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 <machine/xen/xen_intr.h>
84 #include <machine/pmap.h>
85 #include <machine/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>
92 * 32-bit time_t's can't reach leap years before 1904 or after 2036, so we
93 * can use a simple formula for leap years.
95 #define LEAPYEAR(y) (!((y) % 4))
96 #define DAYSPERYEAR (28+30*4+31*7)
99 #define TIMER_FREQ 1193182
102 #ifdef CYC2NS_SCALE_FACTOR
103 #undef CYC2NS_SCALE_FACTOR
105 #define CYC2NS_SCALE_FACTOR 10
107 /* Values for timerX_state: */
109 #define RELEASE_PENDING 1
111 #define ACQUIRE_PENDING 3
113 struct mtx clock_lock;
114 #define RTC_LOCK_INIT \
115 mtx_init(&clock_lock, "clk", NULL, MTX_SPIN | MTX_NOPROFILE)
116 #define RTC_LOCK mtx_lock_spin(&clock_lock)
117 #define RTC_UNLOCK mtx_unlock_spin(&clock_lock)
119 int adjkerntz; /* local offset from GMT in seconds */
123 int statclock_disable;
125 u_int timer_freq = TIMER_FREQ;
126 static int independent_wallclock;
127 static int xen_disable_rtc_set;
128 static u_long cached_gtm; /* cached quotient for TSC -> microseconds */
129 static u_long cyc2ns_scale;
130 static struct timespec shadow_tv;
131 static uint32_t shadow_tv_version; /* XXX: lazy locking */
132 static uint64_t processed_system_time; /* stime (ns) at last processing. */
134 static const u_char daysinmonth[] = {31,28,31,30,31,30,31,31,30,31,30,31};
136 SYSCTL_INT(_machdep, OID_AUTO, independent_wallclock,
137 CTLFLAG_RW, &independent_wallclock, 0, "");
138 SYSCTL_INT(_machdep, OID_AUTO, xen_disable_rtc_set,
139 CTLFLAG_RW, &xen_disable_rtc_set, 1, "");
142 #define do_div(n,base) ({ \
143 unsigned long __upper, __low, __high, __mod, __base; \
145 __asm("":"=a" (__low), "=d" (__high):"A" (n)); \
148 __upper = __high % (__base); \
149 __high = __high / (__base); \
151 __asm("divl %2":"=a" (__low), "=d" (__mod):"rm" (__base), "0" (__low), "1" (__upper)); \
152 __asm("":"=A" (n):"a" (__low),"d" (__high)); \
157 /* These are peridically updated in shared_info, and then copied here. */
158 struct shadow_time_info {
159 uint64_t tsc_timestamp; /* TSC at last update of time vals. */
160 uint64_t system_timestamp; /* Time, in nanosecs, since boot. */
161 uint32_t tsc_to_nsec_mul;
162 uint32_t tsc_to_usec_mul;
166 static DEFINE_PER_CPU(uint64_t, processed_system_time);
167 static DEFINE_PER_CPU(struct shadow_time_info, shadow_time);
170 #define NS_PER_TICK (1000000000ULL/hz)
172 #define rdtscll(val) \
173 __asm__ __volatile__("rdtsc" : "=A" (val))
176 /* convert from cycles(64bits) => nanoseconds (64bits)
178 * ns = cycles / (freq / ns_per_sec)
179 * ns = cycles * (ns_per_sec / freq)
180 * ns = cycles * (10^9 / (cpu_mhz * 10^6))
181 * ns = cycles * (10^3 / cpu_mhz)
183 * Then we use scaling math (suggested by george@mvista.com) to get:
184 * ns = cycles * (10^3 * SC / cpu_mhz) / SC
185 * ns = cycles * cyc2ns_scale / SC
187 * And since SC is a constant power of two, we can convert the div
189 * -johnstul@us.ibm.com "math is hard, lets go shopping!"
