xref: /freebsd/sys/kern/kern_clocksource.c (revision 29fc4075e69fd27de0cded313ac6000165d99f8b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2010-2013 Alexander Motin <mav@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer,
12  *    without modification, immediately at the beginning of the file.
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  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 /*
33  * Common routines to manage event timers hardware.
34  */
35 
36 #include "opt_device_polling.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/bus.h>
41 #include <sys/limits.h>
42 #include <sys/lock.h>
43 #include <sys/kdb.h>
44 #include <sys/ktr.h>
45 #include <sys/mutex.h>
46 #include <sys/proc.h>
47 #include <sys/kernel.h>
48 #include <sys/sched.h>
49 #include <sys/smp.h>
50 #include <sys/sysctl.h>
51 #include <sys/timeet.h>
52 #include <sys/timetc.h>
53 
54 #include <machine/atomic.h>
55 #include <machine/clock.h>
56 #include <machine/cpu.h>
57 #include <machine/smp.h>
58 
59 int			cpu_disable_c2_sleep = 0; /* Timer dies in C2. */
60 int			cpu_disable_c3_sleep = 0; /* Timer dies in C3. */
61 
62 static void		setuptimer(void);
63 static void		loadtimer(sbintime_t now, int first);
64 static int		doconfigtimer(void);
65 static void		configtimer(int start);
66 static int		round_freq(struct eventtimer *et, int freq);
67 
68 struct pcpu_state;
69 static sbintime_t	getnextcpuevent(struct pcpu_state *state, int idle);
70 static sbintime_t	getnextevent(struct pcpu_state *state);
71 static int		handleevents(sbintime_t now, int fake);
72 
73 static struct mtx	et_hw_mtx;
74 
75 #define	ET_HW_LOCK(state)						\
76 	{								\
77 		if (timer->et_flags & ET_FLAGS_PERCPU)			\
78 			mtx_lock_spin(&(state)->et_hw_mtx);		\
79 		else							\
80 			mtx_lock_spin(&et_hw_mtx);			\
81 	}
82 
83 #define	ET_HW_UNLOCK(state)						\
84 	{								\
85 		if (timer->et_flags & ET_FLAGS_PERCPU)			\
86 			mtx_unlock_spin(&(state)->et_hw_mtx);		\
87 		else							\
88 			mtx_unlock_spin(&et_hw_mtx);			\
89 	}
90 
91 static struct eventtimer *timer = NULL;
92 static sbintime_t	timerperiod;	/* Timer period for periodic mode. */
93 static sbintime_t	statperiod;	/* statclock() events period. */
94 static sbintime_t	profperiod;	/* profclock() events period. */
95 static sbintime_t	nexttick;	/* Next global timer tick time. */
96 static u_int		busy = 1;	/* Reconfiguration is in progress. */
97 static int		profiling;	/* Profiling events enabled. */
98 
99 static char		timername[32];	/* Wanted timer. */
100 TUNABLE_STR("kern.eventtimer.timer", timername, sizeof(timername));
101 
102 static int		singlemul;	/* Multiplier for periodic mode. */
103 SYSCTL_INT(_kern_eventtimer, OID_AUTO, singlemul, CTLFLAG_RWTUN, &singlemul,
104     0, "Multiplier for periodic mode");
105 
106 static u_int		idletick;	/* Run periodic events when idle. */
107 SYSCTL_UINT(_kern_eventtimer, OID_AUTO, idletick, CTLFLAG_RWTUN, &idletick,
108     0, "Run periodic events when idle");
109 
110 static int		periodic;	/* Periodic or one-shot mode. */
111 static int		want_periodic;	/* What mode to prefer. */
112 TUNABLE_INT("kern.eventtimer.periodic", &want_periodic);
113 
114 struct pcpu_state {
115 	struct mtx	et_hw_mtx;	/* Per-CPU timer mutex. */
116 	u_int		action;		/* Reconfiguration requests. */
117 	u_int		handle;		/* Immediate handle resuests. */
118 	sbintime_t	now;		/* Last tick time. */
119 	sbintime_t	nextevent;	/* Next scheduled event on this CPU. */
120 	sbintime_t	nexttick;	/* Next timer tick time. */
121 	sbintime_t	nexthard;	/* Next hardclock() event. */
122 	sbintime_t	nextstat;	/* Next statclock() event. */
123 	sbintime_t	nextprof;	/* Next profclock() event. */
124 	sbintime_t	nextcall;	/* Next callout event. */
125 	sbintime_t	nextcallopt;	/* Next optional callout event. */
126 	int		ipi;		/* This CPU needs IPI. */
127 	int		idle;		/* This CPU is in idle mode. */
128 };
129 
130 DPCPU_DEFINE_STATIC(struct pcpu_state, timerstate);
131 DPCPU_DEFINE(sbintime_t, hardclocktime);
132 
133 /*
134  * Timer broadcast IPI handler.
