2 * Copyright (c) 2010-2012 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 struct bintime nexthard; /* Next global hardlock() event. */
103 static u_int busy = 0; /* Reconfiguration is in progress. */
104 static int profiling = 0; /* Profiling events enabled. */
106 static char timername[32]; /* Wanted timer. */
107 TUNABLE_STR("kern.eventtimer.timer", timername, sizeof(timername));
109 static int singlemul = 0; /* Multiplier for periodic mode. */
110 TUNABLE_INT("kern.eventtimer.singlemul", &singlemul);
111 SYSCTL_INT(_kern_eventtimer, OID_AUTO, singlemul, CTLFLAG_RW, &singlemul,
112 0, "Multiplier for periodic mode");
114 static u_int idletick = 0; /* Run periodic events when idle. */
115 TUNABLE_INT("kern.eventtimer.idletick", &idletick);
116 SYSCTL_UINT(_kern_eventtimer, OID_AUTO, idletick, CTLFLAG_RW, &idletick,
117 0, "Run periodic events when idle");
119 static u_int activetick = 1; /* Run all periodic events when active. */
120 TUNABLE_INT("kern.eventtimer.activetick", &activetick);
121 SYSCTL_UINT(_kern_eventtimer, OID_AUTO, activetick, CTLFLAG_RW, &activetick,
122 0, "Run all periodic events when active");
124 static int periodic = 0; /* Periodic or one-shot mode. */
125 static int want_periodic = 0; /* What mode to prefer. */
126 TUNABLE_INT("kern.eventtimer.periodic", &want_periodic);
129 struct mtx et_hw_mtx; /* Per-CPU timer mutex. */
130 u_int action; /* Reconfiguration requests. */
131 u_int handle; /* Immediate handle resuests. */
132 struct bintime now; /* Last tick time. */
133 struct bintime nextevent; /* Next scheduled event on this CPU. */
134 struct bintime nexttick; /* Next timer tick time. */
135 struct bintime nexthard; /* Next hardlock() event. */
136 struct bintime nextstat; /* Next statclock() event. */
137 struct bintime nextprof; /* Next profclock() event. */
139 struct bintime nextcyc; /* Next OpenSolaris cyclics event. */
141 int ipi; /* This CPU needs IPI. */
142 int idle; /* This CPU is in idle mode. */
145 static DPCPU_DEFINE(struct pcpu_state, timerstate);
147 #define FREQ2BT(freq, bt) \
150 (bt)->frac = ((uint64_t)0x8000000000000000 / (freq)) << 1; \
152 #define BT2FREQ(bt) \
153 (((uint64_t)0x8000000000000000 + ((bt)->frac >> 2)) / \
157 * Timer broadcast IPI handler.
163 struct pcpu_state *state;
166 if (doconfigtimer() || busy)
167 return (FILTER_HANDLED);
168 state = DPCPU_PTR(timerstate);
170 CTR4(KTR_SPARE2, "ipi at %d: now %d.%08x%08x",
171 curcpu, now.sec, (unsigned int)(now.frac >> 32),
172 (unsigned int)(now.frac & 0xffffffff));
173 done = handleevents(&now, 0);
174 return (done ? FILTER_HANDLED : FILTER_STRAY);
178 * Handle all events for specified time on this CPU
181 handleevents(struct bintime *now, int fake)
184 struct trapframe *frame;
185 struct pcpu_state *state;
190 CTR4(KTR_SPARE2, "handle at %d: now %d.%08x%08x",
191 curcpu, now->sec, (unsigned int)(now->frac >> 32),
192 (unsigned int)(now->frac & 0xffffffff));
199 frame = curthread->td_intr_frame;
200 usermode = TRAPF_USERMODE(frame);
201 pc = TRAPF_PC(frame);
204 state = DPCPU_PTR(timerstate);
207 while (bintime_cmp(now, &state->nexthard, >=)) {
208 bintime_add(&state->nexthard, &hardperiod);
211 if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 &&
212 bintime_cmp(&state->nexthard, &nexthard, >))
213 nexthard = state->nexthard;
214 if (runs && fake < 2) {
215 hardclock_cnt(runs, usermode);
219 while (bintime_cmp(now, &state->nextstat, >=)) {
220 bintime_add(&state->nextstat, &statperiod);
223 if (runs && fake < 2) {
224 statclock_cnt(runs, usermode);
229 while (bintime_cmp(now, &state->nextprof, >=)) {
230 bintime_add(&state->nextprof, &profperiod);
234 profclock_cnt(runs, usermode, pc);
238 state->nextprof = state->nextstat;
241 if (fake == 0 && cyclic_clock_func != NULL &&
242 state->nextcyc.sec != -1 &&
243 bintime_cmp(now, &state->nextcyc, >=)) {
244 state->nextcyc.sec = -1;
245 (*cyclic_clock_func)(frame);
249 getnextcpuevent(&t, 0);
251 state->nextevent = t;
257 state->nextevent = t;
265 * Schedule binuptime of the next event on current CPU.
