2 * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer,
10 * without modification, immediately at the beginning of the file.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
31 * Common routines to manage event timers hardware.
34 #include "opt_device_polling.h"
35 #include "opt_kdtrace.h"
37 #include <sys/param.h>
38 #include <sys/systm.h>
43 #include <sys/mutex.h>
45 #include <sys/kernel.h>
46 #include <sys/sched.h>
48 #include <sys/sysctl.h>
49 #include <sys/timeet.h>
50 #include <sys/timetc.h>
52 #include <machine/atomic.h>
53 #include <machine/clock.h>
54 #include <machine/cpu.h>
55 #include <machine/smp.h>
58 #include <sys/dtrace_bsd.h>
59 cyclic_clock_func_t cyclic_clock_func = NULL;
62 int cpu_can_deep_sleep = 0; /* C3 state is available. */
63 int cpu_disable_deep_sleep = 0; /* Timer dies in C3. */
65 static void setuptimer(void);
66 static void loadtimer(struct bintime *now, int first);
67 static int doconfigtimer(void);
68 static void configtimer(int start);
69 static int round_freq(struct eventtimer *et, int freq);
71 static void getnextcpuevent(struct bintime *event, int idle);
72 static void getnextevent(struct bintime *event);
73 static int handleevents(struct bintime *now, int fake);
75 static void cpu_new_callout(int cpu, int ticks);
78 static struct mtx et_hw_mtx;
80 #define ET_HW_LOCK(state) \
82 if (timer->et_flags & ET_FLAGS_PERCPU) \
83 mtx_lock_spin(&(state)->et_hw_mtx); \
85 mtx_lock_spin(&et_hw_mtx); \
88 #define ET_HW_UNLOCK(state) \
90 if (timer->et_flags & ET_FLAGS_PERCPU) \
91 mtx_unlock_spin(&(state)->et_hw_mtx); \
93 mtx_unlock_spin(&et_hw_mtx); \
96 static struct eventtimer *timer = NULL;
97 static struct bintime timerperiod; /* Timer period for periodic mode. */
98 static struct bintime hardperiod; /* hardclock() events period. */
99 static struct bintime statperiod; /* statclock() events period. */
100 static struct bintime profperiod; /* profclock() events period. */
101 static struct bintime nexttick; /* Next global timer tick time. */
102 static u_int busy = 0; /* Reconfiguration is in progress. */
103 static int profiling = 0; /* Profiling events enabled. */
105 static char timername[32]; /* Wanted timer. */
106 TUNABLE_STR("kern.eventtimer.timer", timername, sizeof(timername));
108 static int singlemul = 0; /* Multiplier for periodic mode. */
109 TUNABLE_INT("kern.eventtimer.singlemul", &singlemul);
110 SYSCTL_INT(_kern_eventtimer, OID_AUTO, singlemul, CTLFLAG_RW, &singlemul,
111 0, "Multiplier for periodic mode");
113 static u_int idletick = 0; /* Idle mode allowed. */
114 TUNABLE_INT("kern.eventtimer.idletick", &idletick);
115 SYSCTL_UINT(_kern_eventtimer, OID_AUTO, idletick, CTLFLAG_RW, &idletick,
116 0, "Run periodic events when idle");
118 static int periodic = 0; /* Periodic or one-shot mode. */
119 static int want_periodic = 0; /* What mode to prefer. */
120 TUNABLE_INT("kern.eventtimer.periodic", &want_periodic);
123 struct mtx et_hw_mtx; /* Per-CPU timer mutex. */
124 u_int action; /* Reconfiguration requests. */
125 u_int handle; /* Immediate handle resuests. */
126 struct bintime now; /* Last tick time. */
127 struct bintime nextevent; /* Next scheduled event on this CPU. */
128 struct bintime nexttick; /* Next timer tick time. */
129 struct bintime nexthard; /* Next hardlock() event. */
130 struct bintime nextstat; /* Next statclock() event. */
131 struct bintime nextprof; /* Next profclock() event. */
133 struct bintime nextcyc; /* Next OpenSolaris cyclics event. */
135 int ipi; /* This CPU needs IPI. */
136 int idle; /* This CPU is in idle mode. */
139 static DPCPU_DEFINE(struct pcpu_state, timerstate);
141 #define FREQ2BT(freq, bt) \
144 (bt)->frac = ((uint64_t)0x8000000000000000 / (freq)) << 1; \
146 #define BT2FREQ(bt) \
147 (((uint64_t)0x8000000000000000 + ((bt)->frac >> 2)) / \
151 * Timer broadcast IPI handler.
