }
423 EXPORT_SYMBOL(profile_pc);
424 #endif
425
426 #ifdef CONFIG_IRQ_WORK
427
428 /*
429 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
430 */
431 #ifdef CONFIG_PPC64
test_irq_work_pending(void)432 static inline unsigned long test_irq_work_pending(void)
433 {
434 unsigned long x;
435
436 asm volatile("lbz %0,%1(13)"
437 : "=r" (x)
438 : "i" (offsetof(struct paca_struct, irq_work_pending)));
439 return x;
440 }
441
set_irq_work_pending_flag(void)442 static inline void set_irq_work_pending_flag(void)
443 {
444 asm volatile("stb %0,%1(13)" : :
445 "r" (1),
446 "i" (offsetof(struct paca_struct, irq_work_pending)));
447 }
448
clear_irq_work_pending(void)449 static inline void clear_irq_work_pending(void)
450 {
451 asm volatile("stb %0,%1(13)" : :
452 "r" (0),
453 "i" (offsetof(struct paca_struct, irq_work_pending)));
454 }
455
456 #else /* 32-bit */
457
458 DEFINE_PER_CPU(u8, irq_work_pending);
459
460 #define set_irq_work_pending_flag() __get_cpu_var(irq_work_pending) = 1
461 #define test_irq_work_pending() __get_cpu_var(irq_work_pending)
462 #define clear_irq_work_pending() __get_cpu_var(irq_work_pending) = 0
463
464 #endif /* 32 vs 64 bit */
465
arch_irq_work_raise(void)466 void arch_irq_work_raise(void)
467 {
468 preempt_disable();
469 set_irq_work_pending_flag();
470 set_dec(1);
471 preempt_enable();
472 }
473
474 #else /* CONFIG_IRQ_WORK */
475
476 #define test_irq_work_pending() 0
477 #define clear_irq_work_pending()
478
479 #endif /* CONFIG_IRQ_WORK */
480
481 /*
482 * timer_interrupt - gets called when the decrementer overflows,
483 * with interrupts disabled.
484 */
timer_interrupt(struct pt_regs * regs)485 void timer_interrupt(struct pt_regs * regs)
486 {
487 struct pt_regs *old_regs;
488 u64 *next_tb = &__get_cpu_var(decrementers_next_tb);
489 struct clock_event_device *evt = &__get_cpu_var(decrementers);
490 u64 now;
491
492 /* Ensure a positive value is written to the decrementer, or else
493 * some CPUs will continue to take decrementer exceptions.
494 */
495 set_dec(DECREMENTER_MAX);
496
497 /* Some implementations of hotplug will get timer interrupts while
498 * offline, just ignore these and we also need to set
499 * decrementers_next_tb as MAX to make sure __check_irq_replay
500 * don't replay timer interrupt when return, otherwise we'll trap
501 * here infinitely :(
502 */
503 if (!cpu_online(smp_processor_id())) {
504 *next_tb = ~(u64)0;
505 return;
506 }
507
508 /* Conditionally hard-enable interrupts now that the DEC has been
509 * bumped to its maximum value
510 */
511 may_hard_irq_enable();
512
513 __get_cpu_var(irq_stat).timer_irqs++;
514
515 #if defined(CONFIG_PPC32) && defined(CONFIG_PMAC)
516 if (atomic_read(&ppc_n_lost_interrupts) != 0)
517 do_IRQ(regs);
518 #endif
519
520 old_regs = set_irq_regs(regs);
521 irq_enter();
522
523 trace_timer_interrupt_entry(regs);
524
525 if (test_irq_work_pending()) {
526 clear_irq_work_pending();
527 irq_work_run();
528 }
529
530 now = get_tb_or_rtc();
531 if (now >= *next_tb) {
532 *next_tb = ~(u64)0;
533 if (evt->event_handler)
534 evt->event_handler(evt);
535 } else {
536 now = *next_tb - now;
537 if (now <= DECREMENTER_MAX)
538 set_dec((int)now);
539 }
540
541 #ifdef CONFIG_PPC64
542 /* collect purr register values often, for accurate calculations */
543 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
544 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
545 cu->current_tb = mfspr(SPRN_PURR);
546 }
547 #endif
548
549 trace_timer_interrupt_exit(regs);
550
551 irq_exit();
552 set_irq_regs(old_regs);
553 }
554
555 /*
556 * Hypervisor decrementer interrupts shouldn't occur but are sometimes
557 * left pending on exit from a KVM guest. We don't need to do anything
558 * to clear them, as they are edge-triggered.
