2 * Common time routines among all ppc machines.
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)
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).
15 * Speeded up do_gettimeofday by getting rid of references to
16 * xtime (which required locks for consistency). (mikejc@us.ibm.com)
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.
26 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
27 * "A Kernel Model for Precision Timekeeping" by Dave Mills
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.
35 #include <linux/errno.h>
36 #include <linux/export.h>
37 #include <linux/sched.h>
38 #include <linux/kernel.h>
39 #include <linux/param.h>
40 #include <linux/string.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 #include <linux/delay.h>
56 #include <linux/irq_work.h>
57 #include <asm/trace.h>
60 #include <asm/processor.h>
61 #include <asm/nvram.h>
62 #include <asm/cache.h>
63 #include <asm/machdep.h>
64 #include <asm/uaccess.h>
68 #include <asm/div64.h>
70 #include <asm/vdso_datapage.h>
71 #include <asm/firmware.h>
72 #include <asm/cputime.h>
73 #ifdef CONFIG_PPC_ISERIES
74 #include <asm/iseries/it_lp_queue.h>
75 #include <asm/iseries/hv_call_xm.h>
78 /* powerpc clocksource/clockevent code */
80 #include <linux/clockchips.h>
81 #include <linux/clocksource.h>
83 static cycle_t
rtc_read(struct clocksource
*);
84 static struct clocksource clocksource_rtc
= {
87 .flags
= CLOCK_SOURCE_IS_CONTINUOUS
,
88 .mask
= CLOCKSOURCE_MASK(64),
92 static cycle_t
timebase_read(struct clocksource
*);
93 static struct clocksource clocksource_timebase
= {
96 .flags
= CLOCK_SOURCE_IS_CONTINUOUS
,
97 .mask
= CLOCKSOURCE_MASK(64),
98 .read
= timebase_read
,
101 #define DECREMENTER_MAX 0x7fffffff
103 static int decrementer_set_next_event(unsigned long evt
,
104 struct clock_event_device
*dev
);
105 static void decrementer_set_mode(enum clock_event_mode mode
,
106 struct clock_event_device
*dev
);
108 static struct clock_event_device decrementer_clockevent
= {
109 .name
= "decrementer",
112 .set_next_event
= decrementer_set_next_event
,
113 .set_mode
= decrementer_set_mode
,
114 .features
= CLOCK_EVT_FEAT_ONESHOT
,
117 struct decrementer_clock
{
118 struct clock_event_device event
;
122 static DEFINE_PER_CPU(struct decrementer_clock
, decrementers
);
124 #ifdef CONFIG_PPC_ISERIES
125 static unsigned long __initdata iSeries_recal_titan
;
126 static signed long __initdata iSeries_recal_tb
;
128 /* Forward declaration is only needed for iSereis compiles */
129 static void __init
clocksource_init(void);
132 #define XSEC_PER_SEC (1024*1024)
135 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
137 /* compute ((xsec << 12) * max) >> 32 */
138 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
141 unsigned long tb_ticks_per_jiffy
;
142 unsigned long tb_ticks_per_usec
= 100; /* sane default */
143 EXPORT_SYMBOL(tb_ticks_per_usec
);
144 unsigned long tb_ticks_per_sec
;
145 EXPORT_SYMBOL(tb_ticks_per_sec
); /* for cputime_t conversions */
147 DEFINE_SPINLOCK(rtc_lock
);
148 EXPORT_SYMBOL_GPL(rtc_lock
);
150 static u64 tb_to_ns_scale __read_mostly
;
151 static unsigned tb_to_ns_shift __read_mostly
;
152 static u64 boot_tb __read_mostly
;
154 extern struct timezone sys_tz
;
155 static long timezone_offset
;
157 unsigned long ppc_proc_freq
;
158 EXPORT_SYMBOL_GPL(ppc_proc_freq
);
159 unsigned long ppc_tb_freq
;
160 EXPORT_SYMBOL_GPL(ppc_tb_freq
);
162 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
164 * Factors for converting from cputime_t (timebase ticks) to
165 * jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds).
166 * These are all stored as 0.64 fixed-point binary fractions.
