2 * Copyright (c) 1990 The Regents of the University of California.
5 * This code is derived from software contributed to Berkeley by
6 * William Jolitz and Don Ahn.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * from: @(#)clock.c 7.2 (Berkeley) 5/12/91
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 /* #define DELAYDEBUG */
44 * Routines to handle clock hardware.
48 #include "opt_clock.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
53 #include <sys/clock.h>
55 #include <sys/mutex.h>
58 #include <sys/timetc.h>
59 #include <sys/kernel.h>
60 #include <sys/limits.h>
61 #include <sys/sysctl.h>
63 #include <sys/power.h>
65 #include <machine/clock.h>
66 #include <machine/cputypes.h>
67 #include <machine/frame.h>
68 #include <machine/intr_machdep.h>
69 #include <machine/md_var.h>
70 #include <machine/psl.h>
72 #include <machine/smp.h>
74 #include <machine/specialreg.h>
75 #include <machine/timerreg.h>
77 #include <i386/isa/icu.h>
78 #include <i386/isa/isa.h>
81 #include <xen/xen_intr.h>
84 #include <machine/pmap.h>
85 #include <xen/hypervisor.h>
86 #include <machine/xen/xen-os.h>
87 #include <machine/xen/xenfunc.h>
88 #include <xen/interface/vcpu.h>
89 #include <machine/cpu.h>
90 #include <machine/xen/xen_clock_util.h>
93 * 32-bit time_t's can't reach leap years before 1904 or after 2036, so we
94 * can use a simple formula for leap years.
96 #define LEAPYEAR(y) (!((y) % 4))
97 #define DAYSPERYEAR (28+30*4+31*7)
100 #define TIMER_FREQ 1193182
103 #ifdef CYC2NS_SCALE_FACTOR
104 #undef CYC2NS_SCALE_FACTOR
106 #define CYC2NS_SCALE_FACTOR 10
108 /* Values for timerX_state: */
110 #define RELEASE_PENDING 1
112 #define ACQUIRE_PENDING 3
114 struct mtx clock_lock;
115 #define RTC_LOCK_INIT \
116 mtx_init(&clock_lock, "clk", NULL, MTX_SPIN | MTX_NOPROFILE)
117 #define RTC_LOCK mtx_lock_spin(&clock_lock)
118 #define RTC_UNLOCK mtx_unlock_spin(&clock_lock)
120 int adjkerntz; /* local offset from GMT in seconds */
125 u_int timer_freq = TIMER_FREQ;
126 static int independent_wallclock;
127 static int xen_disable_rtc_set;
128 static u_long cyc2ns_scale;
129 static struct timespec shadow_tv;
130 static uint32_t shadow_tv_version; /* XXX: lazy locking */
131 static uint64_t processed_system_time; /* stime (ns) at last processing. */
133 static const u_char daysinmonth[] = {31,28,31,30,31,30,31,31,30,31,30,31};
135 SYSCTL_INT(_machdep, OID_AUTO, independent_wallclock,
136 CTLFLAG_RW, &independent_wallclock, 0, "");
137 SYSCTL_INT(_machdep, OID_AUTO, xen_disable_rtc_set,
138 CTLFLAG_RW, &xen_disable_rtc_set, 1, "");
141 #define do_div(n,base) ({ \
142 unsigned long __upper, __low, __high, __mod, __base; \
144 __asm("":"=a" (__low), "=d" (__high):"A" (n)); \
147 __upper = __high % (__base); \
148 __high = __high / (__base); \
150 __asm("divl %2":"=a" (__low), "=d" (__mod):"rm" (__base), "0" (__low), "1" (__upper)); \
151 __asm("":"=A" (n):"a" (__low),"d" (__high)); \
156 #define NS_PER_TICK (1000000000ULL/hz)
158 #define rdtscll(val) \
159 __asm__ __volatile__("rdtsc" : "=A" (val))
162 /* convert from cycles(64bits) => nanoseconds (64bits)
164 * ns = cycles / (freq / ns_per_sec)
165 * ns = cycles * (ns_per_sec / freq)
166 * ns = cycles * (10^9 / (cpu_mhz * 10^6))
167 * ns = cycles * (10^3 / cpu_mhz)
169 * Then we use scaling math (suggested by george@mvista.com) to get:
170 * ns = cycles * (10^3 * SC / cpu_mhz) / SC
171 * ns = cycles * cyc2ns_scale / SC
173 * And since SC is a constant power of two, we can convert the div
175 * -johnstul@us.ibm.com "math is hard, lets go shopping!"
