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. (for iSeries, we calibrate the timebase
21 * against the Titan chip's clock.)
22 * - for astronomical applications: add a new function to get
23 * non ambiguous timestamps even around leap seconds. This needs
24 * a new timestamp format and a good name.
25 *
26 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
27 * "A Kernel Model for Precision Timekeeping" by Dave Mills
28 *
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
33 */
34
35 #include <linux/errno.h>
36 #include <linux/module.h>
37 #include <linux/sched.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/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
56 #include <asm/io.h>
57 #include <asm/processor.h>
58 #include <asm/nvram.h>
59 #include <asm/cache.h>
60 #include <asm/machdep.h>
61 #include <asm/uaccess.h>
62 #include <asm/time.h>
63 #include <asm/prom.h>
64 #include <asm/irq.h>
65 #include <asm/div64.h>
66 #include <asm/smp.h>
67 #include <asm/vdso_datapage.h>
68 #include <asm/firmware.h>
69 #include <asm/cputime.h>
70 #ifdef CONFIG_PPC_ISERIES
71 #include <asm/iseries/it_lp_queue.h>
72 #include <asm/iseries/hv_call_xm.h>
73 #endif
74
75 /* powerpc clocksource/clockevent code */
76
77 #include <linux/clockchips.h>
78 #include <linux/clocksource.h>
79
80 static cycle_t rtc_read(void);
81 static struct clocksource clocksource_rtc = {
82 .name = "rtc",
83 .rating = 400,
84 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
85 .mask = CLOCKSOURCE_MASK(64),
86 .shift = 22,
87 .mult = 0, /* To be filled in */
88 .read = rtc_read,
89 };
90
91 static cycle_t timebase_read(void);
92 static struct clocksource clocksource_timebase = {
93 .name = "timebase",
94 .rating = 400,
95 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
96 .mask = CLOCKSOURCE_MASK(64),
97 .shift = 22,
98 .mult = 0, /* To be filled in */
99 .read = timebase_read,
100 };
101
102 #define DECREMENTER_MAX 0x7fffffff
103
104 static int decrementer_set_next_event(unsigned long evt,
105 struct clock_event_device *dev);
106 static void decrementer_set_mode(enum clock_event_mode mode,
107 struct clock_event_device *dev);
108
109 static struct clock_event_device decrementer_clockevent = {
110 .name = "decrementer",
111 .rating = 200,
112 .shift = 16,
113 .mult = 0, /* To be filled in */
114 .irq = 0,
115 .set_next_event = decrementer_set_next_event,
116 .set_mode = decrementer_set_mode,
117 .features = CLOCK_EVT_FEAT_ONESHOT,
118 };
119
120 struct decrementer_clock {
121 struct clock_event_device event;
122 u64 next_tb;
123 };
124
125 static DEFINE_PER_CPU(struct decrementer_clock, decrementers);
126
127 #ifdef CONFIG_PPC_ISERIES
128 static unsigned long __initdata iSeries_recal_titan;
129 static signed long __initdata iSeries_recal_tb;
130
131 /* Forward declaration is only needed for iSereis compiles */
132 static void __init clocksource_init(void);
133 #endif
134
135 #define XSEC_PER_SEC (1024*1024)
136
137 #ifdef CONFIG_PPC64
138 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
139 #else
140 /* compute ((xsec << 12) * max) >> 32 */
141 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
142 #endif
143
144 unsigned long tb_ticks_per_jiffy;
145 unsigned long tb_ticks_per_usec = 100; /* sane default */
146 EXPORT_SYMBOL(tb_ticks_per_usec);
147 unsigned long tb_ticks_per_sec;
148 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
149 u64 tb_to_xs;
150 unsigned tb_to_us;
151
152 #define TICKLEN_SCALE NTP_SCALE_SHIFT
153 static u64 last_tick_len; /* units are ns / 2^TICKLEN_SCALE */
154 static u64 ticklen_to_xs; /* 0.64 fraction */
155
156 /* If last_tick_len corresponds to about 1/HZ seconds, then
157 last_tick_len << TICKLEN_SHIFT will be about 2^63. */
158 #define TICKLEN_SHIFT (63 - 30 - TICKLEN_SCALE + SHIFT_HZ)
159
160 DEFINE_SPINLOCK(rtc_lock);
161 EXPORT_SYMBOL_GPL(rtc_lock);
162
163 static u64 tb_to_ns_scale __read_mostly;
164 static unsigned tb_to_ns_shift __read_mostly;
165 static unsigned long boot_tb __read_mostly;
166
167 extern struct timezone sys_tz;
168 static long timezone_offset;
169
170 unsigned long ppc_proc_freq;
171 EXPORT_SYMBOL(ppc_proc_freq);
172 unsigned long ppc_tb_freq;
173
174 static u64 tb_last_jiffy __cacheline_aligned_in_smp;
175 static DEFINE_PER_CPU(u64, last_jiffy);
176
177 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
178 /*
179 * Factors for converting from cputime_t (timebase ticks) to
180 * jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds).