168 u64 __cputime_jiffies_factor
;
169 EXPORT_SYMBOL(__cputime_jiffies_factor
);
170 u64 __cputime_msec_factor
;
171 EXPORT_SYMBOL(__cputime_msec_factor
);
172 u64 __cputime_sec_factor
;
173 EXPORT_SYMBOL(__cputime_sec_factor
);
174 u64 __cputime_clockt_factor
;
175 EXPORT_SYMBOL(__cputime_clockt_factor
);
176 DEFINE_PER_CPU(unsigned long, cputime_last_delta
);
177 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta
);
179 cputime_t cputime_one_jiffy
;
181 void (*dtl_consumer
)(struct dtl_entry
*, u64
);
183 static void calc_cputime_factors(void)
185 struct div_result res
;
187 div128_by_32(HZ
, 0, tb_ticks_per_sec
, &res
);
188 __cputime_jiffies_factor
= res
.result_low
;
189 div128_by_32(1000, 0, tb_ticks_per_sec
, &res
);
190 __cputime_msec_factor
= res
.result_low
;
191 div128_by_32(1, 0, tb_ticks_per_sec
, &res
);
192 __cputime_sec_factor
= res
.result_low
;
193 div128_by_32(USER_HZ
, 0, tb_ticks_per_sec
, &res
);
194 __cputime_clockt_factor
= res
.result_low
;
198 * Read the SPURR on systems that have it, otherwise the PURR,
199 * or if that doesn't exist return the timebase value passed in.
201 static u64
read_spurr(u64 tb
)
203 if (cpu_has_feature(CPU_FTR_SPURR
))
204 return mfspr(SPRN_SPURR
);
205 if (cpu_has_feature(CPU_FTR_PURR
))
206 return mfspr(SPRN_PURR
);
210 #ifdef CONFIG_PPC_SPLPAR
213 * Scan the dispatch trace log and count up the stolen time.
214 * Should be called with interrupts disabled.
216 static u64
scan_dispatch_log(u64 stop_tb
)
218 u64 i
= local_paca
->dtl_ridx
;
219 struct dtl_entry
*dtl
= local_paca
->dtl_curr
;
220 struct dtl_entry
*dtl_end
= local_paca
->dispatch_log_end
;
221 struct lppaca
*vpa
= local_paca
->lppaca_ptr
;
229 if (i
== vpa
->dtl_idx
)
231 while (i
< vpa
->dtl_idx
) {
233 dtl_consumer(dtl
, i
);
235 tb_delta
= dtl
->enqueue_to_dispatch_time
+
236 dtl
->ready_to_enqueue_time
;
238 if (i
+ N_DISPATCH_LOG
< vpa
->dtl_idx
) {
239 /* buffer has overflowed */
240 i
= vpa
->dtl_idx
- N_DISPATCH_LOG
;
241 dtl
= local_paca
->dispatch_log
+ (i
% N_DISPATCH_LOG
);
250 dtl
= local_paca
->dispatch_log
;
252 local_paca
->dtl_ridx
= i
;
253 local_paca
->dtl_curr
= dtl
;
258 * Accumulate stolen time by scanning the dispatch trace log.
259 * Called on entry from user mode.