177 static inline void set_cyc2ns_scale(unsigned long cpu_mhz)
179 cyc2ns_scale = (1000 << CYC2NS_SCALE_FACTOR)/cpu_mhz;
182 static inline unsigned long long cycles_2_ns(unsigned long long cyc)
184 return (cyc * cyc2ns_scale) >> CYC2NS_SCALE_FACTOR;
188 * Scale a 64-bit delta by scaling and multiplying by a 32-bit fraction,
189 * yielding a 64-bit result.
191 static inline uint64_t
192 scale_delta(uint64_t delta, uint32_t mul_frac, int shift)
210 : "=A" (product), "=r" (tmp1), "=r" (tmp2)
211 : "a" ((uint32_t)delta), "1" ((uint32_t)(delta >> 32)), "2" (mul_frac) );
217 get_nsec_offset(struct shadow_time_info *shadow)
221 delta = now - shadow->tsc_timestamp;
222 return scale_delta(delta, shadow->tsc_to_nsec_mul, shadow->tsc_shift);
225 static void update_wallclock(void)
227 shared_info_t *s = HYPERVISOR_shared_info;
230 shadow_tv_version = s->wc_version;
232 shadow_tv.tv_sec = s->wc_sec;
233 shadow_tv.tv_nsec = s->wc_nsec;
236 while ((s->wc_version & 1) | (shadow_tv_version ^ s->wc_version));
241 add_uptime_to_wallclock(void)
245 xen_fetch_uptime(&ut);
246 timespecadd(&shadow_tv, &ut);
250 * Reads a consistent set of time-base values from Xen, into a shadow data
251 * area. Must be called with the xtime_lock held for writing.
253 static void __get_time_values_from_xen(void)
255 shared_info_t *s = HYPERVISOR_shared_info;
256 struct vcpu_time_info *src;
257 struct shadow_time_info *dst;
258 uint32_t pre_version, post_version;
260 src = &s->vcpu_info[smp_processor_id()].time;
261 dst = &per_cpu(shadow_time, smp_processor_id());
265 pre_version = dst->version = src->version;
267 dst->tsc_timestamp = src->tsc_timestamp;
268 dst->system_timestamp = src->system_time;
269 dst->tsc_to_nsec_mul = src->tsc_to_system_mul;
270 dst->tsc_shift = src->tsc_shift;
272 post_version = src->version;
274 while ((pre_version & 1) | (pre_version ^ post_version));
276 dst->tsc_to_usec_mul = dst->tsc_to_nsec_mul / 1000;
281 static inline int time_values_up_to_date(int cpu)
283 struct vcpu_time_info *src;
284 struct shadow_time_info *dst;
286 src = &HYPERVISOR_shared_info->vcpu_info[cpu].time;
287 dst = &per_cpu(shadow_time, cpu);
290 return (dst->version == src->version);
293 static unsigned xen_get_timecount(struct timecounter *tc);
295 static struct timecounter xen_timecounter = {
296 xen_get_timecount, /* get_timecount */
298 ~0u, /* counter_mask */
307 int64_t delta_cpu, delta;
308 struct trapframe *frame = (struct trapframe *)arg;
309 int cpu = smp_processor_id();
310 struct shadow_time_info *shadow = &per_cpu(shadow_time, cpu);
313 __get_time_values_from_xen();
316 shadow->system_timestamp + get_nsec_offset(shadow);
317 delta -= processed_system_time;
318 delta_cpu -= per_cpu(processed_system_time, cpu);
320 } while (!time_values_up_to_date(cpu));
322 if (unlikely(delta < (int64_t)0) || unlikely(delta_cpu < (int64_t)0)) {
323 printf("Timer ISR: Time went backwards: %lld\n", delta);
324 return (FILTER_HANDLED);
327 /* Process elapsed ticks since last call. */
328 while (delta >= NS_PER_TICK) {
329 delta -= NS_PER_TICK;
330 processed_system_time += NS_PER_TICK;
331 per_cpu(processed_system_time, cpu) += NS_PER_TICK;
332 if (PCPU_GET(cpuid) == 0)
333 hardclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
335 hardclock_cpu(TRAPF_USERMODE(frame));
338 * Take synchronised time from Xen once a minute if we're not
339 * synchronised ourselves, and we haven't chosen to keep an independent
343 if (shadow_tv_version != HYPERVISOR_shared_info->wc_version &&
344 !independent_wallclock) {
345 printf("[XEN] hypervisor wallclock nudged; nudging TOD.\n");
347 add_uptime_to_wallclock();
348 tc_setclock(&shadow_tv);
352 return (FILTER_HANDLED);
357 struct shadow_time_info *shadow;
359 uint32_t local_time_version;
361 shadow = &per_cpu(shadow_time, smp_processor_id());
364 local_time_version = shadow->version;
366 time = shadow->system_timestamp + get_nsec_offset(shadow);
367 if (!time_values_up_to_date(smp_processor_id()))
368 __get_time_values_from_xen(/*cpu */);
370 } while (local_time_version != shadow->version);
377 * XXX: timer needs more SMP work.