559 */
hdec_interrupt(struct pt_regs * regs)560 void hdec_interrupt(struct pt_regs *regs)
561 {
562 }
563
564 #ifdef CONFIG_SUSPEND
generic_suspend_disable_irqs(void)565 static void generic_suspend_disable_irqs(void)
566 {
567 /* Disable the decrementer, so that it doesn't interfere
568 * with suspending.
569 */
570
571 set_dec(DECREMENTER_MAX);
572 local_irq_disable();
573 set_dec(DECREMENTER_MAX);
574 }
575
generic_suspend_enable_irqs(void)576 static void generic_suspend_enable_irqs(void)
577 {
578 local_irq_enable();
579 }
580
581 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_disable_irqs(void)582 void arch_suspend_disable_irqs(void)
583 {
584 if (ppc_md.suspend_disable_irqs)
585 ppc_md.suspend_disable_irqs();
586 generic_suspend_disable_irqs();
587 }
588
589 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_enable_irqs(void)590 void arch_suspend_enable_irqs(void)
591 {
592 generic_suspend_enable_irqs();
593 if (ppc_md.suspend_enable_irqs)
594 ppc_md.suspend_enable_irqs();
595 }
596 #endif
597
598 /*
599 * Scheduler clock - returns current time in nanosec units.
600 *
601 * Note: mulhdu(a, b) (multiply high double unsigned) returns
602 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
603 * are 64-bit unsigned numbers.
604 */
sched_clock(void)605 unsigned long long sched_clock(void)
606 {
607 if (__USE_RTC())
608 return get_rtc();
609 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
610 }
611
get_freq(char * name,int cells,unsigned long * val)612 static int __init get_freq(char *name, int cells, unsigned long *val)
613 {
614 struct device_node *cpu;
615 const unsigned int *fp;
616 int found = 0;
617
618 /* The cpu node should have timebase and clock frequency properties */
619 cpu = of_find_node_by_type(NULL, "cpu");
620
621 if (cpu) {
622 fp = of_get_property(cpu, name, NULL);
623 if (fp) {
624 found = 1;
625 *val = of_read_ulong(fp, cells);
626 }
627
628 of_node_put(cpu);
629 }
630
631 return found;
632 }
633
634 /* should become __cpuinit when secondary_cpu_time_init also is */
start_cpu_decrementer(void)635 void start_cpu_decrementer(void)
636 {
637 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
638 /* Clear any pending timer interrupts */
639 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
640
641 /* Enable decrementer interrupt */
642 mtspr(SPRN_TCR, TCR_DIE);
643 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
644 }
645
generic_calibrate_decr(void)646 void __init generic_calibrate_decr(void)
647 {
648 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
649
650 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
651 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
652
653 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
654 "(not found)\n");
655 }
656
657 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
658
659 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
660 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
661
662 printk(KERN_ERR "WARNING: Estimating processor frequency "
663 "(not found)\n");
664 }
665 }
666
update_persistent_clock(struct timespec now)667 int update_persistent_clock(struct timespec now)
668 {
669 struct rtc_time tm;
670
671 if (!ppc_md.set_rtc_time)
672 return -ENODEV;
673
674 to_tm(now.tv_sec + 1 + timezone_offset, &tm);
675 tm.tm_year -= 1900;
676 tm.tm_mon -= 1;
677
678 return ppc_md.set_rtc_time(&tm);
679 }
680
