2 * linux/arch/x86-64/kernel/time.c
4 * "High Precision Event Timer" based timekeeping.
6 * Copyright (c) 1991,1992,1995 Linus Torvalds
7 * Copyright (c) 1994 Alan Modra
8 * Copyright (c) 1995 Markus Kuhn
9 * Copyright (c) 1996 Ingo Molnar
10 * Copyright (c) 1998 Andrea Arcangeli
11 * Copyright (c) 2002 Vojtech Pavlik
12 * Copyright (c) 2003 Andi Kleen
13 * RTC support code taken from arch/i386/kernel/timers/time_hpet.c
16 #include <linux/kernel.h>
17 #include <linux/sched.h>
18 #include <linux/interrupt.h>
19 #include <linux/init.h>
20 #include <linux/mc146818rtc.h>
21 #include <linux/irq.h>
22 #include <linux/time.h>
23 #include <linux/ioport.h>
24 #include <linux/module.h>
25 #include <linux/device.h>
26 #include <linux/sysdev.h>
27 #include <linux/bcd.h>
28 #include <linux/kallsyms.h>
29 #include <linux/acpi.h>
31 #include <acpi/achware.h> /* for PM timer frequency */
33 #include <asm/8253pit.h>
34 #include <asm/pgtable.h>
35 #include <asm/vsyscall.h>
36 #include <asm/timex.h>
37 #include <asm/proto.h>
39 #include <asm/sections.h>
40 #include <linux/cpufreq.h>
41 #include <linux/hpet.h>
42 #ifdef CONFIG_X86_LOCAL_APIC
46 u64 jiffies_64 = INITIAL_JIFFIES;
48 EXPORT_SYMBOL(jiffies_64);
50 #ifdef CONFIG_CPU_FREQ
51 static void cpufreq_delayed_get(void);
53 extern void i8254_timer_resume(void);
54 extern int using_apic_timer;
56 DEFINE_SPINLOCK(rtc_lock);
57 DEFINE_SPINLOCK(i8253_lock);
59 static int nohpet __initdata = 0;
60 static int notsc __initdata = 0;
62 #undef HPET_HACK_ENABLE_DANGEROUS
64 unsigned int cpu_khz; /* TSC clocks / usec, not used here */
65 static unsigned long hpet_period; /* fsecs / HPET clock */
66 unsigned long hpet_tick; /* HPET clocks / interrupt */
67 unsigned long vxtime_hz = PIT_TICK_RATE;
68 int report_lost_ticks; /* command line option */
69 unsigned long long monotonic_base;
71 struct vxtime_data __vxtime __section_vxtime; /* for vsyscalls */
73 volatile unsigned long __jiffies __section_jiffies = INITIAL_JIFFIES;
74 unsigned long __wall_jiffies __section_wall_jiffies = INITIAL_JIFFIES;
75 struct timespec __xtime __section_xtime;
76 struct timezone __sys_tz __section_sys_tz;
78 static inline void rdtscll_sync(unsigned long *tsc)
87 * do_gettimeoffset() returns microseconds since last timer interrupt was
88 * triggered by hardware. A memory read of HPET is slower than a register read
89 * of TSC, but much more reliable. It's also synchronized to the timer
90 * interrupt. Note that do_gettimeoffset() may return more than hpet_tick, if a
91 * timer interrupt has happened already, but vxtime.trigger wasn't updated yet.
92 * This is not a problem, because jiffies hasn't updated either. They are bound
93 * together by xtime_lock.
96 static inline unsigned int do_gettimeoffset_tsc(void)
101 if (t < vxtime.last_tsc) t = vxtime.last_tsc; /* hack */
102 x = ((t - vxtime.last_tsc) * vxtime.tsc_quot) >> 32;
106 static inline unsigned int do_gettimeoffset_hpet(void)
108 return ((hpet_readl(HPET_COUNTER) - vxtime.last) * vxtime.quot) >> 32;
111 unsigned int (*do_gettimeoffset)(void) = do_gettimeoffset_tsc;
114 * This version of gettimeofday() has microsecond resolution and better than
115 * microsecond precision, as we're using at least a 10 MHz (usually 14.31818
119 void do_gettimeofday(struct timeval *tv)
121 unsigned long seq, t;
122 unsigned int sec, usec;
125 seq = read_seqbegin(&xtime_lock);
128 usec = xtime.tv_nsec / 1000;
130 /* i386 does some correction here to keep the clock
131 monotonous even when ntpd is fixing drift.
132 But they didn't work for me, there is a non monotonic
133 clock anyways with ntp.