135  */
136 int
137 hardclockintr(void)
138 {
139 	sbintime_t now;
140 	struct pcpu_state *state;
141 	int done;
142 
143 	if (doconfigtimer() || busy)
144 		return (FILTER_HANDLED);
145 	state = DPCPU_PTR(timerstate);
146 	now = state->now;
147 	CTR2(KTR_SPARE2, "ipi:    now  %d.%08x",
148 	    (int)(now >> 32), (u_int)(now & 0xffffffff));
149 	done = handleevents(now, 0);
150 	return (done ? FILTER_HANDLED : FILTER_STRAY);
151 }
152 
153 /*
154  * Handle all events for specified time on this CPU
155  */
156 static int
157 handleevents(sbintime_t now, int fake)
158 {
159 	sbintime_t t, *hct;
160 	struct trapframe *frame;
161 	struct pcpu_state *state;
162 	int usermode;
163 	int done, runs;
164 
165 	CTR2(KTR_SPARE2, "handle:  now  %d.%08x",
166 	    (int)(now >> 32), (u_int)(now & 0xffffffff));
167 	done = 0;
168 	if (fake) {
169 		frame = NULL;
170 		usermode = 0;
171 	} else {
172 		frame = curthread->td_intr_frame;
173 		usermode = TRAPF_USERMODE(frame);
174 	}
175 
176 	state = DPCPU_PTR(timerstate);
177 
178 	runs = 0;
179 	while (now >= state->nexthard) {
180 		state->nexthard += tick_sbt;
181 		runs++;
182 	}
183 	if (runs) {
184 		hct = DPCPU_PTR(hardclocktime);
185 		*hct = state->nexthard - tick_sbt;
186 		if (fake < 2) {
187 			hardclock(runs, usermode);
188 			done = 1;
189 		}
190 	}
191 	runs = 0;
192 	while (now >= state->nextstat) {
193 		state->nextstat += statperiod;
194 		runs++;
195 	}
196 	if (runs && fake < 2) {
197 		statclock(runs, usermode);
198 		done = 1;
199 	}
200 	if (profiling) {
201 		runs = 0;
202 		while (now >= state->nextprof) {
203 			state->nextprof += profperiod;
204 			runs++;
205 		}
206 		if (runs && !fake) {
207 			profclock(runs, usermode, TRAPF_PC(frame));
208 			done = 1;
209 		}
210 	} else
211 		state->nextprof = state->nextstat;
212 	if (now >= state->nextcallopt || now >= state->nextcall) {
213 		state->nextcall = state->nextcallopt = SBT_MAX;
214 		callout_process(now);
215 	}
216 
217 	ET_HW_LOCK(state);
218 	t = getnextcpuevent(state, 0);
219 	if (!busy) {
220 		state->idle = 0;
221 		state->nextevent = t;
222 		loadtimer(now, (fake == 2) &&
223 		    (timer->et_flags & ET_FLAGS_PERCPU));
224 	}
225 	ET_HW_UNLOCK(state);
226 	return (done);
227 }
228 
229 /*
230  * Schedule binuptime of the next event on current CPU.
231  */
232 static sbintime_t
233 getnextcpuevent(struct pcpu_state *state, int idle)
234 {
235 	sbintime_t event;
236 	u_int hardfreq;
237 
238 	/* Handle hardclock() events, skipping some if CPU is idle. */
239 	event = state->nexthard;
240 	if (idle) {
241 		if (tc_min_ticktock_freq > 1
242 #ifdef SMP
243 		    && curcpu == CPU_FIRST()
244 #endif
245 		    )
246 			hardfreq = hz / tc_min_ticktock_freq;
247 		else
248 			hardfreq = hz;
249 		if (hardfreq > 1)
250 			event += tick_sbt * (hardfreq - 1);
251 	}
252 	/* Handle callout events. */
253 	if (event > state->nextcall)
254 		event = state->nextcall;
255 	if (!idle) { /* If CPU is active - handle other types of events. */
256 		if (event > state->nextstat)
257 			event = state->nextstat;
258 		if (profiling && event > state->nextprof)
259 			event = state->nextprof;
260 	}
261 	return (event);
262 }
263 
264 /*
265  * Schedule binuptime of the next event on all CPUs.