387 * Wait "n" microseconds.
388 * Relies on timer 1 counting down from (timer_freq / hz)
389 * Note: timer had better have been programmed before this is first used!
394 int delta, ticks_left;
395 uint32_t tick, prev_tick;
399 static int state = 0;
403 for (n1 = 1; n1 <= 10000000; n1 *= 10)
408 printf("DELAY(%d)...", n);
411 * Read the counter first, so that the rest of the setup overhead is
412 * counted. Guess the initial overhead is 20 usec (on most systems it
413 * takes about 1.5 usec for each of the i/o's in getit(). The loop
414 * takes about 6 usec on a 486/33 and 13 usec on a 386/20. The
415 * multiplications and divisions to scale the count take a while).
417 * However, if ddb is active then use a fake counter since reading
418 * the i8254 counter involves acquiring a lock. ddb must not go
419 * locking for many reasons, but it calls here for at least atkbd
424 n -= 0; /* XXX actually guess no initial overhead */
426 * Calculate (n * (timer_freq / 1e6)) without using floating point
427 * and without any avoidable overflows.
433 * Use fixed point to avoid a slow division by 1000000.
434 * 39099 = 1193182 * 2^15 / 10^6 rounded to nearest.
435 * 2^15 is the first power of 2 that gives exact results
436 * for n between 0 and 256.
438 ticks_left = ((u_int)n * 39099 + (1 << 15) - 1) >> 15;
441 * Don't bother using fixed point, although gcc-2.7.2
442 * generates particularly poor code for the long long
443 * division, since even the slow way will complete long
444 * before the delay is up (unless we're interrupted).
446 ticks_left = ((u_int)n * (long long)timer_freq + 999999)
449 while (ticks_left > 0) {
454 delta = tick - prev_tick;
458 * Guard against timer0_max_count being wrong.
459 * This shouldn't happen in normal operation,
460 * but it may happen if set_timer_freq() is
463 /* delta += timer0_max_count; ??? */
471 printf(" %d calls to getit() at %d usec each\n",
472 getit_calls, (n + 5) / getit_calls);
478 * Restore all the timers non-atomically (XXX: should be atomically).
480 * This function is called from pmtimer_resume() to restore all the timers.
481 * This should not be necessary, but there are broken laptops that do not
482 * restore all the timers on resume.