268 getnextcpuevent(struct bintime *event, int idle)
271 struct pcpu_state *state;
274 state = DPCPU_PTR(timerstate);
275 /* Handle hardclock() events. */
276 *event = state->nexthard;
277 if (idle || (!activetick && !profiling &&
278 (timer->et_flags & ET_FLAGS_PERCPU) == 0)) {
279 skip = idle ? 4 : (stathz / 2);
280 if (curcpu == CPU_FIRST() && tc_min_ticktock_freq > skip)
281 skip = tc_min_ticktock_freq;
282 skip = callout_tickstofirst(hz / skip) - 1;
283 CTR2(KTR_SPARE2, "skip at %d: %d", curcpu, skip);
285 bintime_mul(&tmp, skip);
286 bintime_add(event, &tmp);
288 if (!idle) { /* If CPU is active - handle other types of events. */
289 if (bintime_cmp(event, &state->nextstat, >))
290 *event = state->nextstat;
291 if (profiling && bintime_cmp(event, &state->nextprof, >))
292 *event = state->nextprof;
295 if (state->nextcyc.sec != -1 && bintime_cmp(event, &state->nextcyc, >))
296 *event = state->nextcyc;
301 * Schedule binuptime of the next event on all CPUs.
304 getnextevent(struct bintime *event)
306 struct pcpu_state *state;
312 state = DPCPU_PTR(timerstate);
313 *event = state->nextevent;
315 nonidle = !state->idle;
316 if ((timer->et_flags & ET_FLAGS_PERCPU) == 0) {
321 state = DPCPU_ID_PTR(cpu, timerstate);
322 nonidle += !state->idle;
323 if (bintime_cmp(event, &state->nextevent, >)) {
324 *event = state->nextevent;
329 if (nonidle != 0 && bintime_cmp(event, &nexthard, >))
332 CTR5(KTR_SPARE2, "next at %d: next %d.%08x%08x by %d",
333 curcpu, event->sec, (unsigned int)(event->frac >> 32),
334 (unsigned int)(event->frac & 0xffffffff), c);
337 /* Hardware timer callback function. */
339 timercb(struct eventtimer *et, void *arg)
342 struct bintime *next;
343 struct pcpu_state *state;
348 /* Do not touch anything if somebody reconfiguring timers. */
351 /* Update present and next tick times. */
352 state = DPCPU_PTR(timerstate);
353 if (et->et_flags & ET_FLAGS_PERCPU) {
354 next = &state->nexttick;
358 now = *next; /* Ex-next tick time becomes present time. */
359 bintime_add(next, &timerperiod); /* Next tick in 1 period. */
361 binuptime(&now); /* Get present time from hardware. */
362 next->sec = -1; /* Next tick is not scheduled yet. */
365 CTR4(KTR_SPARE2, "intr at %d: now %d.%08x%08x",
366 curcpu, now.sec, (unsigned int)(now.frac >> 32),
367 (unsigned int)(now.frac & 0xffffffff));
370 /* Prepare broadcasting to other CPUs for non-per-CPU timers. */
372 if ((et->et_flags & ET_FLAGS_PERCPU) == 0 && smp_started) {
374 state = DPCPU_ID_PTR(cpu, timerstate);
377 if (bintime_cmp(&now, &state->nextevent, >=)) {
378 state->nextevent.sec++;
389 /* Handle events for this time on this CPU. */
390 handleevents(&now, 0);
393 /* Broadcast interrupt to other CPUs for non-per-CPU timers. */
398 state = DPCPU_ID_PTR(cpu, timerstate);
401 ipi_cpu(cpu, IPI_HARDCLOCK);
409 * Load new value into hardware timer.