266  */
267 static sbintime_t
268 getnextevent(struct pcpu_state *state)
269 {
270 	sbintime_t event;
271 #ifdef SMP
272 	int	cpu;
273 #endif
274 #ifdef KTR
275 	int	c;
276 
277 	c = -1;
278 #endif
279 	event = state->nextevent;
280 #ifdef SMP
281 	if ((timer->et_flags & ET_FLAGS_PERCPU) == 0) {
282 		CPU_FOREACH(cpu) {
283 			state = DPCPU_ID_PTR(cpu, timerstate);
284 			if (event > state->nextevent) {
285 				event = state->nextevent;
286 #ifdef KTR
287 				c = cpu;
288 #endif
289 			}
290 		}
291 	}
292 #endif
293 	CTR3(KTR_SPARE2, "next:    next %d.%08x by %d",
294 	    (int)(event >> 32), (u_int)(event & 0xffffffff), c);
295 	return (event);
296 }
297 
298 /* Hardware timer callback function. */
299 static void
300 timercb(struct eventtimer *et, void *arg)
301 {
302 	sbintime_t now;
303 	sbintime_t *next;
304 	struct pcpu_state *state;
305 #ifdef SMP
306 	int cpu, bcast;
307 #endif
308 
309 	/* Do not touch anything if somebody reconfiguring timers. */
310 	if (busy)
311 		return;
312 	/* Update present and next tick times. */
313 	state = DPCPU_PTR(timerstate);
314 	if (et->et_flags & ET_FLAGS_PERCPU) {
315 		next = &state->nexttick;
316 	} else
317 		next = &nexttick;
318 	now = sbinuptime();
319 	if (periodic)
320 		*next = now + timerperiod;
321 	else
322 		*next = -1;	/* Next tick is not scheduled yet. */
323 	state->now = now;
324 	CTR2(KTR_SPARE2, "intr:    now  %d.%08x",
325 	    (int)(now >> 32), (u_int)(now & 0xffffffff));
326 
327 #ifdef SMP
328 #ifdef EARLY_AP_STARTUP
329 	MPASS(mp_ncpus == 1 || smp_started);
330 #endif
331 	/* Prepare broadcasting to other CPUs for non-per-CPU timers. */
332 	bcast = 0;
333 #ifdef EARLY_AP_STARTUP
334 	if ((et->et_flags & ET_FLAGS_PERCPU) == 0) {
335 #else
336 	if ((et->et_flags & ET_FLAGS_PERCPU) == 0 && smp_started) {
337 #endif
338 		CPU_FOREACH(cpu) {
339 			state = DPCPU_ID_PTR(cpu, timerstate);
340 			ET_HW_LOCK(state);
341 			state->now = now;
342 			if (now >= state->nextevent) {
343 				state->nextevent += SBT_1S;
344 				if (curcpu != cpu) {
345 					state->ipi = 1;
346 					bcast = 1;
347 				}
348 			}
349 			ET_HW_UNLOCK(state);
350 		}
351 	}
352 #endif
353 
354 	/* Handle events for this time on this CPU. */
355 	handleevents(now, 0);
356 
357 #ifdef SMP
358 	/* Broadcast interrupt to other CPUs for non-per-CPU timers. */
359 	if (bcast) {
360 		CPU_FOREACH(cpu) {
361 			if (curcpu == cpu)
362 				continue;
363 			state = DPCPU_ID_PTR(cpu, timerstate);
364 			if (state->ipi) {
365 				state->ipi = 0;
366 				ipi_cpu(cpu, IPI_HARDCLOCK);
367 			}
368 		}
369 	}
370 #endif
371 }
372 
373 /*
374  * Load new value into hardware timer.