261 void accumulate_stolen_time(void)
265 u8 save_soft_enabled
= local_paca
->soft_enabled
;
266 u8 save_hard_enabled
= local_paca
->hard_enabled
;
268 /* We are called early in the exception entry, before
269 * soft/hard_enabled are sync'ed to the expected state
270 * for the exception. We are hard disabled but the PACA
271 * needs to reflect that so various debug stuff doesn't
274 local_paca
->soft_enabled
= 0;
275 local_paca
->hard_enabled
= 0;
277 sst
= scan_dispatch_log(local_paca
->starttime_user
);
278 ust
= scan_dispatch_log(local_paca
->starttime
);
279 local_paca
->system_time
-= sst
;
280 local_paca
->user_time
-= ust
;
281 local_paca
->stolen_time
+= ust
+ sst
;
283 local_paca
->soft_enabled
= save_soft_enabled
;
284 local_paca
->hard_enabled
= save_hard_enabled
;
287 static inline u64
calculate_stolen_time(u64 stop_tb
)
291 if (get_paca()->dtl_ridx
!= get_paca()->lppaca_ptr
->dtl_idx
) {
292 stolen
= scan_dispatch_log(stop_tb
);
293 get_paca()->system_time
-= stolen
;
296 stolen
+= get_paca()->stolen_time
;
297 get_paca()->stolen_time
= 0;
301 #else /* CONFIG_PPC_SPLPAR */
302 static inline u64
calculate_stolen_time(u64 stop_tb
)
307 #endif /* CONFIG_PPC_SPLPAR */
310 * Account time for a transition between system, hard irq
313 void account_system_vtime(struct task_struct
*tsk
)
315 u64 now
, nowscaled
, delta
, deltascaled
;
317 u64 stolen
, udelta
, sys_scaled
, user_scaled
;
319 local_irq_save(flags
);
321 nowscaled
= read_spurr(now
);
322 get_paca()->system_time
+= now
- get_paca()->starttime
;
323 get_paca()->starttime
= now
;
324 deltascaled
= nowscaled
- get_paca()->startspurr
;
325 get_paca()->startspurr
= nowscaled
;
327 stolen
= calculate_stolen_time(now
);
329 delta
= get_paca()->system_time
;
330 get_paca()->system_time
= 0;
331 udelta
= get_paca()->user_time
- get_paca()->utime_sspurr
;
332 get_paca()->utime_sspurr
= get_paca()->user_time
;
335 * Because we don't read the SPURR on every kernel entry/exit,
336 * deltascaled includes both user and system SPURR ticks.
337 * Apportion these ticks to system SPURR ticks and user
338 * SPURR ticks in the same ratio as the system time (delta)
339 * and user time (udelta) values obtained from the timebase
340 * over the same interval. The system ticks get accounted here;
341 * the user ticks get saved up in paca->user_time_scaled to be
342 * used by account_process_tick.
345 user_scaled
= udelta
;
346 if (deltascaled
!= delta
+ udelta
) {
348 sys_scaled
= deltascaled
* delta
/ (delta
+ udelta
);
349 user_scaled
= deltascaled
- sys_scaled
;
351 sys_scaled
= deltascaled
;
354 get_paca()->user_time_scaled
+= user_scaled
;
356 if (in_interrupt() || idle_task(smp_processor_id()) != tsk
) {
357 account_system_time(tsk
, 0, delta
, sys_scaled
);
359 account_steal_time(stolen
);
361 account_idle_time(delta
+ stolen
);
363 local_irq_restore(flags
);
365 EXPORT_SYMBOL_GPL(account_system_vtime
);
368 * Transfer the user and system times accumulated in the paca
369 * by the exception entry and exit code to the generic process
370 * user and system time records.
371 * Must be called with interrupts disabled.
372 * Assumes that account_system_vtime() has been called recently
373 * (i.e. since the last entry from usermode) so that
374 * get_paca()->user_time_scaled is up to date.
376 void account_process_tick(struct task_struct
*tsk
, int user_tick
)
378 cputime_t utime
, utimescaled
;
380 utime
= get_paca()->user_time
;
381 utimescaled
= get_paca()->user_time_scaled
;
382 get_paca()->user_time
= 0;
383 get_paca()->user_time_scaled
= 0;
384 get_paca()->utime_sspurr
= 0;
385 account_user_time(tsk
, utime
, utimescaled
);
388 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
389 #define calc_cputime_factors()
392 void __delay(unsigned long loops
)
400 /* the RTCL register wraps at 1000000000 */
401 diff
= get_rtcl() - start
;
404 } while (diff
< loops
);
407 while (get_tbl() - start
< loops
)
412 EXPORT_SYMBOL(__delay
);
414 void udelay(unsigned long usecs
)
416 __delay(tb_ticks_per_usec
* usecs
);
418 EXPORT_SYMBOL(udelay
);
421 unsigned long profile_pc(struct pt_regs
*regs
)
423 unsigned long pc
= instruction_pointer(regs
);
425 if (in_lock_functions(pc
))
430 EXPORT_SYMBOL(profile_pc
);
433 #ifdef CONFIG_PPC_ISERIES
436 * This function recalibrates the timebase based on the 49-bit time-of-day
437 * value in the Titan chip. The Titan is much more accurate than the value
438 * returned by the service processor for the timebase frequency.