191 static inline void set_cyc2ns_scale(unsigned long cpu_mhz)
193 cyc2ns_scale = (1000 << CYC2NS_SCALE_FACTOR)/cpu_mhz;
196 static inline unsigned long long cycles_2_ns(unsigned long long cyc)
198 return (cyc * cyc2ns_scale) >> CYC2NS_SCALE_FACTOR;
202 * Scale a 64-bit delta by scaling and multiplying by a 32-bit fraction,
203 * yielding a 64-bit result.
205 static inline uint64_t
206 scale_delta(uint64_t delta, uint32_t mul_frac, int shift)
224 : "=A" (product), "=r" (tmp1), "=r" (tmp2)
225 : "a" ((uint32_t)delta), "1" ((uint32_t)(delta >> 32)), "2" (mul_frac) );
230 static uint64_t get_nsec_offset(struct shadow_time_info *shadow)
234 delta = now - shadow->tsc_timestamp;
235 return scale_delta(delta, shadow->tsc_to_nsec_mul, shadow->tsc_shift);
238 static void update_wallclock(void)
240 shared_info_t *s = HYPERVISOR_shared_info;
243 shadow_tv_version = s->wc_version;
245 shadow_tv.tv_sec = s->wc_sec;
246 shadow_tv.tv_nsec = s->wc_nsec;
249 while ((s->wc_version & 1) | (shadow_tv_version ^ s->wc_version));
254 * Reads a consistent set of time-base values from Xen, into a shadow data
255 * area. Must be called with the xtime_lock held for writing.
257 static void __get_time_values_from_xen(void)
259 shared_info_t *s = HYPERVISOR_shared_info;
260 struct vcpu_time_info *src;
261 struct shadow_time_info *dst;
263 src = &s->vcpu_info[smp_processor_id()].time;
264 dst = &per_cpu(shadow_time, smp_processor_id());
267 dst->version = src->version;
269 dst->tsc_timestamp = src->tsc_timestamp;
270 dst->system_timestamp = src->system_time;
271 dst->tsc_to_nsec_mul = src->tsc_to_system_mul;
272 dst->tsc_shift = src->tsc_shift;
275 while ((src->version & 1) | (dst->version ^ src->version));
277 dst->tsc_to_usec_mul = dst->tsc_to_nsec_mul / 1000;
280 static inline int time_values_up_to_date(int cpu)
282 struct vcpu_time_info *src;
283 struct shadow_time_info *dst;
285 src = &HYPERVISOR_shared_info->vcpu_info[cpu].time;
286 dst = &per_cpu(shadow_time, cpu);
289 return (dst->version == src->version);
292 static unsigned xen_get_timecount(struct timecounter *tc);
294 static struct timecounter xen_timecounter = {
295 xen_get_timecount, /* get_timecount */
297 ~0u, /* counter_mask */
304 clkintr(struct trapframe *frame)
306 int64_t delta_cpu, delta;
307 int cpu = smp_processor_id();
308 struct shadow_time_info *shadow = &per_cpu(shadow_time, cpu);
311 __get_time_values_from_xen();
314 shadow->system_timestamp + get_nsec_offset(shadow);
315 delta -= processed_system_time;
316 delta_cpu -= per_cpu(processed_system_time, cpu);
318 } while (!time_values_up_to_date(cpu));
320 if (unlikely(delta < (int64_t)0) || unlikely(delta_cpu < (int64_t)0)) {
321 printf("Timer ISR: Time went backwards: %lld\n", delta);
325 /* Process elapsed ticks since last call. */
326 if (delta >= NS_PER_TICK) {
327 processed_system_time += (delta / NS_PER_TICK) * NS_PER_TICK;
328 per_cpu(processed_system_time, cpu) += (delta_cpu / NS_PER_TICK) * NS_PER_TICK;
330 hardclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
333 * Take synchronised time from Xen once a minute if we're not
334 * synchronised ourselves, and we haven't chosen to keep an independent
338 if (shadow_tv_version != HYPERVISOR_shared_info->wc_version) {
340 tc_setclock(&shadow_tv);
349 struct shadow_time_info *shadow;
350 shadow = &per_cpu(shadow_time, smp_processor_id());
351 __get_time_values_from_xen();
352 return shadow->system_timestamp + get_nsec_offset(shadow);
357 * XXX: timer needs more SMP work.