375  */
376 static void
377 loadtimer(sbintime_t now, int start)
378 {
379 	struct pcpu_state *state;
380 	sbintime_t new;
381 	sbintime_t *next;
382 	uint64_t tmp;
383 	int eq;
384 
385 	state = DPCPU_PTR(timerstate);
386 	if (timer->et_flags & ET_FLAGS_PERCPU)
387 		next = &state->nexttick;
388 	else
389 		next = &nexttick;
390 	if (periodic) {
391 		if (start) {
392 			/*
393 			 * Try to start all periodic timers aligned
394 			 * to period to make events synchronous.
395 			 */
396 			tmp = now % timerperiod;
397 			new = timerperiod - tmp;
398 			if (new < tmp)		/* Left less then passed. */
399 				new += timerperiod;
400 			CTR4(KTR_SPARE2, "load p:   now %d.%08x first in %d.%08x",
401 			    (int)(now >> 32), (u_int)(now & 0xffffffff),
402 			    (int)(new >> 32), (u_int)(new & 0xffffffff));
403 			*next = new + now;
404 			et_start(timer, new, timerperiod);
405 		}
406 	} else {
407 		new = getnextevent(state);
408 		eq = (new == *next);
409 		CTR3(KTR_SPARE2, "load:    next %d.%08x eq %d",
410 		    (int)(new >> 32), (u_int)(new & 0xffffffff), eq);
411 		if (!eq) {
412 			*next = new;
413 			et_start(timer, new - now, 0);
414 		}
415 	}
416 }
417 
418 /*
419  * Prepare event timer parameters after configuration changes.
420  */
421 static void
422 setuptimer(void)
423 {
424 	int freq;
425 
426 	if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
427 		periodic = 0;
428 	else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
429 		periodic = 1;
430 	singlemul = MIN(MAX(singlemul, 1), 20);
431 	freq = hz * singlemul;
432 	while (freq < (profiling ? profhz : stathz))
433 		freq += hz;
434 	freq = round_freq(timer, freq);
435 	timerperiod = SBT_1S / freq;
436 }
437 
438 /*
439  * Reconfigure specified per-CPU timer on other CPU. Called from IPI handler.
440  */
441 static int
442 doconfigtimer(void)
443 {
444 	sbintime_t now;
445 	struct pcpu_state *state;
446 
447 	state = DPCPU_PTR(timerstate);
448 	switch (atomic_load_acq_int(&state->action)) {
449 	case 1:
450 		now = sbinuptime();
451 		ET_HW_LOCK(state);
452 		loadtimer(now, 1);
453 		ET_HW_UNLOCK(state);
454 		state->handle = 0;
455 		atomic_store_rel_int(&state->action, 0);
456 		return (1);
457 	case 2:
458 		ET_HW_LOCK(state);
459 		et_stop(timer);
460 		ET_HW_UNLOCK(state);
461 		state->handle = 0;
462 		atomic_store_rel_int(&state->action, 0);
463 		return (1);
464 	}
465 	if (atomic_readandclear_int(&state->handle) && !busy) {
466 		now = sbinuptime();
467 		handleevents(now, 0);
468 		return (1);
469 	}
470 	return (0);
471 }
472 
473 /*
474  * Reconfigure specified timer.
475  * For per-CPU timers use IPI to make other CPUs to reconfigure.