181 * These are all stored as 0.64 fixed-point binary fractions.
182 */
183 u64 __cputime_jiffies_factor;
184 EXPORT_SYMBOL(__cputime_jiffies_factor);
185 u64 __cputime_msec_factor;
186 EXPORT_SYMBOL(__cputime_msec_factor);
187 u64 __cputime_sec_factor;
188 EXPORT_SYMBOL(__cputime_sec_factor);
189 u64 __cputime_clockt_factor;
190 EXPORT_SYMBOL(__cputime_clockt_factor);
191 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
192 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
193
calc_cputime_factors(void)194 static void calc_cputime_factors(void)
195 {
196 struct div_result res;
197
198 div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
199 __cputime_jiffies_factor = res.result_low;
200 div128_by_32(1000, 0, tb_ticks_per_sec, &res);
201 __cputime_msec_factor = res.result_low;
202 div128_by_32(1, 0, tb_ticks_per_sec, &res);
203 __cputime_sec_factor = res.result_low;
204 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
205 __cputime_clockt_factor = res.result_low;
206 }
207
208 /*
209 * Read the PURR on systems that have it, otherwise the timebase.
210 */
read_purr(void)211 static u64 read_purr(void)
212 {
213 if (cpu_has_feature(CPU_FTR_PURR))
214 return mfspr(SPRN_PURR);
215 return mftb();
216 }
217
218 /*
219 * Read the SPURR on systems that have it, otherwise the purr
220 */
read_spurr(u64 purr)221 static u64 read_spurr(u64 purr)
222 {
223 /*
224 * cpus without PURR won't have a SPURR
225 * We already know the former when we use this, so tell gcc
226 */
227 if (cpu_has_feature(CPU_FTR_PURR) && cpu_has_feature(CPU_FTR_SPURR))
228 return mfspr(SPRN_SPURR);
229 return purr;
230 }
231
232 /*
233 * Account time for a transition between system, hard irq
234 * or soft irq state.
235 */
account_system_vtime(struct task_struct * tsk)236 void account_system_vtime(struct task_struct *tsk)
237 {
238 u64 now, nowscaled, delta, deltascaled, sys_time;
239 unsigned long flags;
240
241 local_irq_save(flags);
242 now = read_purr();
243 nowscaled = read_spurr(now);
244 delta = now - get_paca()->startpurr;
245 deltascaled = nowscaled - get_paca()->startspurr;
246 get_paca()->startpurr = now;
247 get_paca()->startspurr = nowscaled;
248 if (!in_interrupt()) {
249 /* deltascaled includes both user and system time.
250 * Hence scale it based on the purr ratio to estimate
251 * the system time */
252 sys_time = get_paca()->system_time;
253 if (get_paca()->user_time)
254 deltascaled = deltascaled * sys_time /
255 (sys_time + get_paca()->user_time);
256 delta += sys_time;
257 get_paca()->system_time = 0;
258 }
259 if (in_irq() || idle_task(smp_processor_id()) != tsk)
260 account_system_time(tsk, 0, delta, deltascaled);
261 else
262 account_idle_time(delta);
263 per_cpu(cputime_last_delta, smp_processor_id()) = delta;
264 per_cpu(cputime_scaled_last_delta, smp_processor_id()) = deltascaled;
265 local_irq_restore(flags);
266 }
267
268 /*
269 * Transfer the user and system times accumulated in the paca
270 * by the exception entry and exit code to the generic process
271 * user and system time records.
272 * Must be called with interrupts disabled.
273 */
account_process_tick(struct task_struct * tsk,int user_tick)274 void account_process_tick(struct task_struct *tsk, int user_tick)
275 {
276 cputime_t utime, utimescaled;
277
278 utime = get_paca()->user_time;
279 get_paca()->user_time = 0;
280 utimescaled = cputime_to_scaled(utime);
281 account_user_time(tsk, utime, utimescaled);
282 }
283
284 /*
285 * Stuff for accounting stolen time.
286 */
287 struct cpu_purr_data {
288 int initialized; /* thread is running */
289 u64 tb; /* last TB value read */
290 u64 purr; /* last PURR value read */
291 u64 spurr; /* last SPURR value read */
292 };
293
294 /*
295 * Each entry in the cpu_purr_data array is manipulated only by its
296 * "owner" cpu -- usually in the timer interrupt but also occasionally
297 * in process context for cpu online. As long as cpus do not touch
298 * each others' cpu_purr_data, disabling local interrupts is
299 * sufficient to serialize accesses.