441 static int __init
iSeries_tb_recal(void)
443 unsigned long titan
, tb
;
445 /* Make sure we only run on iSeries */
446 if (!firmware_has_feature(FW_FEATURE_ISERIES
))
450 titan
= HvCallXm_loadTod();
451 if ( iSeries_recal_titan
) {
452 unsigned long tb_ticks
= tb
- iSeries_recal_tb
;
453 unsigned long titan_usec
= (titan
- iSeries_recal_titan
) >> 12;
454 unsigned long new_tb_ticks_per_sec
= (tb_ticks
* USEC_PER_SEC
)/titan_usec
;
455 unsigned long new_tb_ticks_per_jiffy
=
456 DIV_ROUND_CLOSEST(new_tb_ticks_per_sec
, HZ
);
457 long tick_diff
= new_tb_ticks_per_jiffy
- tb_ticks_per_jiffy
;
459 /* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
460 new_tb_ticks_per_sec
= new_tb_ticks_per_jiffy
* HZ
;
462 if ( tick_diff
< 0 ) {
463 tick_diff
= -tick_diff
;
467 if ( tick_diff
< tb_ticks_per_jiffy
/25 ) {
468 printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
469 new_tb_ticks_per_jiffy
, sign
, tick_diff
);
470 tb_ticks_per_jiffy
= new_tb_ticks_per_jiffy
;
471 tb_ticks_per_sec
= new_tb_ticks_per_sec
;
472 calc_cputime_factors();
473 vdso_data
->tb_ticks_per_sec
= tb_ticks_per_sec
;
474 setup_cputime_one_jiffy();
477 printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
478 " new tb_ticks_per_jiffy = %lu\n"
479 " old tb_ticks_per_jiffy = %lu\n",
480 new_tb_ticks_per_jiffy
, tb_ticks_per_jiffy
);
484 iSeries_recal_titan
= titan
;
485 iSeries_recal_tb
= tb
;
487 /* Called here as now we know accurate values for the timebase */
491 late_initcall(iSeries_tb_recal
);
493 /* Called from platform early init */
494 void __init
iSeries_time_init_early(void)
496 iSeries_recal_tb
= get_tb();
497 iSeries_recal_titan
= HvCallXm_loadTod();
499 #endif /* CONFIG_PPC_ISERIES */
501 #ifdef CONFIG_IRQ_WORK
504 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
507 static inline unsigned long test_irq_work_pending(void)
511 asm volatile("lbz %0,%1(13)"
513 : "i" (offsetof(struct paca_struct
, irq_work_pending
)));
517 static inline void set_irq_work_pending_flag(void)
519 asm volatile("stb %0,%1(13)" : :
521 "i" (offsetof(struct paca_struct
, irq_work_pending
)));
524 static inline void clear_irq_work_pending(void)
526 asm volatile("stb %0,%1(13)" : :
528 "i" (offsetof(struct paca_struct
, irq_work_pending
)));
533 DEFINE_PER_CPU(u8
, irq_work_pending
);
535 #define set_irq_work_pending_flag() __get_cpu_var(irq_work_pending) = 1
536 #define test_irq_work_pending() __get_cpu_var(irq_work_pending)
537 #define clear_irq_work_pending() __get_cpu_var(irq_work_pending) = 0
539 #endif /* 32 vs 64 bit */
541 void arch_irq_work_raise(void)
544 set_irq_work_pending_flag();
549 #else /* CONFIG_IRQ_WORK */
551 #define test_irq_work_pending() 0
552 #define clear_irq_work_pending()
554 #endif /* CONFIG_IRQ_WORK */
557 * For iSeries shared processors, we have to let the hypervisor
558 * set the hardware decrementer. We set a virtual decrementer
559 * in the lppaca and call the hypervisor if the virtual
560 * decrementer is less than the current value in the hardware
561 * decrementer. (almost always the new decrementer value will
562 * be greater than the current hardware decementer so the hypervisor
563 * call will not be needed)
567 * timer_interrupt - gets called when the decrementer overflows,
568 * with interrupts disabled.
570 void timer_interrupt(struct pt_regs
* regs
)
572 struct pt_regs
*old_regs
;
573 struct decrementer_clock
*decrementer
= &__get_cpu_var(decrementers
);
574 struct clock_event_device
*evt
= &decrementer
->event
;
577 /* Ensure a positive value is written to the decrementer, or else
578 * some CPUs will continue to take decrementer exceptions.