367 * Wait "n" microseconds.
368 * Relies on timer 1 counting down from (timer_freq / hz)
369 * Note: timer had better have been programmed before this is first used!
374 int delta, ticks_left;
375 uint32_t tick, prev_tick;
379 static int state = 0;
383 for (n1 = 1; n1 <= 10000000; n1 *= 10)
388 printf("DELAY(%d)...", n);
391 * Read the counter first, so that the rest of the setup overhead is
392 * counted. Guess the initial overhead is 20 usec (on most systems it
393 * takes about 1.5 usec for each of the i/o's in getit(). The loop
394 * takes about 6 usec on a 486/33 and 13 usec on a 386/20. The
395 * multiplications and divisions to scale the count take a while).
397 * However, if ddb is active then use a fake counter since reading
398 * the i8254 counter involves acquiring a lock. ddb must not go
399 * locking for many reasons, but it calls here for at least atkbd
404 n -= 0; /* XXX actually guess no initial overhead */
406 * Calculate (n * (timer_freq / 1e6)) without using floating point
407 * and without any avoidable overflows.
413 * Use fixed point to avoid a slow division by 1000000.
414 * 39099 = 1193182 * 2^15 / 10^6 rounded to nearest.
415 * 2^15 is the first power of 2 that gives exact results
416 * for n between 0 and 256.
418 ticks_left = ((u_int)n * 39099 + (1 << 15) - 1) >> 15;
421 * Don't bother using fixed point, although gcc-2.7.2
422 * generates particularly poor code for the long long
423 * division, since even the slow way will complete long
424 * before the delay is up (unless we're interrupted).
426 ticks_left = ((u_int)n * (long long)timer_freq + 999999)
429 while (ticks_left > 0) {
434 delta = tick - prev_tick;
438 * Guard against timer0_max_count being wrong.
439 * This shouldn't happen in normal operation,
440 * but it may happen if set_timer_freq() is
443 /* delta += timer0_max_count; ??? */
451 printf(" %d calls to getit() at %d usec each\n",
452 getit_calls, (n + 5) / getit_calls);
458 * Restore all the timers non-atomically (XXX: should be atomically).
460 * This function is called from pmtimer_resume() to restore all the timers.
461 * This should not be necessary, but there are broken laptops that do not
462 * restore all the timers on resume.
467 /* Get timebases for new environment. */
468 __get_time_values_from_xen();
470 /* Reset our own concept of passage of system time. */
471 processed_system_time = per_cpu(shadow_time, 0).system_timestamp;
472 per_cpu(processed_system_time, 0) = processed_system_time;
478 unsigned long long alarm;
481 struct vcpu_time_info *info;
483 /* initialize xen values */
484 __get_time_values_from_xen();
485 processed_system_time = per_cpu(shadow_time, 0).system_timestamp;
486 per_cpu(processed_system_time, 0) = processed_system_time;
488 __cpu_khz = 1000000ULL << 32;
489 info = &HYPERVISOR_shared_info->vcpu_info[0].time;
491 do_div(__cpu_khz, info->tsc_to_system_mul);
492 if ( info->tsc_shift < 0 )
493 cpu_khz = __cpu_khz << -info->tsc_shift;
495 cpu_khz = __cpu_khz >> info->tsc_shift;
497 printf("Xen reported: %u.%03u MHz processor.\n",
498 cpu_khz / 1000, cpu_khz % 1000);
500 /* (10^6 * 2^32) / cpu_hz = (10^3 * 2^32) / cpu_khz =
501 (2^32 * 1 / (clocks/us)) */
503 unsigned long eax=0, edx=1000;
505 :"=a" (cached_gtm), "=d" (edx)
507 "0" (eax), "1" (edx));
510 set_cyc2ns_scale(cpu_khz/1000);
511 tsc_freq = cpu_khz * 1000;
513 timer_freq = xen_timecounter.tc_frequency = 1000000000LL;
514 tc_init(&xen_timecounter);
521 * RTC support routines
528 return(bcd2bin(rtcin(port)));
532 #ifdef XEN_PRIVILEGED_GUEST
535 * Initialize the time of day register, based on the time base which is, e.g.