476  */
477 static void
478 configtimer(int start)
479 {
480 	sbintime_t now, next;
481 	struct pcpu_state *state;
482 	int cpu;
483 
484 	if (start) {
485 		setuptimer();
486 		now = sbinuptime();
487 	} else
488 		now = 0;
489 	critical_enter();
490 	ET_HW_LOCK(DPCPU_PTR(timerstate));
491 	if (start) {
492 		/* Initialize time machine parameters. */
493 		next = now + timerperiod;
494 		if (periodic)
495 			nexttick = next;
496 		else
497 			nexttick = -1;
498 #ifdef EARLY_AP_STARTUP
499 		MPASS(mp_ncpus == 1 || smp_started);
500 #endif
501 		CPU_FOREACH(cpu) {
502 			state = DPCPU_ID_PTR(cpu, timerstate);
503 			state->now = now;
504 #ifndef EARLY_AP_STARTUP
505 			if (!smp_started && cpu != CPU_FIRST())
506 				state->nextevent = SBT_MAX;
507 			else
508 #endif
509 				state->nextevent = next;
510 			if (periodic)
511 				state->nexttick = next;
512 			else
513 				state->nexttick = -1;
514 			state->nexthard = next;
515 			state->nextstat = next;
516 			state->nextprof = next;
517 			state->nextcall = next;
518 			state->nextcallopt = next;
519 			hardclock_sync(cpu);
520 		}
521 		busy = 0;
522 		/* Start global timer or per-CPU timer of this CPU. */
523 		loadtimer(now, 1);
524 	} else {
525 		busy = 1;
526 		/* Stop global timer or per-CPU timer of this CPU. */
527 		et_stop(timer);
528 	}
529 	ET_HW_UNLOCK(DPCPU_PTR(timerstate));
530 #ifdef SMP
531 #ifdef EARLY_AP_STARTUP
532 	/* If timer is global we are done. */
533 	if ((timer->et_flags & ET_FLAGS_PERCPU) == 0) {
534 #else
535 	/* If timer is global or there is no other CPUs yet - we are done. */
536 	if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 || !smp_started) {
537 #endif
538 		critical_exit();
539 		return;
540 	}
541 	/* Set reconfigure flags for other CPUs. */
542 	CPU_FOREACH(cpu) {
543 		state = DPCPU_ID_PTR(cpu, timerstate);
544 		atomic_store_rel_int(&state->action,
545 		    (cpu == curcpu) ? 0 : ( start ? 1 : 2));
546 	}
547 	/* Broadcast reconfigure IPI. */
548 	ipi_all_but_self(IPI_HARDCLOCK);
549 	/* Wait for reconfiguration completed. */
550 restart:
551 	cpu_spinwait();
552 	CPU_FOREACH(cpu) {
553 		if (cpu == curcpu)
554 			continue;
555 		state = DPCPU_ID_PTR(cpu, timerstate);
556 		if (atomic_load_acq_int(&state->action))
557 			goto restart;
558 	}
559 #endif
560 	critical_exit();
561 }
562 
563 /*
564  * Calculate nearest frequency supported by hardware timer.
565  */
566 static int
567 round_freq(struct eventtimer *et, int freq)
568 {
569 	uint64_t div;
570 
571 	if (et->et_frequency != 0) {
572 		div = lmax((et->et_frequency + freq / 2) / freq, 1);
573 		if (et->et_flags & ET_FLAGS_POW2DIV)
574 			div = 1 << (flsl(div + div / 2) - 1);
575 		freq = (et->et_frequency + div / 2) / div;
576 	}
577 	if (et->et_min_period > SBT_1S)
578 		panic("Event timer \"%s\" doesn't support sub-second periods!",
579 		    et->et_name);
580 	else if (et->et_min_period != 0)
581 		freq = min(freq, SBT2FREQ(et->et_min_period));
582 	if (et->et_max_period < SBT_1S && et->et_max_period != 0)
583 		freq = max(freq, SBT2FREQ(et->et_max_period));
584 	return (freq);
585 }
586 
587 /*
588  * Configure and start event timers (BSP part).
589  */
590 void
591 cpu_initclocks_bsp(void)
592 {
593 	struct pcpu_state *state;
594 	int base, div, cpu;
595 
596 	mtx_init(&et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
597 	CPU_FOREACH(cpu) {
598 		state = DPCPU_ID_PTR(cpu, timerstate);
599 		mtx_init(&state->et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
600 		state->nextcall = SBT_MAX;
601 		state->nextcallopt = SBT_MAX;
602 	}
603 	periodic = want_periodic;
604 	/* Grab requested timer or the best of present. */
605 	if (timername[0])
606 		timer = et_find(timername, 0, 0);
607 	if (timer == NULL && periodic) {
608 		timer = et_find(NULL,
609 		    ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
610 	}
611 	if (timer == NULL) {
612 		timer = et_find(NULL,
613 		    ET_FLAGS_ONESHOT, ET_FLAGS_ONESHOT);
614 	}
615 	if (timer == NULL && !periodic) {
616 		timer = et_find(NULL,
617 		    ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
618 	}
619 	if (timer == NULL)
620 		panic("No usable event timer found!");
621 	et_init(timer, timercb, NULL, NULL);
622 
623 	/* Adapt to timer capabilities. */
624 	if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
625 		periodic = 0;
626 	else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
627 		periodic = 1;
628 	if (timer->et_flags & ET_FLAGS_C3STOP)
629 		cpu_disable_c3_sleep++;
630 
631 	/*
632 	 * We honor the requested 'hz' value.