157 struct pcpu_state *state;
160 if (doconfigtimer() || busy)
161 return (FILTER_HANDLED);
162 state = DPCPU_PTR(timerstate);
164 CTR4(KTR_SPARE2, "ipi at %d: now %d.%08x%08x",
165 curcpu, now.sec, (unsigned int)(now.frac >> 32),
166 (unsigned int)(now.frac & 0xffffffff));
167 done = handleevents(&now, 0);
168 return (done ? FILTER_HANDLED : FILTER_STRAY);
172 * Handle all events for specified time on this CPU
175 handleevents(struct bintime *now, int fake)
178 struct trapframe *frame;
179 struct pcpu_state *state;
184 CTR4(KTR_SPARE2, "handle at %d: now %d.%08x%08x",
185 curcpu, now->sec, (unsigned int)(now->frac >> 32),
186 (unsigned int)(now->frac & 0xffffffff));
193 frame = curthread->td_intr_frame;
194 usermode = TRAPF_USERMODE(frame);
195 pc = TRAPF_PC(frame);
199 state = DPCPU_PTR(timerstate);
201 while (bintime_cmp(now, &state->nexthard, >=)) {
202 bintime_add(&state->nexthard, &hardperiod);
205 if (runs && fake < 2) {
206 hardclock_anycpu(runs, usermode);
209 while (bintime_cmp(now, &state->nextstat, >=)) {
212 bintime_add(&state->nextstat, &statperiod);
216 while (bintime_cmp(now, &state->nextprof, >=)) {
218 profclock(usermode, pc);
219 bintime_add(&state->nextprof, &profperiod);
223 state->nextprof = state->nextstat;
226 if (fake == 0 && cyclic_clock_func != NULL &&
227 state->nextcyc.sec != -1 &&
228 bintime_cmp(now, &state->nextcyc, >=)) {
229 state->nextcyc.sec = -1;
230 (*cyclic_clock_func)(frame);
234 getnextcpuevent(&t, 0);
236 state->nextevent = t;
242 state->nextevent = t;
250 * Schedule binuptime of the next event on current CPU.
253 getnextcpuevent(struct bintime *event, int idle)
256 struct pcpu_state *state;
259 state = DPCPU_PTR(timerstate);
260 *event = state->nexthard;
261 if (idle) { /* If CPU is idle - ask callouts for how long. */
263 if (curcpu == CPU_FIRST() && tc_min_ticktock_freq > skip)
264 skip = tc_min_ticktock_freq;
265 skip = callout_tickstofirst(hz / skip) - 1;
266 CTR2(KTR_SPARE2, "skip at %d: %d", curcpu, skip);
268 bintime_mul(&tmp, skip);
269 bintime_add(event, &tmp);
270 } else { /* If CPU is active - handle all types of events. */
271 if (bintime_cmp(event, &state->nextstat, >))
272 *event = state->nextstat;
273 if (profiling && bintime_cmp(event, &state->nextprof, >))
274 *event = state->nextprof;
277 if (state->nextcyc.sec != -1 && bintime_cmp(event, &state->nextcyc, >))
278 *event = state->nextcyc;
283 * Schedule binuptime of the next event on all CPUs.