580 set_dec(DECREMENTER_MAX
);
582 /* Some implementations of hotplug will get timer interrupts while
583 * offline, just ignore these
585 if (!cpu_online(smp_processor_id()))
588 trace_timer_interrupt_entry(regs
);
590 __get_cpu_var(irq_stat
).timer_irqs
++;
592 #if defined(CONFIG_PPC32) && defined(CONFIG_PMAC)
593 if (atomic_read(&ppc_n_lost_interrupts
) != 0)
597 old_regs
= set_irq_regs(regs
);
600 if (test_irq_work_pending()) {
601 clear_irq_work_pending();
605 #ifdef CONFIG_PPC_ISERIES
606 if (firmware_has_feature(FW_FEATURE_ISERIES
))
607 get_lppaca()->int_dword
.fields
.decr_int
= 0;
610 now
= get_tb_or_rtc();
611 if (now
>= decrementer
->next_tb
) {
612 decrementer
->next_tb
= ~(u64
)0;
613 if (evt
->event_handler
)
614 evt
->event_handler(evt
);
616 now
= decrementer
->next_tb
- now
;
617 if (now
<= DECREMENTER_MAX
)
621 #ifdef CONFIG_PPC_ISERIES
622 if (firmware_has_feature(FW_FEATURE_ISERIES
) && hvlpevent_is_pending())
623 process_hvlpevents();
627 /* collect purr register values often, for accurate calculations */
628 if (firmware_has_feature(FW_FEATURE_SPLPAR
)) {
629 struct cpu_usage
*cu
= &__get_cpu_var(cpu_usage_array
);
630 cu
->current_tb
= mfspr(SPRN_PURR
);
635 set_irq_regs(old_regs
);
637 trace_timer_interrupt_exit(regs
);
640 #ifdef CONFIG_SUSPEND
641 static void generic_suspend_disable_irqs(void)
643 /* Disable the decrementer, so that it doesn't interfere
652 static void generic_suspend_enable_irqs(void)
657 /* Overrides the weak version in kernel/power/main.c */
658 void arch_suspend_disable_irqs(void)
660 if (ppc_md
.suspend_disable_irqs
)
661 ppc_md
.suspend_disable_irqs();
662 generic_suspend_disable_irqs();
665 /* Overrides the weak version in kernel/power/main.c */
666 void arch_suspend_enable_irqs(void)
668 generic_suspend_enable_irqs();
669 if (ppc_md
.suspend_enable_irqs
)
670 ppc_md
.suspend_enable_irqs();
675 * Scheduler clock - returns current time in nanosec units.
677 * Note: mulhdu(a, b) (multiply high double unsigned) returns
678 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
679 * are 64-bit unsigned numbers.
681 unsigned long long sched_clock(void)
685 return mulhdu(get_tb() - boot_tb
, tb_to_ns_scale
) << tb_to_ns_shift
;
688 static int __init
get_freq(char *name
, int cells
, unsigned long *val
)
690 struct device_node
*cpu
;
691 const unsigned int *fp
;
694 /* The cpu node should have timebase and clock frequency properties */
695 cpu
= of_find_node_by_type(NULL
, "cpu");
698 fp
= of_get_property(cpu
, name
, NULL
);
701 *val
= of_read_ulong(fp
, cells
);
710 /* should become __cpuinit when secondary_cpu_time_init also is */
711 void start_cpu_decrementer(void)
713 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
714 /* Clear any pending timer interrupts */
715 mtspr(SPRN_TSR
, TSR_ENW
| TSR_WIS
| TSR_DIS
| TSR_FIS
);
717 /* Enable decrementer interrupt */
718 mtspr(SPRN_TCR
, TCR_DIE
);
719 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
722 void __init
generic_calibrate_decr(void)