__read_persistent_clock(struct timespec * ts)681 static void __read_persistent_clock(struct timespec *ts)
682 {
683 struct rtc_time tm;
684 static int first = 1;
685
686 ts->tv_nsec = 0;
687 /* XXX this is a litle fragile but will work okay in the short term */
688 if (first) {
689 first = 0;
690 if (ppc_md.time_init)
691 timezone_offset = ppc_md.time_init();
692
693 /* get_boot_time() isn't guaranteed to be safe to call late */
694 if (ppc_md.get_boot_time) {
695 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
696 return;
697 }
698 }
699 if (!ppc_md.get_rtc_time) {
700 ts->tv_sec = 0;
701 return;
702 }
703 ppc_md.get_rtc_time(&tm);
704
705 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
706 tm.tm_hour, tm.tm_min, tm.tm_sec);
707 }
708
read_persistent_clock(struct timespec * ts)709 void read_persistent_clock(struct timespec *ts)
710 {
711 __read_persistent_clock(ts);
712
713 /* Sanitize it in case real time clock is set below EPOCH */
714 if (ts->tv_sec < 0) {
715 ts->tv_sec = 0;
716 ts->tv_nsec = 0;
717 }
718
719 }
720
721 /* clocksource code */
rtc_read(struct clocksource * cs)722 static cycle_t rtc_read(struct clocksource *cs)
723 {
724 return (cycle_t)get_rtc();
725 }
726
timebase_read(struct clocksource * cs)727 static cycle_t timebase_read(struct clocksource *cs)
728 {
729 return (cycle_t)get_tb();
730 }
731
update_vsyscall_old(struct timespec * wall_time,struct timespec * wtm,struct clocksource * clock,u32 mult)732 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
733 struct clocksource *clock, u32 mult)
734 {
735 u64 new_tb_to_xs, new_stamp_xsec;
736 u32 frac_sec;
737
738 if (clock != &clocksource_timebase)
739 return;
740
741 /* Make userspace gettimeofday spin until we're done. */
742 ++vdso_data->tb_update_count;
743 smp_mb();
744
745 /* 19342813113834067 ~= 2^(20+64) / 1e9 */
746 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
747 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
748 do_div(new_stamp_xsec, 1000000000);
749 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
750
751 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
752 /* this is tv_nsec / 1e9 as a 0.32 fraction */
753 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
754
755 /*
756 * tb_update_count is used to allow the userspace gettimeofday code
757 * to assure itself that it sees a consistent view of the tb_to_xs and
758 * stamp_xsec variables. It reads the tb_update_count, then reads
759 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
760 * the two values of tb_update_count match and are even then the
761 * tb_to_xs and stamp_xsec values are consistent. If not, then it
762 * loops back and reads them again until this criteria is met.
763 * We expect the caller to have done the first increment of
764 * vdso_data->tb_update_count already.
765 */
766 vdso_data->tb_orig_stamp = clock->cycle_last;
767 vdso_data->stamp_xsec = new_stamp_xsec;
768 vdso_data->tb_to_xs = new_tb_to_xs;
769 vdso_data->wtom_clock_sec = wtm->tv_sec;
770 vdso_data->wtom_clock_nsec = wtm->tv_nsec;
771 vdso_data->stamp_xtime = *wall_time;
772 vdso_data->stamp_sec_fraction = frac_sec;
773 smp_wmb();
774 ++(vdso_data->tb_update_count);
775 }
776
update_vsyscall_tz(void)777 void update_vsyscall_tz(void)
778 {
779 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
780 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
781 }
782
clocksource_init(void)783 static void __init clocksource_init(void)
784 {