300 */
301 static DEFINE_PER_CPU(struct cpu_purr_data, cpu_purr_data);
302
snapshot_tb_and_purr(void * data)303 static void snapshot_tb_and_purr(void *data)
304 {
305 unsigned long flags;
306 struct cpu_purr_data *p = &__get_cpu_var(cpu_purr_data);
307
308 local_irq_save(flags);
309 p->tb = get_tb_or_rtc();
310 p->purr = mfspr(SPRN_PURR);
311 wmb();
312 p->initialized = 1;
313 local_irq_restore(flags);
314 }
315
316 /*
317 * Called during boot when all cpus have come up.
318 */
snapshot_timebases(void)319 void snapshot_timebases(void)
320 {
321 if (!cpu_has_feature(CPU_FTR_PURR))
322 return;
323 on_each_cpu(snapshot_tb_and_purr, NULL, 1);
324 }
325
326 /*
327 * Must be called with interrupts disabled.
328 */
calculate_steal_time(void)329 void calculate_steal_time(void)
330 {
331 u64 tb, purr;
332 s64 stolen;
333 struct cpu_purr_data *pme;
334
335 pme = &__get_cpu_var(cpu_purr_data);
336 if (!pme->initialized)
337 return; /* !CPU_FTR_PURR or early in early boot */
338 tb = mftb();
339 purr = mfspr(SPRN_PURR);
340 stolen = (tb - pme->tb) - (purr - pme->purr);
341 if (stolen > 0) {
342 if (idle_task(smp_processor_id()) != current)
343 account_steal_time(stolen);
344 else
345 account_idle_time(stolen);
346 }
347 pme->tb = tb;
348 pme->purr = purr;
349 }
350
351 #ifdef CONFIG_PPC_SPLPAR
352 /*
353 * Must be called before the cpu is added to the online map when
354 * a cpu is being brought up at runtime.
355 */
snapshot_purr(void)356 static void snapshot_purr(void)
357 {
358 struct cpu_purr_data *pme;
359 unsigned long flags;
360
361 if (!cpu_has_feature(CPU_FTR_PURR))
362 return;
363 local_irq_save(flags);
364 pme = &__get_cpu_var(cpu_purr_data);
365 pme->tb = mftb();
366 pme->purr = mfspr(SPRN_PURR);
367 pme->initialized = 1;
368 local_irq_restore(flags);
369 }
370
371 #endif /* CONFIG_PPC_SPLPAR */
372
373 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
374 #define calc_cputime_factors()
375 #define calculate_steal_time() do { } while (0)
376 #endif
377
378 #if !(defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR))
379 #define snapshot_purr() do { } while (0)
380 #endif
381
382 /*
383 * Called when a cpu comes up after the system has finished booting,
384 * i.e. as a result of a hotplug cpu action.
385 */
snapshot_timebase(void)386 void snapshot_timebase(void)
387 {
388 __get_cpu_var(last_jiffy) = get_tb_or_rtc();
389 snapshot_purr();
390 }
391
__delay(unsigned long loops)392 void __delay(unsigned long loops)
393 {
394 unsigned long start;
395 int diff;
396
397 if (__USE_RTC()) {
398 start = get_rtcl();
399 do {
400 /* the RTCL register wraps at 1000000000 */
401 diff = get_rtcl() - start;
402 if (diff < 0)
403 diff += 1000000000;
404 } while (diff < loops);
405 } else {
406 start = get_tbl();
407 while (get_tbl() - start < loops)
408 HMT_low();
409 HMT_medium();
410 }
411 }
412 EXPORT_SYMBOL(__delay);
413
udelay(unsigned long usecs)414 void udelay(unsigned long usecs)
415 {
416 __delay(tb_ticks_per_usec * usecs);
417 }
418 EXPORT_SYMBOL(udelay);
419
update_gtod(u64 new_tb_stamp,u64 new_stamp_xsec,u64 new_tb_to_xs)420 static inline void update_gtod(u64 new_tb_stamp, u64 new_stamp_xsec,
421 u64 new_tb_to_xs)
422 {
423 /*
424 * tb_update_count is used to allow the userspace gettimeofday code
425 * to assure itself that it sees a consistent view of the tb_to_xs and
426 * stamp_xsec variables. It reads the tb_update_count, then reads
427 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
428 * the two values of tb_update_count match and are even then the
429 * tb_to_xs and stamp_xsec values are consistent. If not, then it
430 * loops back and reads them again until this criteria is met.
431 * We expect the caller to have done the first increment of
432 * vdso_data->tb_update_count already.