134 I dropped all corrections now until a real solution can
135 be found. Note when you fix it here you need to do the same
136 in arch/x86_64/kernel/vsyscall.c and export all needed
137 variables in vmlinux.lds. -AK */
139 t = (jiffies - wall_jiffies) * (1000000L / HZ) +
143 } while (read_seqretry(&xtime_lock, seq));
145 tv->tv_sec = sec + usec / 1000000;
146 tv->tv_usec = usec % 1000000;
149 EXPORT_SYMBOL(do_gettimeofday);
152 * settimeofday() first undoes the correction that gettimeofday would do
153 * on the time, and then saves it. This is ugly, but has been like this for
157 int do_settimeofday(struct timespec *tv)
159 time_t wtm_sec, sec = tv->tv_sec;
160 long wtm_nsec, nsec = tv->tv_nsec;
162 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
165 write_seqlock_irq(&xtime_lock);
167 nsec -= do_gettimeoffset() * 1000 +
168 (jiffies - wall_jiffies) * (NSEC_PER_SEC/HZ);
170 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
171 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
173 set_normalized_timespec(&xtime, sec, nsec);
174 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
176 time_adjust = 0; /* stop active adjtime() */
177 time_status |= STA_UNSYNC;
178 time_maxerror = NTP_PHASE_LIMIT;
179 time_esterror = NTP_PHASE_LIMIT;
181 write_sequnlock_irq(&xtime_lock);
186 EXPORT_SYMBOL(do_settimeofday);
188 unsigned long profile_pc(struct pt_regs *regs)
190 unsigned long pc = instruction_pointer(regs);
192 /* Assume the lock function has either no stack frame or only a single word.
193 This checks if the address on the stack looks like a kernel text address.
194 There is a small window for false hits, but in that case the tick
195 is just accounted to the spinlock function.
196 Better would be to write these functions in assembler again
197 and check exactly. */
198 if (in_lock_functions(pc)) {
199 char *v = *(char **)regs->rsp;
200 if ((v >= _stext && v <= _etext) ||
201 (v >= _sinittext && v <= _einittext) ||
202 (v >= (char *)MODULES_VADDR && v <= (char *)MODULES_END))
203 return (unsigned long)v;
204 return ((unsigned long *)regs->rsp)[1];
208 EXPORT_SYMBOL(profile_pc);
211 * In order to set the CMOS clock precisely, set_rtc_mmss has to be called 500
212 * ms after the second nowtime has started, because when nowtime is written
213 * into the registers of the CMOS clock, it will jump to the next second
214 * precisely 500 ms later. Check the Motorola MC146818A or Dallas DS12887 data
218 static void set_rtc_mmss(unsigned long nowtime)
220 int real_seconds, real_minutes, cmos_minutes;
221 unsigned char control, freq_select;
224 * IRQs are disabled when we're called from the timer interrupt,
225 * no need for spin_lock_irqsave()
228 spin_lock(&rtc_lock);
231 * Tell the clock it's being set and stop it.
234 control = CMOS_READ(RTC_CONTROL);
235 CMOS_WRITE(control | RTC_SET, RTC_CONTROL);
237 freq_select = CMOS_READ(RTC_FREQ_SELECT);
238 CMOS_WRITE(freq_select | RTC_DIV_RESET2, RTC_FREQ_SELECT);
240 cmos_minutes = CMOS_READ(RTC_MINUTES);
241 BCD_TO_BIN(cmos_minutes);
244 * since we're only adjusting minutes and seconds, don't interfere with hour
245 * overflow. This avoids messing with unknown time zones but requires your RTC
246 * not to be off by more than 15 minutes. Since we're calling it only when
247 * our clock is externally synchronized using NTP, this shouldn't be a problem.
250 real_seconds = nowtime % 60;
251 real_minutes = nowtime / 60;
252 if (((abs(real_minutes - cmos_minutes) + 15) / 30) & 1)
253 real_minutes += 30; /* correct for half hour time zone */
257 /* AMD 8111 is a really bad time keeper and hits this regularly.
258 It probably was an attempt to avoid screwing up DST, but ignore
260 if (abs(real_minutes - cmos_minutes) >= 30) {
261 printk(KERN_WARNING "time.c: can't update CMOS clock "
262 "from %d to %d\n", cmos_minutes, real_minutes);
267 BIN_TO_BCD(real_seconds);
268 BIN_TO_BCD(real_minutes);
269 CMOS_WRITE(real_seconds, RTC_SECONDS);
270 CMOS_WRITE(real_minutes, RTC_MINUTES);
274 * The following flags have to be released exactly in this order, otherwise the
275 * DS12887 (popular MC146818A clone with integrated battery and quartz) will
276 * not reset the oscillator and will not update precisely 500 ms later. You
277 * won't find this mentioned in the Dallas Semiconductor data sheets, but who
278 * believes data sheets anyway ... -- Markus Kuhn
281 CMOS_WRITE(control, RTC_CONTROL);
282 CMOS_WRITE(freq_select, RTC_FREQ_SELECT);
284 spin_unlock(&rtc_lock);
288 /* monotonic_clock(): returns # of nanoseconds passed since time_init()
289 * Note: This function is required to return accurate
290 * time even in the absence of multiple timer ticks.