286 getnextevent(struct bintime *event)
288 struct pcpu_state *state;
294 state = DPCPU_PTR(timerstate);
295 *event = state->nextevent;
298 if ((timer->et_flags & ET_FLAGS_PERCPU) == 0) {
302 state = DPCPU_ID_PTR(cpu, timerstate);
303 if (bintime_cmp(event, &state->nextevent, >)) {
304 *event = state->nextevent;
310 CTR5(KTR_SPARE2, "next at %d: next %d.%08x%08x by %d",
311 curcpu, event->sec, (unsigned int)(event->frac >> 32),
312 (unsigned int)(event->frac & 0xffffffff), c);
315 /* Hardware timer callback function. */
317 timercb(struct eventtimer *et, void *arg)
320 struct bintime *next;
321 struct pcpu_state *state;
326 /* Do not touch anything if somebody reconfiguring timers. */
329 /* Update present and next tick times. */
330 state = DPCPU_PTR(timerstate);
331 if (et->et_flags & ET_FLAGS_PERCPU) {
332 next = &state->nexttick;
336 now = *next; /* Ex-next tick time becomes present time. */
337 bintime_add(next, &timerperiod); /* Next tick in 1 period. */
339 binuptime(&now); /* Get present time from hardware. */
340 next->sec = -1; /* Next tick is not scheduled yet. */
343 CTR4(KTR_SPARE2, "intr at %d: now %d.%08x%08x",
344 curcpu, now.sec, (unsigned int)(now.frac >> 32),
345 (unsigned int)(now.frac & 0xffffffff));
348 /* Prepare broadcasting to other CPUs for non-per-CPU timers. */
350 if ((et->et_flags & ET_FLAGS_PERCPU) == 0 && smp_started) {
352 state = DPCPU_ID_PTR(cpu, timerstate);
355 if (bintime_cmp(&now, &state->nextevent, >=)) {
356 state->nextevent.sec++;
367 /* Handle events for this time on this CPU. */
368 handleevents(&now, 0);
371 /* Broadcast interrupt to other CPUs for non-per-CPU timers. */
376 state = DPCPU_ID_PTR(cpu, timerstate);
379 ipi_cpu(cpu, IPI_HARDCLOCK);
387 * Load new value into hardware timer.
390 loadtimer(struct bintime *now, int start)
392 struct pcpu_state *state;
394 struct bintime *next;
398 if (timer->et_flags & ET_FLAGS_PERCPU) {
399 state = DPCPU_PTR(timerstate);
400 next = &state->nexttick;
406 * Try to start all periodic timers aligned
407 * to period to make events synchronous.
409 tmp = ((uint64_t)now->sec << 36) + (now->frac >> 28);
410 tmp = (tmp % (timerperiod.frac >> 28)) << 28;
412 new.frac = timerperiod.frac - tmp;
413 if (new.frac < tmp) /* Left less then passed. */
414 bintime_add(&new, &timerperiod);
415 CTR5(KTR_SPARE2, "load p at %d: now %d.%08x first in %d.%08x",
416 curcpu, now->sec, (unsigned int)(now->frac >> 32),
417 new.sec, (unsigned int)(new.frac >> 32));
419 bintime_add(next, now);
420 et_start(timer, &new, &timerperiod);
424 eq = bintime_cmp(&new, next, ==);
425 CTR5(KTR_SPARE2, "load at %d: next %d.%08x%08x eq %d",
426 curcpu, new.sec, (unsigned int)(new.frac >> 32),
427 (unsigned int)(new.frac & 0xffffffff),
431 bintime_sub(&new, now);
432 et_start(timer, &new, NULL);
438 * Prepare event timer parameters after configuration changes.
445 if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
447 else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
449 singlemul = MIN(MAX(singlemul, 1), 20);
450 freq = hz * singlemul;
451 while (freq < (profiling ? profhz : stathz))
453 freq = round_freq(timer, freq);
454 FREQ2BT(freq, &timerperiod);
458 * Reconfigure specified per-CPU timer on other CPU. Called from IPI handler.
464 struct pcpu_state *state;
466 state = DPCPU_PTR(timerstate);
467 switch (atomic_load_acq_int(&state->action)) {
474 atomic_store_rel_int(&state->action, 0);
481 atomic_store_rel_int(&state->action, 0);
484 if (atomic_readandclear_int(&state->handle) && !busy) {
486 handleevents(&now, 0);
493 * Reconfigure specified timer.
494 * For per-CPU timers use IPI to make other CPUs to reconfigure.