539 domu_inittodr(time_t base)
555 sec += tz_minuteswest * 60 + (wall_cmos_clock ? adjkerntz : 0);
557 y = time_second - shadow_tv.tv_sec;
558 if (y <= -2 || y >= 2) {
559 /* badly off, adjust it */
560 tc_setclock(&shadow_tv);
566 * Write system time back to RTC.
574 struct shadow_time_info *shadow;
576 shadow = &per_cpu(shadow_time, smp_processor_id());
577 if (xen_disable_rtc_set)
584 tm -= tz_minuteswest * 60 + (wall_cmos_clock ? adjkerntz : 0);
586 if ((xen_start_info->flags & SIF_INITDOMAIN) &&
587 !independent_wallclock)
589 op.cmd = DOM0_SETTIME;
590 op.u.settime.secs = tm;
591 op.u.settime.nsecs = 0;
592 op.u.settime.system_time = shadow->system_timestamp;
593 HYPERVISOR_dom0_op(&op);
595 } else if (independent_wallclock) {
602 * Initialize the time of day register, based on the time base which is, e.g.
606 inittodr(time_t base)
608 unsigned long sec, days;
613 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
626 /* Look if we have a RTC present and the time is valid */
627 if (!(rtcin(RTC_STATUSD) & RTCSD_PWR))
630 /* wait for time update to complete */
631 /* If RTCSA_TUP is zero, we have at least 244us before next update */
633 while (rtcin(RTC_STATUSA) & RTCSA_TUP) {
639 #ifdef USE_RTC_CENTURY
640 year = readrtc(RTC_YEAR) + readrtc(RTC_CENTURY) * 100;
642 year = readrtc(RTC_YEAR) + 1900;
650 month = readrtc(RTC_MONTH);
651 for (m = 1; m < month; m++)
652 days += daysinmonth[m-1];
653 if ((month > 2) && LEAPYEAR(year))
655 days += readrtc(RTC_DAY) - 1;
656 for (y = 1970; y < year; y++)
657 days += DAYSPERYEAR + LEAPYEAR(y);
658 sec = ((( days * 24 +
659 readrtc(RTC_HRS)) * 60 +
660 readrtc(RTC_MIN)) * 60 +
662 /* sec now contains the number of seconds, since Jan 1 1970,
663 in the local time zone */
665 sec += tz_minuteswest * 60 + (wall_cmos_clock ? adjkerntz : 0);
667 y = time_second - sec;
668 if (y <= -2 || y >= 2) {
669 /* badly off, adjust it */
678 printf("Invalid time in real time clock.\n");
679 printf("Check and reset the date immediately!\n");
684 * Write system time back to RTC
692 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
697 if (xen_disable_rtc_set)
704 /* Disable RTC updates and interrupts. */
705 writertc(RTC_STATUSB, RTCSB_HALT | RTCSB_24HR);
707 /* Calculate local time to put in RTC */
709 tm -= tz_minuteswest * 60 + (wall_cmos_clock ? adjkerntz : 0);
711 writertc(RTC_SEC, bin2bcd(tm%60)); tm /= 60; /* Write back Seconds */
712 writertc(RTC_MIN, bin2bcd(tm%60)); tm /= 60; /* Write back Minutes */
713 writertc(RTC_HRS, bin2bcd(tm%24)); tm /= 24; /* Write back Hours */
715 /* We have now the days since 01-01-1970 in tm */
716 writertc(RTC_WDAY, (tm + 4) % 7 + 1); /* Write back Weekday */
717 for (y = 1970, m = DAYSPERYEAR + LEAPYEAR(y);
719 y++, m = DAYSPERYEAR + LEAPYEAR(y))
722 /* Now we have the years in y and the day-of-the-year in tm */
723 writertc(RTC_YEAR, bin2bcd(y%100)); /* Write back Year */
724 #ifdef USE_RTC_CENTURY
725 writertc(RTC_CENTURY, bin2bcd(y/100)); /* ... and Century */
731 if (m == 1 && LEAPYEAR(y))
738 writertc(RTC_MONTH, bin2bcd(m + 1)); /* Write back Month */
739 writertc(RTC_DAY, bin2bcd(tm + 1)); /* Write back Month Day */
741 /* Reenable RTC updates and interrupts. */
742 writertc(RTC_STATUSB, RTCSB_24HR);
747 static struct vcpu_set_periodic_timer xen_set_periodic_tick;
750 * Start clocks running.