412 loadtimer(struct bintime *now, int start)
414 struct pcpu_state *state;
416 struct bintime *next;
420 if (timer->et_flags & ET_FLAGS_PERCPU) {
421 state = DPCPU_PTR(timerstate);
422 next = &state->nexttick;
428 * Try to start all periodic timers aligned
429 * to period to make events synchronous.
431 tmp = ((uint64_t)now->sec << 36) + (now->frac >> 28);
432 tmp = (tmp % (timerperiod.frac >> 28)) << 28;
434 new.frac = timerperiod.frac - tmp;
435 if (new.frac < tmp) /* Left less then passed. */
436 bintime_add(&new, &timerperiod);
437 CTR5(KTR_SPARE2, "load p at %d: now %d.%08x first in %d.%08x",
438 curcpu, now->sec, (unsigned int)(now->frac >> 32),
439 new.sec, (unsigned int)(new.frac >> 32));
441 bintime_add(next, now);
442 et_start(timer, &new, &timerperiod);
446 eq = bintime_cmp(&new, next, ==);
447 CTR5(KTR_SPARE2, "load at %d: next %d.%08x%08x eq %d",
448 curcpu, new.sec, (unsigned int)(new.frac >> 32),
449 (unsigned int)(new.frac & 0xffffffff),
453 bintime_sub(&new, now);
454 et_start(timer, &new, NULL);
460 * Prepare event timer parameters after configuration changes.
467 if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
469 else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
471 singlemul = MIN(MAX(singlemul, 1), 20);
472 freq = hz * singlemul;
473 while (freq < (profiling ? profhz : stathz))
475 freq = round_freq(timer, freq);
476 FREQ2BT(freq, &timerperiod);
480 * Reconfigure specified per-CPU timer on other CPU. Called from IPI handler.
486 struct pcpu_state *state;
488 state = DPCPU_PTR(timerstate);
489 switch (atomic_load_acq_int(&state->action)) {
496 atomic_store_rel_int(&state->action, 0);
503 atomic_store_rel_int(&state->action, 0);
506 if (atomic_readandclear_int(&state->handle) && !busy) {
508 handleevents(&now, 0);
515 * Reconfigure specified timer.
516 * For per-CPU timers use IPI to make other CPUs to reconfigure.
519 configtimer(int start)
521 struct bintime now, next;
522 struct pcpu_state *state;
530 ET_HW_LOCK(DPCPU_PTR(timerstate));
532 /* Initialize time machine parameters. */
534 bintime_add(&next, &timerperiod);
540 state = DPCPU_ID_PTR(cpu, timerstate);
542 state->nextevent = next;
544 state->nexttick = next;
546 state->nexttick.sec = -1;
547 state->nexthard = next;
548 state->nextstat = next;
549 state->nextprof = next;
553 /* Start global timer or per-CPU timer of this CPU. */
557 /* Stop global timer or per-CPU timer of this CPU. */
560 ET_HW_UNLOCK(DPCPU_PTR(timerstate));
562 /* If timer is global or there is no other CPUs yet - we are done. */
563 if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 || !smp_started) {
567 /* Set reconfigure flags for other CPUs. */
569 state = DPCPU_ID_PTR(cpu, timerstate);
570 atomic_store_rel_int(&state->action,
571 (cpu == curcpu) ? 0 : ( start ? 1 : 2));
573 /* Broadcast reconfigure IPI. */
574 ipi_all_but_self(IPI_HARDCLOCK);
575 /* Wait for reconfiguration completed. */
581 state = DPCPU_ID_PTR(cpu, timerstate);
582 if (atomic_load_acq_int(&state->action))
590 * Calculate nearest frequency supported by hardware timer.