433 */
434 vdso_data->tb_orig_stamp = new_tb_stamp;
435 vdso_data->stamp_xsec = new_stamp_xsec;
436 vdso_data->tb_to_xs = new_tb_to_xs;
437 vdso_data->wtom_clock_sec = wall_to_monotonic.tv_sec;
438 vdso_data->wtom_clock_nsec = wall_to_monotonic.tv_nsec;
439 vdso_data->stamp_xtime = xtime;
440 smp_wmb();
441 ++(vdso_data->tb_update_count);
442 }
443
444 #ifdef CONFIG_SMP
profile_pc(struct pt_regs * regs)445 unsigned long profile_pc(struct pt_regs *regs)
446 {
447 unsigned long pc = instruction_pointer(regs);
448
449 if (in_lock_functions(pc))
450 return regs->link;
451
452 return pc;
453 }
454 EXPORT_SYMBOL(profile_pc);
455 #endif
456
457 #ifdef CONFIG_PPC_ISERIES
458
459 /*
460 * This function recalibrates the timebase based on the 49-bit time-of-day
461 * value in the Titan chip. The Titan is much more accurate than the value
462 * returned by the service processor for the timebase frequency.
463 */
464
iSeries_tb_recal(void)465 static int __init iSeries_tb_recal(void)
466 {
467 struct div_result divres;
468 unsigned long titan, tb;
469
470 /* Make sure we only run on iSeries */
471 if (!firmware_has_feature(FW_FEATURE_ISERIES))
472 return -ENODEV;
473
474 tb = get_tb();
475 titan = HvCallXm_loadTod();
476 if ( iSeries_recal_titan ) {
477 unsigned long tb_ticks = tb - iSeries_recal_tb;
478 unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
479 unsigned long new_tb_ticks_per_sec = (tb_ticks * USEC_PER_SEC)/titan_usec;
480 unsigned long new_tb_ticks_per_jiffy = (new_tb_ticks_per_sec+(HZ/2))/HZ;
481 long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
482 char sign = '+';
483 /* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
484 new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;
485
486 if ( tick_diff < 0 ) {
487 tick_diff = -tick_diff;
488 sign = '-';
489 }
490 if ( tick_diff ) {
491 if ( tick_diff < tb_ticks_per_jiffy/25 ) {
492 printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
493 new_tb_ticks_per_jiffy, sign, tick_diff );
494 tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
495 tb_ticks_per_sec = new_tb_ticks_per_sec;
496 calc_cputime_factors();
497 div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres );
498 tb_to_xs = divres.result_low;
499 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
500 vdso_data->tb_to_xs = tb_to_xs;
501 }
502 else {
503 printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
504 " new tb_ticks_per_jiffy = %lu\n"
505 " old tb_ticks_per_jiffy = %lu\n",
506 new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
507 }
508 }
509 }
510 iSeries_recal_titan = titan;
511 iSeries_recal_tb = tb;
512
513 /* Called here as now we know accurate values for the timebase */
514 clocksource_init();
515 return 0;
516 }
517 late_initcall(iSeries_tb_recal);
518
519 /* Called from platform early init */
iSeries_time_init_early(void)520 void __init iSeries_time_init_early(void)
521 {
522 iSeries_recal_tb = get_tb();
523 iSeries_recal_titan = HvCallXm_loadTod();
524 }
525 #endif /* CONFIG_PPC_ISERIES */
526
527 /*
528 * For iSeries shared processors, we have to let the hypervisor
529 * set the hardware decrementer. We set a virtual decrementer
530 * in the lppaca and call the hypervisor if the virtual
531 * decrementer is less than the current value in the hardware
532 * decrementer. (almost always the new decrementer value will
533 * be greater than the current hardware decementer so the hypervisor
534 * call will not be needed)
535 */
536
537 /*
538 * timer_interrupt - gets called when the decrementer overflows,
539 * with interrupts disabled.
540 */
timer_interrupt(struct pt_regs * regs)541 void timer_interrupt(struct pt_regs * regs)
542 {
543 struct pt_regs *old_regs;
544 struct decrementer_clock *decrementer = &__get_cpu_var(decrementers);
545 struct clock_event_device *evt = &decrementer->event;
546 u64 now;
547
548 /* Ensure a positive value is written to the decrementer, or else
549 * some CPUs will continuue to take decrementer exceptions */
550 set_dec(DECREMENTER_MAX);
551
552 #ifdef CONFIG_PPC32
553 if (atomic_read(&ppc_n_lost_interrupts) != 0)
554 do_IRQ(regs);
555 #endif
556
557 now = get_tb_or_rtc();
558 if (now < decrementer->next_tb) {
559 /* not time for this event yet */
560 now = decrementer->next_tb - now;
561 if (now <= DECREMENTER_MAX)
562 set_dec((int)now);
563 return;
564 }
565 old_regs = set_irq_regs(regs);
566 irq_enter();
567
568 calculate_steal_time();
569
570 #ifdef CONFIG_PPC_ISERIES
571 if (firmware_has_feature(FW_FEATURE_ISERIES))
572 get_lppaca()->int_dword.fields.decr_int = 0;
573 #endif
574
575 if (evt->event_handler)
576 evt->event_handler(evt);
577
578 #ifdef CONFIG_PPC_ISERIES
579 if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending())
580 process_hvlpevents();
581 #endif
582
583 #ifdef CONFIG_PPC64
584 /* collect purr register values often, for accurate calculations */
585 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
586 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
587 cu->current_tb = mfspr(SPRN_PURR);
588 }
589 #endif
590
591 irq_exit();
592 set_irq_regs(old_regs);
593 }
594
wakeup_decrementer(void)595 void wakeup_decrementer(void)
596 {
597 unsigned long ticks;
598
599 /*
600 * The timebase gets saved on sleep and restored on wakeup,
601 * so all we need to do is to reset the decrementer.