292 unsigned long long monotonic_clock(void)
295 u32 last_offset, this_offset, offset;
296 unsigned long long base;
298 if (vxtime.mode == VXTIME_HPET) {
300 seq = read_seqbegin(&xtime_lock);
302 last_offset = vxtime.last;
303 base = monotonic_base;
304 this_offset = hpet_readl(HPET_T0_CMP) - hpet_tick;
306 } while (read_seqretry(&xtime_lock, seq));
307 offset = (this_offset - last_offset);
308 offset *=(NSEC_PER_SEC/HZ)/hpet_tick;
309 return base + offset;
312 seq = read_seqbegin(&xtime_lock);
314 last_offset = vxtime.last_tsc;
315 base = monotonic_base;
316 } while (read_seqretry(&xtime_lock, seq));
318 rdtscll(this_offset);
319 offset = (this_offset - last_offset)*1000/cpu_khz;
320 return base + offset;
325 EXPORT_SYMBOL(monotonic_clock);
327 static noinline void handle_lost_ticks(int lost, struct pt_regs *regs)
329 static long lost_count;
332 if (report_lost_ticks) {
333 printk(KERN_WARNING "time.c: Lost %d timer "
335 print_symbol("rip %s)\n", regs->rip);
338 if (lost_count == 1000 && !warned) {
340 "warning: many lost ticks.\n"
341 KERN_WARNING "Your time source seems to be instable or "
342 "some driver is hogging interupts\n");
343 print_symbol("rip %s\n", regs->rip);
344 if (vxtime.mode == VXTIME_TSC && vxtime.hpet_address) {
345 printk(KERN_WARNING "Falling back to HPET\n");
346 vxtime.last = hpet_readl(HPET_T0_CMP) - hpet_tick;
347 vxtime.mode = VXTIME_HPET;
348 do_gettimeoffset = do_gettimeoffset_hpet;
350 /* else should fall back to PIT, but code missing. */
355 #ifdef CONFIG_CPU_FREQ
356 /* In some cases the CPU can change frequency without us noticing
357 (like going into thermal throttle)
358 Give cpufreq a change to catch up. */
359 if ((lost_count+1) % 25 == 0) {
360 cpufreq_delayed_get();
365 static irqreturn_t timer_interrupt(int irq, void *dev_id, struct pt_regs *regs)
367 static unsigned long rtc_update = 0;
369 int delay, offset = 0, lost = 0;
372 * Here we are in the timer irq handler. We have irqs locally disabled (so we
373 * don't need spin_lock_irqsave()) but we don't know if the timer_bh is running
374 * on the other CPU, so we need a lock. We also need to lock the vsyscall
375 * variables, because both do_timer() and us change them -arca+vojtech
378 write_seqlock(&xtime_lock);
380 if (vxtime.hpet_address) {
381 offset = hpet_readl(HPET_T0_CMP) - hpet_tick;
382 delay = hpet_readl(HPET_COUNTER) - offset;
384 spin_lock(&i8253_lock);
387 delay |= inb(0x40) << 8;
388 spin_unlock(&i8253_lock);
389 delay = LATCH - 1 - delay;
394 if (vxtime.mode == VXTIME_HPET) {
395 if (offset - vxtime.last > hpet_tick) {
396 lost = (offset - vxtime.last) / hpet_tick - 1;
400 (offset - vxtime.last)*(NSEC_PER_SEC/HZ) / hpet_tick;
402 vxtime.last = offset;
403 #ifdef CONFIG_X86_PM_TIMER
404 } else if (vxtime.mode == VXTIME_PMTMR) {
405 lost = pmtimer_mark_offset();
408 offset = (((tsc - vxtime.last_tsc) *
409 vxtime.tsc_quot) >> 32) - (USEC_PER_SEC / HZ);
414 if (offset > (USEC_PER_SEC / HZ)) {
415 lost = offset / (USEC_PER_SEC / HZ);
416 offset %= (USEC_PER_SEC / HZ);
419 monotonic_base += (tsc - vxtime.last_tsc)*1000000/cpu_khz ;
421 vxtime.last_tsc = tsc - vxtime.quot * delay / vxtime.tsc_quot;
423 if ((((tsc - vxtime.last_tsc) *
424 vxtime.tsc_quot) >> 32) < offset)
425 vxtime.last_tsc = tsc -
426 (((long) offset << 32) / vxtime.tsc_quot) - 1;
430 handle_lost_ticks(lost, regs);
435 * Do the timer stuff.
440 update_process_times(user_mode(regs));
444 * In the SMP case we use the local APIC timer interrupt to do the profiling,
445 * except when we simulate SMP mode on a uniprocessor system, in that case we
446 * have to call the local interrupt handler.