724 ppc_tb_freq
= DEFAULT_TB_FREQ
; /* hardcoded default */
726 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq
) &&
727 !get_freq("timebase-frequency", 1, &ppc_tb_freq
)) {
729 printk(KERN_ERR
"WARNING: Estimating decrementer frequency "
733 ppc_proc_freq
= DEFAULT_PROC_FREQ
; /* hardcoded default */
735 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq
) &&
736 !get_freq("clock-frequency", 1, &ppc_proc_freq
)) {
738 printk(KERN_ERR
"WARNING: Estimating processor frequency "
743 int update_persistent_clock(struct timespec now
)
747 if (!ppc_md
.set_rtc_time
)
750 to_tm(now
.tv_sec
+ 1 + timezone_offset
, &tm
);
754 return ppc_md
.set_rtc_time(&tm
);
757 static void __read_persistent_clock(struct timespec
*ts
)
760 static int first
= 1;
763 /* XXX this is a litle fragile but will work okay in the short term */
766 if (ppc_md
.time_init
)
767 timezone_offset
= ppc_md
.time_init();
769 /* get_boot_time() isn't guaranteed to be safe to call late */
770 if (ppc_md
.get_boot_time
) {
771 ts
->tv_sec
= ppc_md
.get_boot_time() - timezone_offset
;
775 if (!ppc_md
.get_rtc_time
) {
779 ppc_md
.get_rtc_time(&tm
);
781 ts
->tv_sec
= mktime(tm
.tm_year
+1900, tm
.tm_mon
+1, tm
.tm_mday
,
782 tm
.tm_hour
, tm
.tm_min
, tm
.tm_sec
);
785 void read_persistent_clock(struct timespec
*ts
)
787 __read_persistent_clock(ts
);
789 /* Sanitize it in case real time clock is set below EPOCH */
790 if (ts
->tv_sec
< 0) {
797 /* clocksource code */
798 static cycle_t
rtc_read(struct clocksource
*cs
)
800 return (cycle_t
)get_rtc();
803 static cycle_t
timebase_read(struct clocksource
*cs
)
805 return (cycle_t
)get_tb();
808 void update_vsyscall(struct timespec
*wall_time
, struct timespec
*wtm
,
809 struct clocksource
*clock
, u32 mult
)
811 u64 new_tb_to_xs
, new_stamp_xsec
;
814 if (clock
!= &clocksource_timebase
)
817 /* Make userspace gettimeofday spin until we're done. */
818 ++vdso_data
->tb_update_count
;
821 /* 19342813113834067 ~= 2^(20+64) / 1e9 */
822 new_tb_to_xs
= (u64
) mult
* (19342813113834067ULL >> clock
->shift
);
823 new_stamp_xsec
= (u64
) wall_time
->tv_nsec
* XSEC_PER_SEC
;
824 do_div(new_stamp_xsec
, 1000000000);
825 new_stamp_xsec
+= (u64
) wall_time
->tv_sec
* XSEC_PER_SEC
;
827 BUG_ON(wall_time
->tv_nsec
>= NSEC_PER_SEC
);
828 /* this is tv_nsec / 1e9 as a 0.32 fraction */
829 frac_sec
= ((u64
) wall_time
->tv_nsec
* 18446744073ULL) >> 32;
832 * tb_update_count is used to allow the userspace gettimeofday code
833 * to assure itself that it sees a consistent view of the tb_to_xs and
834 * stamp_xsec variables. It reads the tb_update_count, then reads
835 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
836 * the two values of tb_update_count match and are even then the
837 * tb_to_xs and stamp_xsec values are consistent. If not, then it
838 * loops back and reads them again until this criteria is met.
839 * We expect the caller to have done the first increment of
840 * vdso_data->tb_update_count already.