602 */
603 ticks = tb_ticks_since(__get_cpu_var(last_jiffy));
604 if (ticks < tb_ticks_per_jiffy)
605 ticks = tb_ticks_per_jiffy - ticks;
606 else
607 ticks = 1;
608 set_dec(ticks);
609 }
610
611 #ifdef CONFIG_SUSPEND
generic_suspend_disable_irqs(void)612 void generic_suspend_disable_irqs(void)
613 {
614 preempt_disable();
615
616 /* Disable the decrementer, so that it doesn't interfere
617 * with suspending.
618 */
619
620 set_dec(0x7fffffff);
621 local_irq_disable();
622 set_dec(0x7fffffff);
623 }
624
generic_suspend_enable_irqs(void)625 void generic_suspend_enable_irqs(void)
626 {
627 wakeup_decrementer();
628
629 local_irq_enable();
630 preempt_enable();
631 }
632
633 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_disable_irqs(void)634 void arch_suspend_disable_irqs(void)
635 {
636 if (ppc_md.suspend_disable_irqs)
637 ppc_md.suspend_disable_irqs();
638 generic_suspend_disable_irqs();
639 }
640
641 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_enable_irqs(void)642 void arch_suspend_enable_irqs(void)
643 {
644 generic_suspend_enable_irqs();
645 if (ppc_md.suspend_enable_irqs)
646 ppc_md.suspend_enable_irqs();
647 }
648 #endif
649
650 #ifdef CONFIG_SMP
smp_space_timers(unsigned int max_cpus)651 void __init smp_space_timers(unsigned int max_cpus)
652 {
653 int i;
654 u64 previous_tb = per_cpu(last_jiffy, boot_cpuid);
655
656 /* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */
657 previous_tb -= tb_ticks_per_jiffy;
658
659 for_each_possible_cpu(i) {
660 if (i == boot_cpuid)
661 continue;
662 per_cpu(last_jiffy, i) = previous_tb;
663 }
664 }
665 #endif
666
667 /*
668 * Scheduler clock - returns current time in nanosec units.
669 *
670 * Note: mulhdu(a, b) (multiply high double unsigned) returns
671 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
672 * are 64-bit unsigned numbers.
673 */
sched_clock(void)674 unsigned long long sched_clock(void)
675 {
676 if (__USE_RTC())
677 return get_rtc();
678 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
679 }
680
get_freq(char * name,int cells,unsigned long * val)681 static int __init get_freq(char *name, int cells, unsigned long *val)
682 {
683 struct device_node *cpu;
684 const unsigned int *fp;
685 int found = 0;
686
687 /* The cpu node should have timebase and clock frequency properties */
688 cpu = of_find_node_by_type(NULL, "cpu");
689
690 if (cpu) {
691 fp = of_get_property(cpu, name, NULL);
692 if (fp) {
693 found = 1;
694 *val = of_read_ulong(fp, cells);
695 }
696
697 of_node_put(cpu);
698 }
699
700 return found;
701 }
702
generic_calibrate_decr(void)703 void __init generic_calibrate_decr(void)
704 {
705 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
706
707 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
708 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
709
710 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
711 "(not found)\n");
712 }
713
714 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
715
716 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
717 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
718
719 printk(KERN_ERR "WARNING: Estimating processor frequency "
720 "(not found)\n");
721 }
722
723 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
724 /* Clear any pending timer interrupts */
725 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
726
727 /* Enable decrementer interrupt */
728 mtspr(SPRN_TCR, TCR_DIE);
729 #endif
730 }
731
update_persistent_clock(struct timespec now)732 int update_persistent_clock(struct timespec now)
733 {
734 struct rtc_time tm;
735
736 if (!ppc_md.set_rtc_time)
737 return 0;
738
739 to_tm(now.tv_sec + 1 + timezone_offset, &tm);
740 tm.tm_year -= 1900;
741 tm.tm_mon -= 1;
742
743 return ppc_md.set_rtc_time(&tm);
744 }
745
read_persistent_clock(void)746 unsigned long read_persistent_clock(void)
747 {
748 struct rtc_time tm;
749 static int first = 1;
750
751 /* XXX this is a litle fragile but will work okay in the short term */
752 if (first) {
753 first = 0;
754 if (ppc_md.time_init)
755 timezone_offset = ppc_md.time_init();
756
757 /* get_boot_time() isn't guaranteed to be safe to call late */
758 if (ppc_md.get_boot_time)
759 return ppc_md.get_boot_time() -timezone_offset;
760 }
761 if (!ppc_md.get_rtc_time)
762 return 0;
763 ppc_md.get_rtc_time(&tm);
764 return mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
765 tm.tm_hour, tm.tm_min, tm.tm_sec);
766 }
767
768 /* clocksource code */
rtc_read(void)769 static cycle_t rtc_read(void)
770 {