449 #ifndef CONFIG_X86_LOCAL_APIC
450 profile_tick(CPU_PROFILING, regs);
452 if (!using_apic_timer)
453 smp_local_timer_interrupt(regs);
457 * If we have an externally synchronized Linux clock, then update CMOS clock
458 * accordingly every ~11 minutes. set_rtc_mmss() will be called in the jiffy
459 * closest to exactly 500 ms before the next second. If the update fails, we
460 * don't care, as it'll be updated on the next turn, and the problem (time way
461 * off) isn't likely to go away much sooner anyway.
464 if ((~time_status & STA_UNSYNC) && xtime.tv_sec > rtc_update &&
465 abs(xtime.tv_nsec - 500000000) <= tick_nsec / 2) {
466 set_rtc_mmss(xtime.tv_sec);
467 rtc_update = xtime.tv_sec + 660;
470 write_sequnlock(&xtime_lock);
475 static unsigned int cyc2ns_scale;
476 #define CYC2NS_SCALE_FACTOR 10 /* 2^10, carefully chosen */
478 static inline void set_cyc2ns_scale(unsigned long cpu_mhz)
480 cyc2ns_scale = (1000 << CYC2NS_SCALE_FACTOR)/cpu_mhz;
483 static inline unsigned long long cycles_2_ns(unsigned long long cyc)
485 return (cyc * cyc2ns_scale) >> CYC2NS_SCALE_FACTOR;
488 unsigned long long sched_clock(void)
493 /* Don't do a HPET read here. Using TSC always is much faster
494 and HPET may not be mapped yet when the scheduler first runs.
495 Disadvantage is a small drift between CPUs in some configurations,
496 but that should be tolerable. */
497 if (__vxtime.mode == VXTIME_HPET)
498 return (hpet_readl(HPET_COUNTER) * vxtime.quot) >> 32;
501 /* Could do CPU core sync here. Opteron can execute rdtsc speculatively,
502 which means it is not completely exact and may not be monotonous between
503 CPUs. But the errors should be too small to matter for scheduling
507 return cycles_2_ns(a);
510 unsigned long get_cmos_time(void)
512 unsigned int timeout, year, mon, day, hour, min, sec;
513 unsigned char last, this;
517 * The Linux interpretation of the CMOS clock register contents: When the
518 * Update-In-Progress (UIP) flag goes from 1 to 0, the RTC registers show the
519 * second which has precisely just started. Waiting for this can take up to 1
520 * second, we timeout approximately after 2.4 seconds on a machine with
521 * standard 8.3 MHz ISA bus.
524 spin_lock_irqsave(&rtc_lock, flags);
529 while (timeout && last && !this) {
531 this = CMOS_READ(RTC_FREQ_SELECT) & RTC_UIP;
536 * Here we are safe to assume the registers won't change for a whole second, so
537 * we just go ahead and read them.
540 sec = CMOS_READ(RTC_SECONDS);
541 min = CMOS_READ(RTC_MINUTES);
542 hour = CMOS_READ(RTC_HOURS);
543 day = CMOS_READ(RTC_DAY_OF_MONTH);
544 mon = CMOS_READ(RTC_MONTH);
545 year = CMOS_READ(RTC_YEAR);
547 spin_unlock_irqrestore(&rtc_lock, flags);
550 * We know that x86-64 always uses BCD format, no need to check the config
562 * x86-64 systems only exists since 2002.
563 * This will work up to Dec 31, 2100
567 return mktime(year, mon, day, hour, min, sec);
570 #ifdef CONFIG_CPU_FREQ
572 /* Frequency scaling support. Adjust the TSC based timer when the cpu frequency
575 RED-PEN: On SMP we assume all CPUs run with the same frequency. It's
576 not that important because current Opteron setups do not support
577 scaling on SMP anyroads.
579 Should fix up last_tsc too. Currently gettimeofday in the
580 first tick after the change will be slightly wrong. */
582 #include <linux/workqueue.h>
584 static unsigned int cpufreq_delayed_issched = 0;
585 static unsigned int cpufreq_init = 0;
586 static struct work_struct cpufreq_delayed_get_work;
588 static void handle_cpufreq_delayed_get(void *v)
591 for_each_online_cpu(cpu) {
594 cpufreq_delayed_issched = 0;
597 /* if we notice lost ticks, schedule a call to cpufreq_get() as it tries
598 * to verify the CPU frequency the timing core thinks the CPU is running
599 * at is still correct.