785 struct clocksource *clock;
786
787 if (__USE_RTC())
788 clock = &clocksource_rtc;
789 else
790 clock = &clocksource_timebase;
791
792 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
793 printk(KERN_ERR "clocksource: %s is already registered\n",
794 clock->name);
795 return;
796 }
797
798 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
799 clock->name, clock->mult, clock->shift);
800 }
801
decrementer_set_next_event(unsigned long evt,struct clock_event_device * dev)802 static int decrementer_set_next_event(unsigned long evt,
803 struct clock_event_device *dev)
804 {
805 __get_cpu_var(decrementers_next_tb) = get_tb_or_rtc() + evt;
806 set_dec(evt);
807 return 0;
808 }
809
decrementer_set_mode(enum clock_event_mode mode,struct clock_event_device * dev)810 static void decrementer_set_mode(enum clock_event_mode mode,
811 struct clock_event_device *dev)
812 {
813 if (mode != CLOCK_EVT_MODE_ONESHOT)
814 decrementer_set_next_event(DECREMENTER_MAX, dev);
815 }
816
register_decrementer_clockevent(int cpu)817 static void register_decrementer_clockevent(int cpu)
818 {
819 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
820
821 *dec = decrementer_clockevent;
822 dec->cpumask = cpumask_of(cpu);
823
824 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
825 dec->name, dec->mult, dec->shift, cpu);
826
827 clockevents_register_device(dec);
828 }
829
init_decrementer_clockevent(void)830 static void __init init_decrementer_clockevent(void)
831 {
832 int cpu = smp_processor_id();
833
834 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
835
836 decrementer_clockevent.max_delta_ns =
837 clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
838 decrementer_clockevent.min_delta_ns =
839 clockevent_delta2ns(2, &decrementer_clockevent);
840
841 register_decrementer_clockevent(cpu);
842 }
843
secondary_cpu_time_init(void)844 void secondary_cpu_time_init(void)
845 {
846 /* Start the decrementer on CPUs that have manual control
847 * such as BookE
848 */
849 start_cpu_decrementer();
850
851 /* FIME: Should make unrelatred change to move snapshot_timebase
852 * call here ! */
853 register_decrementer_clockevent(smp_processor_id());
854 }
855
856 /* This function is only called on the boot processor */
time_init(void)857 void __init time_init(void)
858 {
859 struct div_result res;
860 u64 scale;
861 unsigned shift;
862
863 if (__USE_RTC()) {
864 /* 601 processor: dec counts down by 128 every 128ns */
865 ppc_tb_freq = 1000000000;
866 } else {
867 /* Normal PowerPC with timebase register */
868 ppc_md.calibrate_decr();
869 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
870 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
871 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
872 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
873 }
874
875 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
876 tb_ticks_per_sec = ppc_tb_freq;
877 tb_ticks_per_usec = ppc_tb_freq / 1000000;
878 calc_cputime_factors();
879 setup_cputime_one_jiffy();
880
881 /*
882 * Compute scale factor for sched_clock.
883 * The calibrate_decr() function has set tb_ticks_per_sec,
884 * which is the timebase frequency.
885 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
886 * the 128-bit result as a 64.64 fixed-point number.
887 * We then shift that number right until it is less than 1.0,
888 * giving us the scale factor and shift count to use in
889 * sched_clock().