771 return (cycle_t)get_rtc();
772 }
773
timebase_read(void)774 static cycle_t timebase_read(void)
775 {
776 return (cycle_t)get_tb();
777 }
778
update_vsyscall(struct timespec * wall_time,struct clocksource * clock)779 void update_vsyscall(struct timespec *wall_time, struct clocksource *clock)
780 {
781 u64 t2x, stamp_xsec;
782
783 if (clock != &clocksource_timebase)
784 return;
785
786 /* Make userspace gettimeofday spin until we're done. */
787 ++vdso_data->tb_update_count;
788 smp_mb();
789
790 /* XXX this assumes clock->shift == 22 */
791 /* 4611686018 ~= 2^(20+64-22) / 1e9 */
792 t2x = (u64) clock->mult * 4611686018ULL;
793 stamp_xsec = (u64) xtime.tv_nsec * XSEC_PER_SEC;
794 do_div(stamp_xsec, 1000000000);
795 stamp_xsec += (u64) xtime.tv_sec * XSEC_PER_SEC;
796 update_gtod(clock->cycle_last, stamp_xsec, t2x);
797 }
798
update_vsyscall_tz(void)799 void update_vsyscall_tz(void)
800 {
801 /* Make userspace gettimeofday spin until we're done. */
802 ++vdso_data->tb_update_count;
803 smp_mb();
804 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
805 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
806 smp_mb();
807 ++vdso_data->tb_update_count;
808 }
809
clocksource_init(void)810 static void __init clocksource_init(void)
811 {
812 struct clocksource *clock;
813
814 if (__USE_RTC())
815 clock = &clocksource_rtc;
816 else
817 clock = &clocksource_timebase;
818
819 clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
820
821 if (clocksource_register(clock)) {
822 printk(KERN_ERR "clocksource: %s is already registered\n",
823 clock->name);
824 return;
825 }
826
827 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
828 clock->name, clock->mult, clock->shift);
829 }
830
decrementer_set_next_event(unsigned long evt,struct clock_event_device * dev)831 static int decrementer_set_next_event(unsigned long evt,
832 struct clock_event_device *dev)
833 {
834 __get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
835 set_dec(evt);
836 return 0;
837 }
838
decrementer_set_mode(enum clock_event_mode mode,struct clock_event_device * dev)839 static void decrementer_set_mode(enum clock_event_mode mode,
840 struct clock_event_device *dev)
841 {
842 if (mode != CLOCK_EVT_MODE_ONESHOT)
843 decrementer_set_next_event(DECREMENTER_MAX, dev);
844 }
845
register_decrementer_clockevent(int cpu)846 static void register_decrementer_clockevent(int cpu)
847 {
848 struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
849
850 *dec = decrementer_clockevent;
851 dec->cpumask = cpumask_of(cpu);
852
853 printk(KERN_DEBUG "clockevent: %s mult[%lx] shift[%d] cpu[%d]\n",
854 dec->name, dec->mult, dec->shift, cpu);
855
856 clockevents_register_device(dec);
857 }
858
init_decrementer_clockevent(void)859 static void __init init_decrementer_clockevent(void)
860 {
861 int cpu = smp_processor_id();
862
863 decrementer_clockevent.mult = div_sc(ppc_tb_freq, NSEC_PER_SEC,
864 decrementer_clockevent.shift);
865 decrementer_clockevent.max_delta_ns =
866 clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
867 decrementer_clockevent.min_delta_ns =
868 clockevent_delta2ns(2, &decrementer_clockevent);
869
870 register_decrementer_clockevent(cpu);
871 }
872
secondary_cpu_time_init(void)873 void secondary_cpu_time_init(void)
874 {
875 /* FIME: Should make unrelatred change to move snapshot_timebase
876 * call here ! */
877 register_decrementer_clockevent(smp_processor_id());
878 }
879
880 /* This function is only called on the boot processor */
time_init(void)881 void __init time_init(void)
882 {
883 unsigned long flags;
884 struct div_result res;
885 u64 scale, x;
886 unsigned shift;
887
888 if (__USE_RTC()) {
889 /* 601 processor: dec counts down by 128 every 128ns */
890 ppc_tb_freq = 1000000000;
891 tb_last_jiffy = get_rtcl();
892 } else {
893 /* Normal PowerPC with timebase register */
894 ppc_md.calibrate_decr();
895 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
896 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
897 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
898 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
899 tb_last_jiffy = get_tb();
900 }
901
902 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
903 tb_ticks_per_sec = ppc_tb_freq;
904 tb_ticks_per_usec = ppc_tb_freq / 1000000;
905 tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000);
906 calc_cputime_factors();
907
908 /*
909 * Calculate the length of each tick in ns. It will not be
910 * exactly 1e9/HZ unless ppc_tb_freq is divisible by HZ.