497 configtimer(int start)
499 struct bintime now, next;
500 struct pcpu_state *state;
508 ET_HW_LOCK(DPCPU_PTR(timerstate));
510 /* Initialize time machine parameters. */
512 bintime_add(&next, &timerperiod);
518 state = DPCPU_ID_PTR(cpu, timerstate);
520 state->nextevent = next;
522 state->nexttick = next;
524 state->nexttick.sec = -1;
525 state->nexthard = next;
526 state->nextstat = next;
527 state->nextprof = next;
531 /* Start global timer or per-CPU timer of this CPU. */
535 /* Stop global timer or per-CPU timer of this CPU. */
538 ET_HW_UNLOCK(DPCPU_PTR(timerstate));
540 /* If timer is global or there is no other CPUs yet - we are done. */
541 if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 || !smp_started) {
545 /* Set reconfigure flags for other CPUs. */
547 state = DPCPU_ID_PTR(cpu, timerstate);
548 atomic_store_rel_int(&state->action,
549 (cpu == curcpu) ? 0 : ( start ? 1 : 2));
551 /* Broadcast reconfigure IPI. */
552 ipi_all_but_self(IPI_HARDCLOCK);
553 /* Wait for reconfiguration completed. */
559 state = DPCPU_ID_PTR(cpu, timerstate);
560 if (atomic_load_acq_int(&state->action))
568 * Calculate nearest frequency supported by hardware timer.
571 round_freq(struct eventtimer *et, int freq)
575 if (et->et_frequency != 0) {
576 div = lmax((et->et_frequency + freq / 2) / freq, 1);
577 if (et->et_flags & ET_FLAGS_POW2DIV)
578 div = 1 << (flsl(div + div / 2) - 1);
579 freq = (et->et_frequency + div / 2) / div;
581 if (et->et_min_period.sec > 0)
583 else if (et->et_min_period.frac != 0)
584 freq = min(freq, BT2FREQ(&et->et_min_period));
585 if (et->et_max_period.sec == 0 && et->et_max_period.frac != 0)
586 freq = max(freq, BT2FREQ(&et->et_max_period));
591 * Configure and start event timers (BSP part).
594 cpu_initclocks_bsp(void)
596 struct pcpu_state *state;
599 mtx_init(&et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
601 state = DPCPU_ID_PTR(cpu, timerstate);
602 mtx_init(&state->et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
604 state->nextcyc.sec = -1;
608 callout_new_inserted = cpu_new_callout;
610 periodic = want_periodic;
611 /* Grab requested timer or the best of present. */
613 timer = et_find(timername, 0, 0);
614 if (timer == NULL && periodic) {
615 timer = et_find(NULL,
616 ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
619 timer = et_find(NULL,
620 ET_FLAGS_ONESHOT, ET_FLAGS_ONESHOT);
622 if (timer == NULL && !periodic) {
623 timer = et_find(NULL,
624 ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
627 panic("No usable event timer found!");
628 et_init(timer, timercb, NULL, NULL);
630 /* Adapt to timer capabilities. */
631 if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
633 else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
635 if (timer->et_flags & ET_FLAGS_C3STOP)
636 cpu_disable_deep_sleep++;
639 * We honor the requested 'hz' value.
640 * We want to run stathz in the neighborhood of 128hz.
641 * We would like profhz to run as often as possible.
643 if (singlemul <= 0 || singlemul > 20) {
644 if (hz >= 1500 || (hz % 128) == 0)
652 base = round_freq(timer, hz * singlemul);
653 singlemul = max((base + hz / 2) / hz, 1);
654 hz = (base + singlemul / 2) / singlemul;
659 if (div >= singlemul && (div % singlemul) == 0)
664 while ((profhz + stathz) <= 128 * 64)
666 profhz = round_freq(timer, profhz);
668 hz = round_freq(timer, hz);
669 stathz = round_freq(timer, 127);
670 profhz = round_freq(timer, stathz * 64);
673 FREQ2BT(hz, &hardperiod);
674 FREQ2BT(stathz, &statperiod);
675 FREQ2BT(profhz, &profperiod);
682 * Start per-CPU event timers on APs.
685 cpu_initclocks_ap(void)
688 struct pcpu_state *state;
690 state = DPCPU_PTR(timerstate);
693 if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 && periodic) {
694 state->now = nexttick;
695 bintime_sub(&state->now, &timerperiod);
698 hardclock_sync(curcpu);
699 handleevents(&state->now, 2);
700 if (timer->et_flags & ET_FLAGS_PERCPU)
706 * Switch to profiling clock rates.
709 cpu_startprofclock(void)
723 * Switch to regular clock rates.
726 cpu_stopprofclock(void)
740 * Switch to idle mode (all ticks handled).