842 vdso_data
->tb_orig_stamp
= clock
->cycle_last
;
843 vdso_data
->stamp_xsec
= new_stamp_xsec
;
844 vdso_data
->tb_to_xs
= new_tb_to_xs
;
845 vdso_data
->wtom_clock_sec
= wtm
->tv_sec
;
846 vdso_data
->wtom_clock_nsec
= wtm
->tv_nsec
;
847 vdso_data
->stamp_xtime
= *wall_time
;
848 vdso_data
->stamp_sec_fraction
= frac_sec
;
850 ++(vdso_data
->tb_update_count
);
853 void update_vsyscall_tz(void)
855 /* Make userspace gettimeofday spin until we're done. */
856 ++vdso_data
->tb_update_count
;
858 vdso_data
->tz_minuteswest
= sys_tz
.tz_minuteswest
;
859 vdso_data
->tz_dsttime
= sys_tz
.tz_dsttime
;
861 ++vdso_data
->tb_update_count
;
864 static void __init
clocksource_init(void)
866 struct clocksource
*clock
;
869 clock
= &clocksource_rtc
;
871 clock
= &clocksource_timebase
;
873 if (clocksource_register_hz(clock
, tb_ticks_per_sec
)) {
874 printk(KERN_ERR
"clocksource: %s is already registered\n",
879 printk(KERN_INFO
"clocksource: %s mult[%x] shift[%d] registered\n",
880 clock
->name
, clock
->mult
, clock
->shift
);
883 void decrementer_check_overflow(void)
885 u64 now
= get_tb_or_rtc();
886 struct decrementer_clock
*decrementer
= &__get_cpu_var(decrementers
);
888 if (now
>= decrementer
->next_tb
)
892 static int decrementer_set_next_event(unsigned long evt
,
893 struct clock_event_device
*dev
)
895 __get_cpu_var(decrementers
).next_tb
= get_tb_or_rtc() + evt
;
900 static void decrementer_set_mode(enum clock_event_mode mode
,
901 struct clock_event_device
*dev
)
903 if (mode
!= CLOCK_EVT_MODE_ONESHOT
)
904 decrementer_set_next_event(DECREMENTER_MAX
, dev
);
907 static void register_decrementer_clockevent(int cpu
)
909 struct clock_event_device
*dec
= &per_cpu(decrementers
, cpu
).event
;
911 *dec
= decrementer_clockevent
;
912 dec
->cpumask
= cpumask_of(cpu
);
914 printk_once(KERN_DEBUG
"clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
915 dec
->name
, dec
->mult
, dec
->shift
, cpu
);
917 clockevents_register_device(dec
);
920 static void __init
init_decrementer_clockevent(void)
922 int cpu
= smp_processor_id();
924 clockevents_calc_mult_shift(&decrementer_clockevent
, ppc_tb_freq
, 4);
926 decrementer_clockevent
.max_delta_ns
=
927 clockevent_delta2ns(DECREMENTER_MAX
, &decrementer_clockevent
);
928 decrementer_clockevent
.min_delta_ns
=
929 clockevent_delta2ns(2, &decrementer_clockevent
);
931 register_decrementer_clockevent(cpu
);
934 void secondary_cpu_time_init(void)
936 /* Start the decrementer on CPUs that have manual control
939 start_cpu_decrementer();
941 /* FIME: Should make unrelatred change to move snapshot_timebase
943 register_decrementer_clockevent(smp_processor_id());
946 /* This function is only called on the boot processor */
947 void __init
time_init(void)
949 struct div_result res
;
954 /* 601 processor: dec counts down by 128 every 128ns */
955 ppc_tb_freq
= 1000000000;
957 /* Normal PowerPC with timebase register */
958 ppc_md
.calibrate_decr();
959 printk(KERN_DEBUG
"time_init: decrementer frequency = %lu.%.6lu MHz\n",
960 ppc_tb_freq
/ 1000000, ppc_tb_freq
% 1000000);
961 printk(KERN_DEBUG
"time_init: processor frequency = %lu.%.6lu MHz\n",
962 ppc_proc_freq
/ 1000000, ppc_proc_freq
% 1000000);
965 tb_ticks_per_jiffy
= ppc_tb_freq
/ HZ
;
966 tb_ticks_per_sec
= ppc_tb_freq
;
967 tb_ticks_per_usec
= ppc_tb_freq
/ 1000000;
968 calc_cputime_factors();
969 setup_cputime_one_jiffy();
972 * Compute scale factor for sched_clock.
973 * The calibrate_decr() function has set tb_ticks_per_sec,
974 * which is the timebase frequency.
975 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
976 * the 128-bit result as a 64.64 fixed-point number.