601 static void cpufreq_delayed_get(void)
604 if (cpufreq_init && !cpufreq_delayed_issched) {
605 cpufreq_delayed_issched = 1;
608 printk(KERN_DEBUG "Losing some ticks... checking if CPU frequency changed.\n");
610 schedule_work(&cpufreq_delayed_get_work);
614 static unsigned int ref_freq = 0;
615 static unsigned long loops_per_jiffy_ref = 0;
617 static unsigned long cpu_khz_ref = 0;
619 static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
622 struct cpufreq_freqs *freq = data;
623 unsigned long *lpj, dummy;
625 if (cpu_has(&cpu_data[freq->cpu], X86_FEATURE_CONSTANT_TSC))
629 if (!(freq->flags & CPUFREQ_CONST_LOOPS))
631 lpj = &cpu_data[freq->cpu].loops_per_jiffy;
633 lpj = &boot_cpu_data.loops_per_jiffy;
637 ref_freq = freq->old;
638 loops_per_jiffy_ref = *lpj;
639 cpu_khz_ref = cpu_khz;
641 if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) ||
642 (val == CPUFREQ_POSTCHANGE && freq->old > freq->new) ||
643 (val == CPUFREQ_RESUMECHANGE)) {
645 cpufreq_scale(loops_per_jiffy_ref, ref_freq, freq->new);
647 cpu_khz = cpufreq_scale(cpu_khz_ref, ref_freq, freq->new);
648 if (!(freq->flags & CPUFREQ_CONST_LOOPS))
649 vxtime.tsc_quot = (1000L << 32) / cpu_khz;
652 set_cyc2ns_scale(cpu_khz_ref / 1000);
657 static struct notifier_block time_cpufreq_notifier_block = {
658 .notifier_call = time_cpufreq_notifier
661 static int __init cpufreq_tsc(void)
663 INIT_WORK(&cpufreq_delayed_get_work, handle_cpufreq_delayed_get, NULL);
664 if (!cpufreq_register_notifier(&time_cpufreq_notifier_block,
665 CPUFREQ_TRANSITION_NOTIFIER))
670 core_initcall(cpufreq_tsc);
675 * calibrate_tsc() calibrates the processor TSC in a very simple way, comparing
676 * it to the HPET timer of known frequency.
679 #define TICK_COUNT 100000000
681 static unsigned int __init hpet_calibrate_tsc(void)
683 int tsc_start, hpet_start;
684 int tsc_now, hpet_now;
687 local_irq_save(flags);
690 hpet_start = hpet_readl(HPET_COUNTER);
695 hpet_now = hpet_readl(HPET_COUNTER);
698 local_irq_restore(flags);
699 } while ((tsc_now - tsc_start) < TICK_COUNT &&
700 (hpet_now - hpet_start) < TICK_COUNT);
702 return (tsc_now - tsc_start) * 1000000000L
703 / ((hpet_now - hpet_start) * hpet_period / 1000);
708 * pit_calibrate_tsc() uses the speaker output (channel 2) of
709 * the PIT. This is better than using the timer interrupt output,
710 * because we can read the value of the speaker with just one inb(),
711 * where we need three i/o operations for the interrupt channel.
712 * We count how many ticks the TSC does in 50 ms.
715 static unsigned int __init pit_calibrate_tsc(void)
717 unsigned long start, end;
720 spin_lock_irqsave(&i8253_lock, flags);
722 outb((inb(0x61) & ~0x02) | 0x01, 0x61);
725 outb((PIT_TICK_RATE / (1000 / 50)) & 0xff, 0x42);
726 outb((PIT_TICK_RATE / (1000 / 50)) >> 8, 0x42);
729 while ((inb(0x61) & 0x20) == 0);
733 spin_unlock_irqrestore(&i8253_lock, flags);
735 return (end - start) / 50;
739 static __init int late_hpet_init(void)
744 if (!vxtime.hpet_address)
747 memset(&hd, 0, sizeof (hd));
749 ntimer = hpet_readl(HPET_ID);
750 ntimer = (ntimer & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT;
754 * Register with driver.
755 * Timer0 and Timer1 is used by platform.
757 hd.hd_phys_address = vxtime.hpet_address;
758 hd.hd_address = (void *)fix_to_virt(FIX_HPET_BASE);
759 hd.hd_nirqs = ntimer;
760 hd.hd_flags = HPET_DATA_PLATFORM;
761 hpet_reserve_timer(&hd, 0);
762 #ifdef CONFIG_HPET_EMULATE_RTC
763 hpet_reserve_timer(&hd, 1);
765 hd.hd_irq[0] = HPET_LEGACY_8254;
766 hd.hd_irq[1] = HPET_LEGACY_RTC;
769 struct hpet_timer *timer;
772 hpet = (struct hpet *) fix_to_virt(FIX_HPET_BASE);
774 for (i = 2, timer = &hpet->hpet_timers[2]; i < ntimer;
776 hd.hd_irq[i] = (timer->hpet_config &
777 Tn_INT_ROUTE_CNF_MASK) >>
778 Tn_INT_ROUTE_CNF_SHIFT;
785 fs_initcall(late_hpet_init);
788 static int hpet_timer_stop_set_go(unsigned long tick)
793 * Stop the timers and reset the main counter.