911 * We compute 1e9 * tb_ticks_per_jiffy / ppc_tb_freq,
912 * rounded up.
913 */
914 x = (u64) NSEC_PER_SEC * tb_ticks_per_jiffy + ppc_tb_freq - 1;
915 do_div(x, ppc_tb_freq);
916 tick_nsec = x;
917 last_tick_len = x << TICKLEN_SCALE;
918
919 /*
920 * Compute ticklen_to_xs, which is a factor which gets multiplied
921 * by (last_tick_len << TICKLEN_SHIFT) to get a tb_to_xs value.
922 * It is computed as:
923 * ticklen_to_xs = 2^N / (tb_ticks_per_jiffy * 1e9)
924 * where N = 64 + 20 - TICKLEN_SCALE - TICKLEN_SHIFT
925 * which turns out to be N = 51 - SHIFT_HZ.
926 * This gives the result as a 0.64 fixed-point fraction.
927 * That value is reduced by an offset amounting to 1 xsec per
928 * 2^31 timebase ticks to avoid problems with time going backwards
929 * by 1 xsec when we do timer_recalc_offset due to losing the
930 * fractional xsec. That offset is equal to ppc_tb_freq/2^51
931 * since there are 2^20 xsec in a second.
932 */
933 div128_by_32((1ULL << 51) - ppc_tb_freq, 0,
934 tb_ticks_per_jiffy << SHIFT_HZ, &res);
935 div128_by_32(res.result_high, res.result_low, NSEC_PER_SEC, &res);
936 ticklen_to_xs = res.result_low;
937
938 /* Compute tb_to_xs from tick_nsec */
939 tb_to_xs = mulhdu(last_tick_len << TICKLEN_SHIFT, ticklen_to_xs);
940
941 /*
942 * Compute scale factor for sched_clock.
943 * The calibrate_decr() function has set tb_ticks_per_sec,
944 * which is the timebase frequency.
945 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
946 * the 128-bit result as a 64.64 fixed-point number.
947 * We then shift that number right until it is less than 1.0,
948 * giving us the scale factor and shift count to use in
949 * sched_clock().
950 */
951 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
952 scale = res.result_low;
953 for (shift = 0; res.result_high != 0; ++shift) {
954 scale = (scale >> 1) | (res.result_high << 63);
955 res.result_high >>= 1;
956 }
957 tb_to_ns_scale = scale;
958 tb_to_ns_shift = shift;
959 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
960 boot_tb = get_tb_or_rtc();
961
962 write_seqlock_irqsave(&xtime_lock, flags);
963
964 /* If platform provided a timezone (pmac), we correct the time */
965 if (timezone_offset) {
966 sys_tz.tz_minuteswest = -timezone_offset / 60;
967 sys_tz.tz_dsttime = 0;
968 }
969
970 vdso_data->tb_orig_stamp = tb_last_jiffy;
971 vdso_data->tb_update_count = 0;
972 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
973 vdso_data->stamp_xsec = (u64) xtime.tv_sec * XSEC_PER_SEC;
974 vdso_data->tb_to_xs = tb_to_xs;
975
976 write_sequnlock_irqrestore(&xtime_lock, flags);
977
978 /* Register the clocksource, if we're not running on iSeries */
979 if (!firmware_has_feature(FW_FEATURE_ISERIES))
980 clocksource_init();
981
982 init_decrementer_clockevent();
983 }
984
985
986 #define FEBRUARY 2
987 #define STARTOFTIME 1970
988 #define SECDAY 86400L
989 #define SECYR (SECDAY * 365)
990 #define leapyear(year) ((year) % 4 == 0 && \
991 ((year) % 100 != 0 || (year) % 400 == 0))
992 #define days_in_year(a) (leapyear(a) ? 366 : 365)
993 #define days_in_month(a) (month_days[(a) - 1])
994
995 static int month_days[12] = {
996 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
997 };
998
999 /*
1000 * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1001 */
GregorianDay(struct rtc_time * tm)1002 void GregorianDay(struct rtc_time * tm)
1003 {
1004 int leapsToDate;
1005 int lastYear;
1006 int day;
1007 int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1008
1009 lastYear = tm->tm_year - 1;
1010
1011 /*
1012 * Number of leap corrections to apply up to end of last year