593 round_freq(struct eventtimer *et, int freq)
597 if (et->et_frequency != 0) {
598 div = lmax((et->et_frequency + freq / 2) / freq, 1);
599 if (et->et_flags & ET_FLAGS_POW2DIV)
600 div = 1 << (flsl(div + div / 2) - 1);
601 freq = (et->et_frequency + div / 2) / div;
603 if (et->et_min_period.sec > 0)
605 else if (et->et_min_period.frac != 0)
606 freq = min(freq, BT2FREQ(&et->et_min_period));
607 if (et->et_max_period.sec == 0 && et->et_max_period.frac != 0)
608 freq = max(freq, BT2FREQ(&et->et_max_period));
613 * Configure and start event timers (BSP part).
616 cpu_initclocks_bsp(void)
618 struct pcpu_state *state;
621 mtx_init(&et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
623 state = DPCPU_ID_PTR(cpu, timerstate);
624 mtx_init(&state->et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
626 state->nextcyc.sec = -1;
630 callout_new_inserted = cpu_new_callout;
632 periodic = want_periodic;
633 /* Grab requested timer or the best of present. */
635 timer = et_find(timername, 0, 0);
636 if (timer == NULL && periodic) {
637 timer = et_find(NULL,
638 ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
641 timer = et_find(NULL,
642 ET_FLAGS_ONESHOT, ET_FLAGS_ONESHOT);
644 if (timer == NULL && !periodic) {
645 timer = et_find(NULL,
646 ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
649 panic("No usable event timer found!");
650 et_init(timer, timercb, NULL, NULL);
652 /* Adapt to timer capabilities. */
653 if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
655 else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
657 if (timer->et_flags & ET_FLAGS_C3STOP)
658 cpu_disable_deep_sleep++;
661 * We honor the requested 'hz' value.
662 * We want to run stathz in the neighborhood of 128hz.
663 * We would like profhz to run as often as possible.
665 if (singlemul <= 0 || singlemul > 20) {
666 if (hz >= 1500 || (hz % 128) == 0)
674 base = round_freq(timer, hz * singlemul);
675 singlemul = max((base + hz / 2) / hz, 1);
676 hz = (base + singlemul / 2) / singlemul;
681 if (div >= singlemul && (div % singlemul) == 0)
686 while ((profhz + stathz) <= 128 * 64)
688 profhz = round_freq(timer, profhz);
690 hz = round_freq(timer, hz);
691 stathz = round_freq(timer, 127);
692 profhz = round_freq(timer, stathz * 64);
695 FREQ2BT(hz, &hardperiod);
696 FREQ2BT(stathz, &statperiod);
697 FREQ2BT(profhz, &profperiod);
704 * Start per-CPU event timers on APs.
707 cpu_initclocks_ap(void)
710 struct pcpu_state *state;
712 state = DPCPU_PTR(timerstate);
715 if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 && periodic) {
716 state->now = nexttick;
717 bintime_sub(&state->now, &timerperiod);
720 hardclock_sync(curcpu);
721 handleevents(&state->now, 2);
722 if (timer->et_flags & ET_FLAGS_PERCPU)
728 * Switch to profiling clock rates.
731 cpu_startprofclock(void)
735 if (profiling == 0) {
748 * Switch to regular clock rates.
751 cpu_stopprofclock(void)
755 if (profiling == 1) {
768 * Switch to idle mode (all ticks handled).
773 struct bintime now, t;
774 struct pcpu_state *state;
776 if (idletick || busy ||
777 (periodic && (timer->et_flags & ET_FLAGS_PERCPU))
778 #ifdef DEVICE_POLLING
779 || curcpu == CPU_FIRST()
783 state = DPCPU_PTR(timerstate);
788 CTR4(KTR_SPARE2, "idle at %d: now %d.%08x%08x",
789 curcpu, now.sec, (unsigned int)(now.frac >> 32),
790 (unsigned int)(now.frac & 0xffffffff));
791 getnextcpuevent(&t, 1);
794 state->nextevent = t;
801 * Switch to active mode (skip empty ticks).