796 cfg = hpet_readl(HPET_CFG);
797 cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY);
798 hpet_writel(cfg, HPET_CFG);
799 hpet_writel(0, HPET_COUNTER);
800 hpet_writel(0, HPET_COUNTER + 4);
803 * Set up timer 0, as periodic with first interrupt to happen at hpet_tick,
804 * and period also hpet_tick.
807 hpet_writel(HPET_TN_ENABLE | HPET_TN_PERIODIC | HPET_TN_SETVAL |
808 HPET_TN_32BIT, HPET_T0_CFG);
809 hpet_writel(hpet_tick, HPET_T0_CMP);
810 hpet_writel(hpet_tick, HPET_T0_CMP); /* AK: why twice? */
816 cfg |= HPET_CFG_ENABLE | HPET_CFG_LEGACY;
817 hpet_writel(cfg, HPET_CFG);
822 static int hpet_init(void)
826 if (!vxtime.hpet_address)
828 set_fixmap_nocache(FIX_HPET_BASE, vxtime.hpet_address);
829 __set_fixmap(VSYSCALL_HPET, vxtime.hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
832 * Read the period, compute tick and quotient.
835 id = hpet_readl(HPET_ID);
837 if (!(id & HPET_ID_VENDOR) || !(id & HPET_ID_NUMBER) ||
838 !(id & HPET_ID_LEGSUP))
841 hpet_period = hpet_readl(HPET_PERIOD);
842 if (hpet_period < 100000 || hpet_period > 100000000)
845 hpet_tick = (1000000000L * (USEC_PER_SEC / HZ) + hpet_period / 2) /
848 return hpet_timer_stop_set_go(hpet_tick);
851 static int hpet_reenable(void)
853 return hpet_timer_stop_set_go(hpet_tick);
856 void __init pit_init(void)
860 spin_lock_irqsave(&i8253_lock, flags);
861 outb_p(0x34, 0x43); /* binary, mode 2, LSB/MSB, ch 0 */
862 outb_p(LATCH & 0xff, 0x40); /* LSB */
863 outb_p(LATCH >> 8, 0x40); /* MSB */
864 spin_unlock_irqrestore(&i8253_lock, flags);
867 int __init time_setup(char *str)
869 report_lost_ticks = 1;
873 static struct irqaction irq0 = {
874 timer_interrupt, SA_INTERRUPT, CPU_MASK_NONE, "timer", NULL, NULL
877 extern void __init config_acpi_tables(void);
879 void __init time_init(void)
883 #ifdef HPET_HACK_ENABLE_DANGEROUS
884 if (!vxtime.hpet_address) {
885 printk(KERN_WARNING "time.c: WARNING: Enabling HPET base "
887 outl(0x800038a0, 0xcf8);
888 outl(0xff000001, 0xcfc);
889 outl(0x800038a0, 0xcf8);
890 vxtime.hpet_address = inl(0xcfc) & 0xfffffffe;
891 printk(KERN_WARNING "time.c: WARNING: Enabled HPET "
892 "at %#lx.\n", vxtime.hpet_address);
896 vxtime.hpet_address = 0;
898 xtime.tv_sec = get_cmos_time();
901 set_normalized_timespec(&wall_to_monotonic,
902 -xtime.tv_sec, -xtime.tv_nsec);
905 vxtime_hz = (1000000000000000L + hpet_period / 2) /
907 cpu_khz = hpet_calibrate_tsc();
909 #ifdef CONFIG_X86_PM_TIMER
910 } else if (pmtmr_ioport) {
911 vxtime_hz = PM_TIMER_FREQUENCY;
914 cpu_khz = pit_calibrate_tsc();
918 cpu_khz = pit_calibrate_tsc();
922 printk(KERN_INFO "time.c: Using %ld.%06ld MHz %s timer.\n",
923 vxtime_hz / 1000000, vxtime_hz % 1000000, timename);
924 printk(KERN_INFO "time.c: Detected %d.%03d MHz processor.\n",
925 cpu_khz / 1000, cpu_khz % 1000);
926 vxtime.mode = VXTIME_TSC;
927 vxtime.quot = (1000000L << 32) / vxtime_hz;
928 vxtime.tsc_quot = (1000L << 32) / cpu_khz;
929 vxtime.hz = vxtime_hz;
930 rdtscll_sync(&vxtime.last_tsc);
933 set_cyc2ns_scale(cpu_khz / 1000);
941 * Make an educated guess if the TSC is trustworthy and synchronized
944 static __init int unsynchronized_tsc(void)
947 if (oem_force_hpet_timer())
949 /* Intel systems are normally all synchronized. Exceptions
950 are handled in the OEM check above. */
951 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
953 /* All in a single socket - should be synchronized */
954 if (cpus_weight(cpu_core_map[0]) == num_online_cpus())
957 /* Assume multi socket systems are not synchronized */
958 return num_online_cpus() > 1;
962 * Decide after all CPUs are booted what mode gettimeofday should use.