757 xen_set_periodic_tick.period_ns = NS_PER_TICK;
759 HYPERVISOR_vcpu_op(VCPUOP_set_periodic_timer, 0,
760 &xen_set_periodic_tick);
762 if ((time_irq = bind_virq_to_irqhandler(VIRQ_TIMER, 0, "clk",
763 (driver_filter_t *)clkintr, NULL,
764 INTR_TYPE_CLK | INTR_FAST)) < 0) {
765 panic("failed to register clock interrupt\n");
768 /* should fast clock be enabled ? */
775 #if 0 && defined(SMP)
777 ap_cpu_initclocks(void)
780 int cpu = smp_processor_id();
782 per_cpu(processed_system_time, cpu) = processed_system_time;
784 irq = bind_virq_to_irq(VIRQ_TIMER);
785 PCPU_SET(time_irq, irq);
786 PANIC_IF(intr_add_handler("clk", irq, (driver_filter_t *)clkintr, NULL,
787 NULL, INTR_TYPE_CLK | INTR_FAST, NULL));
792 cpu_startprofclock(void)
795 printf("cpu_startprofclock: profiling clock is not supported\n");
799 cpu_stopprofclock(void)
802 printf("cpu_stopprofclock: profiling clock is not supported\n");
804 #define NSEC_PER_USEC 1000
807 xen_get_timecount(struct timecounter *tc)
810 struct shadow_time_info *shadow;
811 shadow = &per_cpu(shadow_time, smp_processor_id());
813 __get_time_values_from_xen();
815 clk = shadow->system_timestamp + get_nsec_offset(shadow);
817 return (uint32_t)((clk / NS_PER_TICK) * NS_PER_TICK);
821 /* Return system time offset by ticks */
823 get_system_time(int ticks)
825 return processed_system_time + (ticks * NS_PER_TICK);
829 * Track behavior of cur_timer->get_offset() functionality in timer_tsc.c
836 register unsigned long eax, edx;
838 /* Read the Time Stamp Counter */
842 /* .. relative to previous jiffy (32 bits is enough) */
843 eax -= shadow_tsc_stamp;
846 * Time offset = (tsc_low delta) * cached_gtm
847 * = (tsc_low delta) * (usecs_per_clock)
848 * = (tsc_low delta) * (usecs_per_jiffy / clocks_per_jiffy)
850 * Using a mull instead of a divl saves up to 31 clock cycles
851 * in the critical path.
855 :"=a" (eax), "=d" (edx)
859 /* our adjusted time offset in microseconds */
867 __get_time_values_from_xen();
868 PANIC_IF(HYPERVISOR_set_timer_op(processed_system_time + NS_PER_TICK) != 0);
869 HYPERVISOR_sched_op(SCHEDOP_block, 0);
873 timer_spkr_acquire(void)
880 timer_spkr_release(void)
887 timer_spkr_setfreq(int freq)