804 cpu_activeclock(void)
807 struct pcpu_state *state;
810 state = DPCPU_PTR(timerstate);
811 if (state->idle == 0 || busy)
817 CTR4(KTR_SPARE2, "active at %d: now %d.%08x%08x",
818 curcpu, now.sec, (unsigned int)(now.frac >> 32),
819 (unsigned int)(now.frac & 0xffffffff));
822 td->td_intr_nesting_level++;
823 handleevents(&now, 1);
824 td->td_intr_nesting_level--;
830 clocksource_cyc_set(const struct bintime *t)
833 struct pcpu_state *state;
835 state = DPCPU_PTR(timerstate);
841 CTR4(KTR_SPARE2, "set_cyc at %d: now %d.%08x%08x",
842 curcpu, now.sec, (unsigned int)(now.frac >> 32),
843 (unsigned int)(now.frac & 0xffffffff));
844 CTR4(KTR_SPARE2, "set_cyc at %d: t %d.%08x%08x",
845 curcpu, t->sec, (unsigned int)(t->frac >> 32),
846 (unsigned int)(t->frac & 0xffffffff));
849 if (bintime_cmp(t, &state->nextcyc, ==)) {
854 if (bintime_cmp(&state->nextcyc, &state->nextevent, >=)) {
858 state->nextevent = state->nextcyc;
867 cpu_new_callout(int cpu, int ticks)
870 struct pcpu_state *state;
872 CTR3(KTR_SPARE2, "new co at %d: on %d in %d",
874 state = DPCPU_ID_PTR(cpu, timerstate);
876 if (state->idle == 0 || busy) {
881 * If timer is periodic - just update next event time for target CPU.
882 * If timer is global - there is chance it is already programmed.
884 if (periodic || (timer->et_flags & ET_FLAGS_PERCPU) == 0) {
886 bintime_mul(&tmp, ticks - 1);
887 bintime_add(&tmp, &state->nexthard);
888 if (bintime_cmp(&tmp, &state->nextevent, <))
889 state->nextevent = tmp;
891 bintime_cmp(&state->nextevent, &nexttick, >=)) {
897 * Otherwise we have to wake that CPU up, as we can't get present
898 * bintime to reprogram global timer from here. If timer is per-CPU,
899 * we by definition can't do it from here.
902 if (timer->et_flags & ET_FLAGS_PERCPU) {
904 ipi_cpu(cpu, IPI_HARDCLOCK);
906 if (!cpu_idle_wakeup(cpu))
907 ipi_cpu(cpu, IPI_AST);
913 * Report or change the active event timers hardware.
916 sysctl_kern_eventtimer_timer(SYSCTL_HANDLER_ARGS)
919 struct eventtimer *et;
924 snprintf(buf, sizeof(buf), "%s", et->et_name);
926 error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
929 if (error != 0 || req->newptr == NULL ||
930 strcasecmp(buf, et->et_name) == 0) {
934 et = et_find(buf, 0, 0);
941 if (et->et_flags & ET_FLAGS_C3STOP)
942 cpu_disable_deep_sleep++;
943 if (timer->et_flags & ET_FLAGS_C3STOP)
944 cpu_disable_deep_sleep--;
945 periodic = want_periodic;
947 et_init(timer, timercb, NULL, NULL);
952 SYSCTL_PROC(_kern_eventtimer, OID_AUTO, timer,
953 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE,
954 0, 0, sysctl_kern_eventtimer_timer, "A", "Chosen event timer");
957 * Report or change the active event timer periodicity.
960 sysctl_kern_eventtimer_periodic(SYSCTL_HANDLER_ARGS)
965 error = sysctl_handle_int(oidp, &val, 0, req);
966 if (error != 0 || req->newptr == NULL)
970 periodic = want_periodic = val;
975 SYSCTL_PROC(_kern_eventtimer, OID_AUTO, periodic,
976 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
977 0, 0, sysctl_kern_eventtimer_periodic, "I", "Enable event timer periodic mode");