977 * We then shift that number right until it is less than 1.0,
978 * giving us the scale factor and shift count to use in
981 div128_by_32(1000000000, 0, tb_ticks_per_sec
, &res
);
982 scale
= res
.result_low
;
983 for (shift
= 0; res
.result_high
!= 0; ++shift
) {
984 scale
= (scale
>> 1) | (res
.result_high
<< 63);
985 res
.result_high
>>= 1;
987 tb_to_ns_scale
= scale
;
988 tb_to_ns_shift
= shift
;
989 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
990 boot_tb
= get_tb_or_rtc();
992 /* If platform provided a timezone (pmac), we correct the time */
993 if (timezone_offset
) {
994 sys_tz
.tz_minuteswest
= -timezone_offset
/ 60;
995 sys_tz
.tz_dsttime
= 0;
998 vdso_data
->tb_update_count
= 0;
999 vdso_data
->tb_ticks_per_sec
= tb_ticks_per_sec
;
1001 /* Start the decrementer on CPUs that have manual control
1004 start_cpu_decrementer();
1006 /* Register the clocksource, if we're not running on iSeries */
1007 if (!firmware_has_feature(FW_FEATURE_ISERIES
))
1010 init_decrementer_clockevent();
1015 #define STARTOFTIME 1970
1016 #define SECDAY 86400L
1017 #define SECYR (SECDAY * 365)
1018 #define leapyear(year) ((year) % 4 == 0 && \
1019 ((year) % 100 != 0 || (year) % 400 == 0))
1020 #define days_in_year(a) (leapyear(a) ? 366 : 365)
1021 #define days_in_month(a) (month_days[(a) - 1])
1023 static int month_days
[12] = {
1024 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1028 * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1030 void GregorianDay(struct rtc_time
* tm
)
1035 int MonthOffset
[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1037 lastYear
= tm
->tm_year
- 1;
1040 * Number of leap corrections to apply up to end of last year
1042 leapsToDate
= lastYear
/ 4 - lastYear
/ 100 + lastYear
/ 400;
1045 * This year is a leap year if it is divisible by 4 except when it is
1046 * divisible by 100 unless it is divisible by 400
1048 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1050 day
= tm
->tm_mon
> 2 && leapyear(tm
->tm_year
);
1052 day
+= lastYear
*365 + leapsToDate
+ MonthOffset
[tm
->tm_mon
-1] +
1055 tm
->tm_wday
= day
% 7;
1058 void to_tm(int tim
, struct rtc_time
* tm
)
1061 register long hms
, day
;
1066 /* Hours, minutes, seconds are easy */
1067 tm
->tm_hour
= hms
/ 3600;
1068 tm
->tm_min
= (hms
% 3600) / 60;
1069 tm
->tm_sec
= (hms
% 3600) % 60;
1071 /* Number of years in days */
1072 for (i
= STARTOFTIME
; day
>= days_in_year(i
); i
++)
1073 day
-= days_in_year(i
);
1076 /* Number of months in days left */
1077 if (leapyear(tm
->tm_year
))
1078 days_in_month(FEBRUARY
) = 29;
1079 for (i
= 1; day
>= days_in_month(i
); i
++)
1080 day
-= days_in_month(i
);
1081 days_in_month(FEBRUARY
) = 28;
1084 /* Days are what is left over (+1) from all that. */
1085 tm
->tm_mday
= day
+ 1;
1088 * Determine the day of week
1094 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1097 void div128_by_32(u64 dividend_high
, u64 dividend_low
,
1098 unsigned divisor
, struct div_result
*dr
)
1100 unsigned long a
, b
, c
, d
;
1101 unsigned long w
, x
, y
, z
;
1104 a
= dividend_high
>> 32;
1105 b
= dividend_high
& 0xffffffff;
1106 c
= dividend_low
>> 32;
1107 d
= dividend_low
& 0xffffffff;
1110 ra
= ((u64
)(a
- (w
* divisor
)) << 32) + b
;
1112 rb
= ((u64
) do_div(ra
, divisor
) << 32) + c
;
1115 rc
= ((u64
) do_div(rb
, divisor
) << 32) + d
;
1118 do_div(rc
, divisor
);
1121 dr
->result_high
= ((u64
)w
<< 32) + x
;
1122 dr
->result_low
= ((u64
)y
<< 32) + z
;
1126 /* We don't need to calibrate delay, we use the CPU timebase for that */
1127 void calibrate_delay(void)
1129 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1130 * as the number of __delay(1) in a jiffy, so make it so
1132 loops_per_jiffy
= tb_ticks_per_jiffy
;
1135 static int __init
rtc_init(void)
1137 struct platform_device
*pdev
;
1139 if (!ppc_md
.get_rtc_time
)
1142 pdev
= platform_device_register_simple("rtc-generic", -1, NULL
, 0);
1144 return PTR_ERR(pdev
);
1149 module_init(rtc_init
);