633 	 * We want to run stathz in the neighborhood of 128hz.
634 	 * We would like profhz to run as often as possible.
635 	 */
636 	if (singlemul <= 0 || singlemul > 20) {
637 		if (hz >= 1500 || (hz % 128) == 0)
638 			singlemul = 1;
639 		else if (hz >= 750)
640 			singlemul = 2;
641 		else
642 			singlemul = 4;
643 	}
644 	if (periodic) {
645 		base = round_freq(timer, hz * singlemul);
646 		singlemul = max((base + hz / 2) / hz, 1);
647 		hz = (base + singlemul / 2) / singlemul;
648 		if (base <= 128)
649 			stathz = base;
650 		else {
651 			div = base / 128;
652 			if (div >= singlemul && (div % singlemul) == 0)
653 				div++;
654 			stathz = base / div;
655 		}
656 		profhz = stathz;
657 		while ((profhz + stathz) <= 128 * 64)
658 			profhz += stathz;
659 		profhz = round_freq(timer, profhz);
660 	} else {
661 		hz = round_freq(timer, hz);
662 		stathz = round_freq(timer, 127);
663 		profhz = round_freq(timer, stathz * 64);
664 	}
665 	tick = 1000000 / hz;
666 	tick_sbt = SBT_1S / hz;
667 	tick_bt = sbttobt(tick_sbt);
668 	statperiod = SBT_1S / stathz;
669 	profperiod = SBT_1S / profhz;
670 	ET_LOCK();
671 	configtimer(1);
672 	ET_UNLOCK();
673 }
674 
675 /*
676  * Start per-CPU event timers on APs.
677  */
678 void
679 cpu_initclocks_ap(void)
680 {
681 	struct pcpu_state *state;
682 	struct thread *td;
683 
684 	state = DPCPU_PTR(timerstate);
685 	ET_HW_LOCK(state);
686 	state->now = sbinuptime();
687 	hardclock_sync(curcpu);
688 	spinlock_enter();
689 	ET_HW_UNLOCK(state);
690 	td = curthread;
691 	td->td_intr_nesting_level++;
692 	handleevents(state->now, 2);
693 	td->td_intr_nesting_level--;
694 	spinlock_exit();
695 }
696 
697 void
698 suspendclock(void)
699 {
700 	ET_LOCK();
701 	configtimer(0);
702 	ET_UNLOCK();
703 }
704 
705 void
706 resumeclock(void)
707 {
708 	ET_LOCK();
709 	configtimer(1);
710 	ET_UNLOCK();
711 }
712 
713 /*
714  * Switch to profiling clock rates.
715  */
716 void
717 cpu_startprofclock(void)
718 {
719 
720 	ET_LOCK();
721 	if (profiling == 0) {
722 		if (periodic) {
723 			configtimer(0);
724 			profiling = 1;
725 			configtimer(1);
726 		} else
727 			profiling = 1;
728 	} else
729 		profiling++;
730 	ET_UNLOCK();
731 }
732 
733 /*
734  * Switch to regular clock rates.
735  */
736 void
737 cpu_stopprofclock(void)
738 {
739 
740 	ET_LOCK();
741 	if (profiling == 1) {
742 		if (periodic) {
743 			configtimer(0);
744 			profiling = 0;
745 			configtimer(1);
746 		} else
747 		profiling = 0;
748 	} else
749 		profiling--;
750 	ET_UNLOCK();
751 }
752 
753 /*
754  * Switch to idle mode (all ticks handled).
755  */
756 sbintime_t
757 cpu_idleclock(void)
758 {
759 	sbintime_t now, t;
760 	struct pcpu_state *state;
761 
762 	if (idletick || busy ||
763 	    (periodic && (timer->et_flags & ET_FLAGS_PERCPU))
764 #ifdef DEVICE_POLLING
765 	    || curcpu == CPU_FIRST()
766 #endif
767 	    )
768 		return (-1);
769 	state = DPCPU_PTR(timerstate);
770 	ET_HW_LOCK(state);
771 	if (periodic)
772 		now = state->now;
773 	else
774 		now = sbinuptime();
775 	CTR2(KTR_SPARE2, "idle:    now  %d.%08x",
776 	    (int)(now >> 32), (u_int)(now & 0xffffffff));
777 	t = getnextcpuevent(state, 1);
778 	state->idle = 1;
779 	state->nextevent = t;
780 	if (!periodic)
781 		loadtimer(now, 0);
782 	ET_HW_UNLOCK(state);
783 	return (MAX(t - now, 0));
784 }
785 
786 /*
787  * Switch to active mode (skip empty ticks).