964 void __init time_init_gtod(void)
968 if (unsynchronized_tsc())
970 if (vxtime.hpet_address && notsc) {
972 vxtime.last = hpet_readl(HPET_T0_CMP) - hpet_tick;
973 vxtime.mode = VXTIME_HPET;
974 do_gettimeoffset = do_gettimeoffset_hpet;
975 #ifdef CONFIG_X86_PM_TIMER
976 /* Using PM for gettimeofday is quite slow, but we have no other
977 choice because the TSC is too unreliable on some systems. */
978 } else if (pmtmr_ioport && !vxtime.hpet_address && notsc) {
980 do_gettimeoffset = do_gettimeoffset_pm;
981 vxtime.mode = VXTIME_PMTMR;
983 printk(KERN_INFO "Disabling vsyscall due to use of PM timer\n");
986 timetype = vxtime.hpet_address ? "HPET/TSC" : "PIT/TSC";
987 vxtime.mode = VXTIME_TSC;
990 printk(KERN_INFO "time.c: Using %s based timekeeping.\n", timetype);
993 __setup("report_lost_ticks", time_setup);
995 static long clock_cmos_diff;
996 static unsigned long sleep_start;
998 static int timer_suspend(struct sys_device *dev, pm_message_t state)
1001 * Estimate time zone so that set_time can update the clock
1003 long cmos_time = get_cmos_time();
1005 clock_cmos_diff = -cmos_time;
1006 clock_cmos_diff += get_seconds();
1007 sleep_start = cmos_time;
1011 static int timer_resume(struct sys_device *dev)
1013 unsigned long flags;
1015 unsigned long ctime = get_cmos_time();
1016 unsigned long sleep_length = (ctime - sleep_start) * HZ;
1018 if (vxtime.hpet_address)
1021 i8254_timer_resume();
1023 sec = ctime + clock_cmos_diff;
1024 write_seqlock_irqsave(&xtime_lock,flags);
1027 write_sequnlock_irqrestore(&xtime_lock,flags);
1028 jiffies += sleep_length;
1029 wall_jiffies += sleep_length;
1033 static struct sysdev_class timer_sysclass = {
1034 .resume = timer_resume,
1035 .suspend = timer_suspend,
1036 set_kset_name("timer"),
1040 /* XXX this driverfs stuff should probably go elsewhere later -john */
1041 static struct sys_device device_timer = {
1043 .cls = &timer_sysclass,
1046 static int time_init_device(void)
1048 int error = sysdev_class_register(&timer_sysclass);
1050 error = sysdev_register(&device_timer);
1054 device_initcall(time_init_device);
1056 #ifdef CONFIG_HPET_EMULATE_RTC
1057 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
1058 * is enabled, we support RTC interrupt functionality in software.
1059 * RTC has 3 kinds of interrupts:
1060 * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
1062 * 2) Alarm Interrupt - generate an interrupt at a specific time of day
1063 * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
1064 * 2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
1065 * (1) and (2) above are implemented using polling at a frequency of
1066 * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
1067 * overhead. (DEFAULT_RTC_INT_FREQ)
1068 * For (3), we use interrupts at 64Hz or user specified periodic
1069 * frequency, whichever is higher.
1071 #include <linux/rtc.h>
1073 extern irqreturn_t rtc_interrupt(int irq, void *dev_id, struct pt_regs *regs);
1075 #define DEFAULT_RTC_INT_FREQ 64
1076 #define RTC_NUM_INTS 1
1078 static unsigned long UIE_on;
1079 static unsigned long prev_update_sec;
1081 static unsigned long AIE_on;
1082 static struct rtc_time alarm_time;
1084 static unsigned long PIE_on;
1085 static unsigned long PIE_freq = DEFAULT_RTC_INT_FREQ;
1086 static unsigned long PIE_count;
1088 static unsigned long hpet_rtc_int_freq; /* RTC interrupt frequency */
1090 int is_hpet_enabled(void)
1092 return vxtime.hpet_address != 0;
1096 * Timer 1 for RTC, we do not use periodic interrupt feature,
1097 * even if HPET supports periodic interrupts on Timer 1.
1098 * The reason being, to set up a periodic interrupt in HPET, we need to
1099 * stop the main counter. And if we do that everytime someone diables/enables
1100 * RTC, we will have adverse effect on main kernel timer running on Timer 0.
1101 * So, for the time being, simulate the periodic interrupt in software.
1103 * hpet_rtc_timer_init() is called for the first time and during subsequent
1104 * interuppts reinit happens through hpet_rtc_timer_reinit().