487 /* Get timebases for new environment. */
488 __get_time_values_from_xen();
490 /* Reset our own concept of passage of system time. */
491 processed_system_time = per_cpu(shadow_time, 0).system_timestamp;
492 per_cpu(processed_system_time, 0) = processed_system_time;
498 unsigned long long alarm;
501 struct vcpu_time_info *info;
503 /* initialize xen values */
504 __get_time_values_from_xen();
505 processed_system_time = per_cpu(shadow_time, 0).system_timestamp;
506 per_cpu(processed_system_time, 0) = processed_system_time;
508 __cpu_khz = 1000000ULL << 32;
509 info = &HYPERVISOR_shared_info->vcpu_info[0].time;
511 do_div(__cpu_khz, info->tsc_to_system_mul);
512 if ( info->tsc_shift < 0 )
513 cpu_khz = __cpu_khz << -info->tsc_shift;
515 cpu_khz = __cpu_khz >> info->tsc_shift;
517 printf("Xen reported: %u.%03u MHz processor.\n",
518 cpu_khz / 1000, cpu_khz % 1000);
520 /* (10^6 * 2^32) / cpu_hz = (10^3 * 2^32) / cpu_khz =
521 (2^32 * 1 / (clocks/us)) */
523 set_cyc2ns_scale(cpu_khz/1000);
524 tsc_freq = cpu_khz * 1000;
526 timer_freq = 1000000000LL;
527 xen_timecounter.tc_frequency = timer_freq >> 9;
528 tc_init(&xen_timecounter);
534 * RTC support routines
541 return(bcd2bin(rtcin(port)));
545 #ifdef XEN_PRIVILEGED_GUEST
548 * Initialize the time of day register, based on the time base which is, e.g.
552 domu_inittodr(time_t base)
559 add_uptime_to_wallclock();
569 sec += tz_minuteswest * 60 + (wall_cmos_clock ? adjkerntz : 0);
571 y = time_second - shadow_tv.tv_sec;
572 if (y <= -2 || y >= 2) {
573 /* badly off, adjust it */
574 tc_setclock(&shadow_tv);
580 * Write system time back to RTC.
588 struct shadow_time_info *shadow;
590 shadow = &per_cpu(shadow_time, smp_processor_id());
591 if (xen_disable_rtc_set)
598 tm -= tz_minuteswest * 60 + (wall_cmos_clock ? adjkerntz : 0);
600 if ((xen_start_info->flags & SIF_INITDOMAIN) &&
601 !independent_wallclock)
603 op.cmd = DOM0_SETTIME;
604 op.u.settime.secs = tm;
605 op.u.settime.nsecs = 0;
606 op.u.settime.system_time = shadow->system_timestamp;
607 HYPERVISOR_dom0_op(&op);
609 add_uptime_to_wallclock();
610 } else if (independent_wallclock) {
617 * Initialize the time of day register, based on the time base which is, e.g.
621 inittodr(time_t base)
623 unsigned long sec, days;
628 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
641 /* Look if we have a RTC present and the time is valid */
642 if (!(rtcin(RTC_STATUSD) & RTCSD_PWR))
645 /* wait for time update to complete */
646 /* If RTCSA_TUP is zero, we have at least 244us before next update */
648 while (rtcin(RTC_STATUSA) & RTCSA_TUP) {
654 #ifdef USE_RTC_CENTURY
655 year = readrtc(RTC_YEAR) + readrtc(RTC_CENTURY) * 100;
657 year = readrtc(RTC_YEAR) + 1900;
665 month = readrtc(RTC_MONTH);
666 for (m = 1; m < month; m++)
667 days += daysinmonth[m-1];
668 if ((month > 2) && LEAPYEAR(year))
670 days += readrtc(RTC_DAY) - 1;
671 for (y = 1970; y < year; y++)
672 days += DAYSPERYEAR + LEAPYEAR(y);
673 sec = ((( days * 24 +
674 readrtc(RTC_HRS)) * 60 +
675 readrtc(RTC_MIN)) * 60 +
677 /* sec now contains the number of seconds, since Jan 1 1970,
678 in the local time zone */
680 sec += tz_minuteswest * 60 + (wall_cmos_clock ? adjkerntz : 0);
682 y = time_second - sec;
683 if (y <= -2 || y >= 2) {
684 /* badly off, adjust it */
693 printf("Invalid time in real time clock.\n");
694 printf("Check and reset the date immediately!\n");
699 * Write system time back to RTC