745 struct bintime now, t;
746 struct pcpu_state *state;
748 if (idletick || busy ||
749 (periodic && (timer->et_flags & ET_FLAGS_PERCPU))
750 #ifdef DEVICE_POLLING
751 || curcpu == CPU_FIRST()
755 state = DPCPU_PTR(timerstate);
760 CTR4(KTR_SPARE2, "idle at %d: now %d.%08x%08x",
761 curcpu, now.sec, (unsigned int)(now.frac >> 32),
762 (unsigned int)(now.frac & 0xffffffff));
763 getnextcpuevent(&t, 1);
766 state->nextevent = t;
773 * Switch to active mode (skip empty ticks).
776 cpu_activeclock(void)
779 struct pcpu_state *state;
782 state = DPCPU_PTR(timerstate);
783 if (state->idle == 0 || busy)
789 CTR4(KTR_SPARE2, "active at %d: now %d.%08x%08x",
790 curcpu, now.sec, (unsigned int)(now.frac >> 32),
791 (unsigned int)(now.frac & 0xffffffff));
794 td->td_intr_nesting_level++;
795 handleevents(&now, 1);
796 td->td_intr_nesting_level--;
802 clocksource_cyc_set(const struct bintime *t)
805 struct pcpu_state *state;
807 state = DPCPU_PTR(timerstate);
813 CTR4(KTR_SPARE2, "set_cyc at %d: now %d.%08x%08x",
814 curcpu, now.sec, (unsigned int)(now.frac >> 32),
815 (unsigned int)(now.frac & 0xffffffff));
816 CTR4(KTR_SPARE2, "set_cyc at %d: t %d.%08x%08x",
817 curcpu, t->sec, (unsigned int)(t->frac >> 32),
818 (unsigned int)(t->frac & 0xffffffff));
821 if (bintime_cmp(t, &state->nextcyc, ==)) {
826 if (bintime_cmp(&state->nextcyc, &state->nextevent, >=)) {
830 state->nextevent = state->nextcyc;
839 cpu_new_callout(int cpu, int ticks)
842 struct pcpu_state *state;
844 CTR3(KTR_SPARE2, "new co at %d: on %d in %d",
846 state = DPCPU_ID_PTR(cpu, timerstate);
848 if (state->idle == 0 || busy) {
853 * If timer is periodic - just update next event time for target CPU.
854 * If timer is global - there is chance it is already programmed.
856 if (periodic || (timer->et_flags & ET_FLAGS_PERCPU) == 0) {
857 state->nextevent = state->nexthard;
859 bintime_mul(&tmp, ticks - 1);
860 bintime_add(&state->nextevent, &tmp);
862 bintime_cmp(&state->nextevent, &nexttick, >=)) {
868 * Otherwise we have to wake that CPU up, as we can't get present
869 * bintime to reprogram global timer from here. If timer is per-CPU,
870 * we by definition can't do it from here.
873 if (timer->et_flags & ET_FLAGS_PERCPU) {
875 ipi_cpu(cpu, IPI_HARDCLOCK);
877 if (!cpu_idle_wakeup(cpu))
878 ipi_cpu(cpu, IPI_AST);
884 * Report or change the active event timers hardware.
887 sysctl_kern_eventtimer_timer(SYSCTL_HANDLER_ARGS)
890 struct eventtimer *et;
895 snprintf(buf, sizeof(buf), "%s", et->et_name);
897 error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
900 if (error != 0 || req->newptr == NULL ||
901 strcasecmp(buf, et->et_name) == 0) {
905 et = et_find(buf, 0, 0);
912 if (et->et_flags & ET_FLAGS_C3STOP)
913 cpu_disable_deep_sleep++;
914 if (timer->et_flags & ET_FLAGS_C3STOP)
915 cpu_disable_deep_sleep--;
916 periodic = want_periodic;
918 et_init(timer, timercb, NULL, NULL);
923 SYSCTL_PROC(_kern_eventtimer, OID_AUTO, timer,
924 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE,
925 0, 0, sysctl_kern_eventtimer_timer, "A", "Chosen event timer");
928 * Report or change the active event timer periodicity.
931 sysctl_kern_eventtimer_periodic(SYSCTL_HANDLER_ARGS)
936 error = sysctl_handle_int(oidp, &val, 0, req);
937 if (error != 0 || req->newptr == NULL)
941 periodic = want_periodic = val;
946 SYSCTL_PROC(_kern_eventtimer, OID_AUTO, periodic,
947 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
948 0, 0, sysctl_kern_eventtimer_periodic, "I", "Enable event timer periodic mode");