1013 */
1014 leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
1015
1016 /*
1017 * This year is a leap year if it is divisible by 4 except when it is
1018 * divisible by 100 unless it is divisible by 400
1019 *
1020 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1021 */
1022 day = tm->tm_mon > 2 && leapyear(tm->tm_year);
1023
1024 day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
1025 tm->tm_mday;
1026
1027 tm->tm_wday = day % 7;
1028 }
1029
to_tm(int tim,struct rtc_time * tm)1030 void to_tm(int tim, struct rtc_time * tm)
1031 {
1032 register int i;
1033 register long hms, day;
1034
1035 day = tim / SECDAY;
1036 hms = tim % SECDAY;
1037
1038 /* Hours, minutes, seconds are easy */
1039 tm->tm_hour = hms / 3600;
1040 tm->tm_min = (hms % 3600) / 60;
1041 tm->tm_sec = (hms % 3600) % 60;
1042
1043 /* Number of years in days */
1044 for (i = STARTOFTIME; day >= days_in_year(i); i++)
1045 day -= days_in_year(i);
1046 tm->tm_year = i;
1047
1048 /* Number of months in days left */
1049 if (leapyear(tm->tm_year))
1050 days_in_month(FEBRUARY) = 29;
1051 for (i = 1; day >= days_in_month(i); i++)
1052 day -= days_in_month(i);
1053 days_in_month(FEBRUARY) = 28;
1054 tm->tm_mon = i;
1055
1056 /* Days are what is left over (+1) from all that. */
1057 tm->tm_mday = day + 1;
1058
1059 /*
1060 * Determine the day of week
1061 */
1062 GregorianDay(tm);
1063 }
1064
1065 /* Auxiliary function to compute scaling factors */
1066 /* Actually the choice of a timebase running at 1/4 the of the bus
1067 * frequency giving resolution of a few tens of nanoseconds is quite nice.
1068 * It makes this computation very precise (27-28 bits typically) which
1069 * is optimistic considering the stability of most processor clock
1070 * oscillators and the precision with which the timebase frequency
1071 * is measured but does not harm.
1072 */
mulhwu_scale_factor(unsigned inscale,unsigned outscale)1073 unsigned mulhwu_scale_factor(unsigned inscale, unsigned outscale)
1074 {
1075 unsigned mlt=0, tmp, err;
1076 /* No concern for performance, it's done once: use a stupid
1077 * but safe and compact method to find the multiplier.
1078 */
1079
1080 for (tmp = 1U<<31; tmp != 0; tmp >>= 1) {
1081 if (mulhwu(inscale, mlt|tmp) < outscale)
1082 mlt |= tmp;
1083 }
1084
1085 /* We might still be off by 1 for the best approximation.
1086 * A side effect of this is that if outscale is too large
1087 * the returned value will be zero.
1088 * Many corner cases have been checked and seem to work,
1089 * some might have been forgotten in the test however.
1090 */
1091
1092 err = inscale * (mlt+1);
1093 if (err <= inscale/2)
1094 mlt++;
1095 return mlt;
1096 }
1097
1098 /*
1099 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1100 * result.
1101 */
div128_by_32(u64 dividend_high,u64 dividend_low,unsigned divisor,struct div_result * dr)1102 void div128_by_32(u64 dividend_high, u64 dividend_low,
1103 unsigned divisor, struct div_result *dr)
1104 {
1105 unsigned long a, b, c, d;
1106 unsigned long w, x, y, z;
1107 u64 ra, rb, rc;
1108
1109 a = dividend_high >> 32;
1110 b = dividend_high & 0xffffffff;
1111 c = dividend_low >> 32;
1112 d = dividend_low & 0xffffffff;
1113
1114 w = a / divisor;
1115 ra = ((u64)(a - (w * divisor)) << 32) + b;
1116
1117 rb = ((u64) do_div(ra, divisor) << 32) + c;
1118 x = ra;
1119
1120 rc = ((u64) do_div(rb, divisor) << 32) + d;
1121 y = rb;
1122
1123 do_div(rc, divisor);
1124 z = rc;
1125
1126 dr->result_high = ((u64)w << 32) + x;
1127 dr->result_low = ((u64)y << 32) + z;
1128
1129 }
1130