890 */
891 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
892 scale = res.result_low;
893 for (shift = 0; res.result_high != 0; ++shift) {
894 scale = (scale >> 1) | (res.result_high << 63);
895 res.result_high >>= 1;
896 }
897 tb_to_ns_scale = scale;
898 tb_to_ns_shift = shift;
899 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
900 boot_tb = get_tb_or_rtc();
901
902 /* If platform provided a timezone (pmac), we correct the time */
903 if (timezone_offset) {
904 sys_tz.tz_minuteswest = -timezone_offset / 60;
905 sys_tz.tz_dsttime = 0;
906 }
907
908 vdso_data->tb_update_count = 0;
909 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
910
911 /* Start the decrementer on CPUs that have manual control
912 * such as BookE
913 */
914 start_cpu_decrementer();
915
916 /* Register the clocksource */
917 clocksource_init();
918
919 init_decrementer_clockevent();
920 }
921
922
923 #define FEBRUARY 2
924 #define STARTOFTIME 1970
925 #define SECDAY 86400L
926 #define SECYR (SECDAY * 365)
927 #define leapyear(year) ((year) % 4 == 0 && \
928 ((year) % 100 != 0 || (year) % 400 == 0))
929 #define days_in_year(a) (leapyear(a) ? 366 : 365)
930 #define days_in_month(a) (month_days[(a) - 1])
931
932 static int month_days[12] = {
933 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
934 };
935
936 /*
937 * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
938 */
GregorianDay(struct rtc_time * tm)939 void GregorianDay(struct rtc_time * tm)
940 {
941 int leapsToDate;
942 int lastYear;
943 int day;
944 int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
945
946 lastYear = tm->tm_year - 1;
947
948 /*
949 * Number of leap corrections to apply up to end of last year
950 */
951 leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
952
953 /*
954 * This year is a leap year if it is divisible by 4 except when it is
955 * divisible by 100 unless it is divisible by 400
956 *
957 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
958 */
959 day = tm->tm_mon > 2 && leapyear(tm->tm_year);
960
961 day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
962 tm->tm_mday;
963
964 tm->tm_wday = day % 7;
965 }
966
to_tm(int tim,struct rtc_time * tm)967 void to_tm(int tim, struct rtc_time * tm)
968 {
969 register int i;
970 register long hms, day;
971
972 day = tim / SECDAY;
973 hms = tim % SECDAY;
974
975 /* Hours, minutes, seconds are easy */
976 tm->tm_hour = hms / 3600;
977 tm->tm_min = (hms % 3600) / 60;
978 tm->tm_sec = (hms % 3600) % 60;
979
980 /* Number of years in days */
981 for (i = STARTOFTIME; day >= days_in_year(i); i++)
982 day -= days_in_year(i);
983 tm->tm_year = i;
984
985 /* Number of months in days left */
986 if (leapyear(tm->tm_year))
987 days_in_month(FEBRUARY) = 29;
988 for (i = 1; day >= days_in_month(i); i++)
989 day -= days_in_month(i);
990 days_in_month(FEBRUARY) = 28;
991 tm->tm_mon = i;
992
993 /* Days are what is left over (+1) from all that. */
994 tm->tm_mday = day + 1;
995
996 /*
997 * Determine the day of week
998 */
999 GregorianDay(tm);
1000 }
1001
1002 /*
1003 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1004 * result.
1005 */
div128_by_32(u64 dividend_high,u64 dividend_low,unsigned divisor,struct div_result * dr)1006 void div128_by_32(u64 dividend_high, u64 dividend_low,
1007 unsigned divisor, struct div_result *dr)
1008 {
1009 unsigned long a, b, c, d;
1010 unsigned long w, x, y, z;
1011 u64 ra, rb, rc;
1012
1013 a = dividend_high >> 32;
1014 b = dividend_high & 0xffffffff;
1015 c = dividend_low >> 32;
1016 d = dividend_low & 0xffffffff;
1017
1018 w = a / divisor;
1019 ra = ((u64)(a - (w * divisor)) << 32) + b;
1020
1021 rb = ((u64) do_div(ra, divisor) << 32) + c;
1022 x = ra;
1023
1024 rc = ((u64) do_div(rb, divisor) << 32) + d;
1025 y = rb;
1026
1027 do_div(rc, divisor);
1028 z = rc;
1029
1030 dr->result_high = ((u64)w << 32) + x;
1031 dr->result_low = ((u64)y << 32) + z;
1032
1033 }
1034
1035 /* We don't need to calibrate delay, we use the CPU timebase for that */
calibrate_delay(void)1036 void calibrate_delay(void)
1037 {
1038 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1039 * as the number of __delay(1) in a jiffy, so make it so
1040 */
1041 loops_per_jiffy = tb_ticks_per_jiffy;
1042 }
1043
rtc_init(void)1044 static int __init rtc_init(void)
1045 {
1046 struct platform_device *pdev;
1047
1048 if (!ppc_md.get_rtc_time)
1049 return -ENODEV;
1050
1051 pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1052
1053 return PTR_RET(pdev);
1054 }
1055
1056 module_init(rtc_init);
1057