1106 int hpet_rtc_timer_init(void)
1108 unsigned int cfg, cnt;
1109 unsigned long flags;
1111 if (!is_hpet_enabled())
1114 * Set the counter 1 and enable the interrupts.
1116 if (PIE_on && (PIE_freq > DEFAULT_RTC_INT_FREQ))
1117 hpet_rtc_int_freq = PIE_freq;
1119 hpet_rtc_int_freq = DEFAULT_RTC_INT_FREQ;
1121 local_irq_save(flags);
1122 cnt = hpet_readl(HPET_COUNTER);
1123 cnt += ((hpet_tick*HZ)/hpet_rtc_int_freq);
1124 hpet_writel(cnt, HPET_T1_CMP);
1125 local_irq_restore(flags);
1127 cfg = hpet_readl(HPET_T1_CFG);
1128 cfg |= HPET_TN_ENABLE | HPET_TN_SETVAL | HPET_TN_32BIT;
1129 hpet_writel(cfg, HPET_T1_CFG);
1134 static void hpet_rtc_timer_reinit(void)
1136 unsigned int cfg, cnt;
1138 if (!(PIE_on | AIE_on | UIE_on))
1141 if (PIE_on && (PIE_freq > DEFAULT_RTC_INT_FREQ))
1142 hpet_rtc_int_freq = PIE_freq;
1144 hpet_rtc_int_freq = DEFAULT_RTC_INT_FREQ;
1146 /* It is more accurate to use the comparator value than current count.*/
1147 cnt = hpet_readl(HPET_T1_CMP);
1148 cnt += hpet_tick*HZ/hpet_rtc_int_freq;
1149 hpet_writel(cnt, HPET_T1_CMP);
1151 cfg = hpet_readl(HPET_T1_CFG);
1152 cfg |= HPET_TN_ENABLE | HPET_TN_SETVAL | HPET_TN_32BIT;
1153 hpet_writel(cfg, HPET_T1_CFG);
1159 * The functions below are called from rtc driver.
1160 * Return 0 if HPET is not being used.
1161 * Otherwise do the necessary changes and return 1.
1163 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1165 if (!is_hpet_enabled())
1168 if (bit_mask & RTC_UIE)
1170 if (bit_mask & RTC_PIE)
1172 if (bit_mask & RTC_AIE)
1178 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1180 int timer_init_reqd = 0;
1182 if (!is_hpet_enabled())
1185 if (!(PIE_on | AIE_on | UIE_on))
1186 timer_init_reqd = 1;
1188 if (bit_mask & RTC_UIE) {
1191 if (bit_mask & RTC_PIE) {
1195 if (bit_mask & RTC_AIE) {
1199 if (timer_init_reqd)
1200 hpet_rtc_timer_init();
1205 int hpet_set_alarm_time(unsigned char hrs, unsigned char min, unsigned char sec)
1207 if (!is_hpet_enabled())
1210 alarm_time.tm_hour = hrs;
1211 alarm_time.tm_min = min;
1212 alarm_time.tm_sec = sec;
1217 int hpet_set_periodic_freq(unsigned long freq)
1219 if (!is_hpet_enabled())
1228 int hpet_rtc_dropped_irq(void)
1230 if (!is_hpet_enabled())
1236 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id, struct pt_regs *regs)
1238 struct rtc_time curr_time;
1239 unsigned long rtc_int_flag = 0;
1240 int call_rtc_interrupt = 0;
1242 hpet_rtc_timer_reinit();
1244 if (UIE_on | AIE_on) {
1245 rtc_get_rtc_time(&curr_time);
1248 if (curr_time.tm_sec != prev_update_sec) {
1249 /* Set update int info, call real rtc int routine */
1250 call_rtc_interrupt = 1;
1251 rtc_int_flag = RTC_UF;
1252 prev_update_sec = curr_time.tm_sec;
1257 if (PIE_count >= hpet_rtc_int_freq/PIE_freq) {
1258 /* Set periodic int info, call real rtc int routine */
1259 call_rtc_interrupt = 1;
1260 rtc_int_flag |= RTC_PF;
1265 if ((curr_time.tm_sec == alarm_time.tm_sec) &&
1266 (curr_time.tm_min == alarm_time.tm_min) &&
1267 (curr_time.tm_hour == alarm_time.tm_hour)) {
1268 /* Set alarm int info, call real rtc int routine */
1269 call_rtc_interrupt = 1;
1270 rtc_int_flag |= RTC_AF;
1273 if (call_rtc_interrupt) {
1274 rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1275 rtc_interrupt(rtc_int_flag, dev_id, regs);
1283 static int __init nohpet_setup(char *s)
1289 __setup("nohpet", nohpet_setup);
1292 static int __init notsc_setup(char *s)
1298 __setup("notsc", notsc_setup);