707 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
712 if (xen_disable_rtc_set)
719 /* Disable RTC updates and interrupts. */
720 writertc(RTC_STATUSB, RTCSB_HALT | RTCSB_24HR);
722 /* Calculate local time to put in RTC */
724 tm -= tz_minuteswest * 60 + (wall_cmos_clock ? adjkerntz : 0);
726 writertc(RTC_SEC, bin2bcd(tm%60)); tm /= 60; /* Write back Seconds */
727 writertc(RTC_MIN, bin2bcd(tm%60)); tm /= 60; /* Write back Minutes */
728 writertc(RTC_HRS, bin2bcd(tm%24)); tm /= 24; /* Write back Hours */
730 /* We have now the days since 01-01-1970 in tm */
731 writertc(RTC_WDAY, (tm + 4) % 7 + 1); /* Write back Weekday */
732 for (y = 1970, m = DAYSPERYEAR + LEAPYEAR(y);
734 y++, m = DAYSPERYEAR + LEAPYEAR(y))
737 /* Now we have the years in y and the day-of-the-year in tm */
738 writertc(RTC_YEAR, bin2bcd(y%100)); /* Write back Year */
739 #ifdef USE_RTC_CENTURY
740 writertc(RTC_CENTURY, bin2bcd(y/100)); /* ... and Century */
746 if (m == 1 && LEAPYEAR(y))
753 writertc(RTC_MONTH, bin2bcd(m + 1)); /* Write back Month */
754 writertc(RTC_DAY, bin2bcd(tm + 1)); /* Write back Month Day */
756 /* Reenable RTC updates and interrupts. */
757 writertc(RTC_STATUSB, RTCSB_24HR);
762 static struct vcpu_set_periodic_timer xen_set_periodic_tick;
765 * Start clocks running.
770 unsigned int time_irq;
773 xen_set_periodic_tick.period_ns = NS_PER_TICK;
775 HYPERVISOR_vcpu_op(VCPUOP_set_periodic_timer, 0,
776 &xen_set_periodic_tick);
778 error = bind_virq_to_irqhandler(VIRQ_TIMER, 0, "clk",
780 INTR_TYPE_CLK | INTR_FAST, &time_irq);
782 panic("failed to register clock interrupt\n");
783 /* should fast clock be enabled ? */
788 ap_cpu_initclocks(int cpu)
790 unsigned int time_irq;
793 xen_set_periodic_tick.period_ns = NS_PER_TICK;
795 HYPERVISOR_vcpu_op(VCPUOP_set_periodic_timer, cpu,
796 &xen_set_periodic_tick);
797 error = bind_virq_to_irqhandler(VIRQ_TIMER, 0, "clk",
799 INTR_TYPE_CLK | INTR_FAST, &time_irq);
801 panic("failed to register clock interrupt\n");
809 cpu_startprofclock(void)
812 printf("cpu_startprofclock: profiling clock is not supported\n");
816 cpu_stopprofclock(void)
819 printf("cpu_stopprofclock: profiling clock is not supported\n");
821 #define NSEC_PER_USEC 1000
824 xen_get_timecount(struct timecounter *tc)
827 struct shadow_time_info *shadow;
828 shadow = &per_cpu(shadow_time, smp_processor_id());
830 __get_time_values_from_xen();
832 clk = shadow->system_timestamp + get_nsec_offset(shadow);
834 return (uint32_t)(clk >> 9);
838 /* Return system time offset by ticks */
840 get_system_time(int ticks)
842 return processed_system_time + (ticks * NS_PER_TICK);
846 * Track behavior of cur_timer->get_offset() functionality in timer_tsc.c
850 /* Convert jiffies to system time. */
852 ticks_to_system_time(int newticks)
857 delta = newticks - ticks;
859 /* Triggers in some wrap-around cases,
861 * we just end up with a shorter timeout. */
862 st = processed_system_time + NS_PER_TICK;
863 } else if (((unsigned int)delta >> (BITS_PER_LONG-3)) != 0) {
864 /* Very long timeout means there is no pending timer.
865 * We indicate this to Xen by passing zero timeout. */
868 st = processed_system_time + delta * (uint64_t)NS_PER_TICK;
879 timeout = ticks_to_system_time(ticks + 1) + NS_PER_TICK/2;
881 __get_time_values_from_xen();
882 PANIC_IF(HYPERVISOR_set_timer_op(timeout) != 0);
883 HYPERVISOR_sched_op(SCHEDOP_block, 0);
887 timer_spkr_acquire(void)
894 timer_spkr_release(void)
901 timer_spkr_setfreq(int freq)