788  */
789 void
790 cpu_activeclock(void)
791 {
792 	sbintime_t now;
793 	struct pcpu_state *state;
794 	struct thread *td;
795 
796 	state = DPCPU_PTR(timerstate);
797 	if (atomic_load_int(&state->idle) == 0 || busy)
798 		return;
799 	spinlock_enter();
800 	if (periodic)
801 		now = state->now;
802 	else
803 		now = sbinuptime();
804 	CTR2(KTR_SPARE2, "active:  now  %d.%08x",
805 	    (int)(now >> 32), (u_int)(now & 0xffffffff));
806 	td = curthread;
807 	td->td_intr_nesting_level++;
808 	handleevents(now, 1);
809 	td->td_intr_nesting_level--;
810 	spinlock_exit();
811 }
812 
813 /*
814  * Change the frequency of the given timer.  This changes et->et_frequency and
815  * if et is the active timer it reconfigures the timer on all CPUs.  This is
816  * intended to be a private interface for the use of et_change_frequency() only.
817  */
818 void
819 cpu_et_frequency(struct eventtimer *et, uint64_t newfreq)
820 {
821 
822 	ET_LOCK();
823 	if (et == timer) {
824 		configtimer(0);
825 		et->et_frequency = newfreq;
826 		configtimer(1);
827 	} else
828 		et->et_frequency = newfreq;
829 	ET_UNLOCK();
830 }
831 
832 void
833 cpu_new_callout(int cpu, sbintime_t bt, sbintime_t bt_opt)
834 {
835 	struct pcpu_state *state;
836 
837 	/* Do not touch anything if somebody reconfiguring timers. */
838 	if (busy)
839 		return;
840 
841 	CTR5(KTR_SPARE2, "new co:  on %d at %d.%08x - %d.%08x",
842 	    cpu, (int)(bt_opt >> 32), (u_int)(bt_opt & 0xffffffff),
843 	    (int)(bt >> 32), (u_int)(bt & 0xffffffff));
844 
845 	KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
846 	state = DPCPU_ID_PTR(cpu, timerstate);
847 	ET_HW_LOCK(state);
848 
849 	/*
850 	 * If there is callout time already set earlier -- do nothing.
851 	 * This check may appear redundant because we check already in
852 	 * callout_process() but this double check guarantees we're safe
853 	 * with respect to race conditions between interrupts execution
854 	 * and scheduling.
855 	 */
856 	state->nextcallopt = bt_opt;
857 	if (bt >= state->nextcall)
858 		goto done;
859 	state->nextcall = bt;
860 	/* If there is some other event set earlier -- do nothing. */
861 	if (bt >= state->nextevent)
862 		goto done;
863 	state->nextevent = bt;
864 	/* If timer is periodic -- there is nothing to reprogram. */
865 	if (periodic)
866 		goto done;
867 	/* If timer is global or of the current CPU -- reprogram it. */
868 	if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 || cpu == curcpu) {
869 		loadtimer(sbinuptime(), 0);
870 done:
871 		ET_HW_UNLOCK(state);
872 		return;
873 	}
874 	/* Otherwise make other CPU to reprogram it. */
875 	state->handle = 1;
876 	ET_HW_UNLOCK(state);
877 #ifdef SMP
878 	ipi_cpu(cpu, IPI_HARDCLOCK);
879 #endif
880 }
881 
882 /*
883  * Report or change the active event timers hardware.
884  */
885 static int
886 sysctl_kern_eventtimer_timer(SYSCTL_HANDLER_ARGS)
887 {
888 	char buf[32];
889 	struct eventtimer *et;
890 	int error;
891 
892 	ET_LOCK();
893 	et = timer;
894 	snprintf(buf, sizeof(buf), "%s", et->et_name);
895 	ET_UNLOCK();
896 	error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
897 	ET_LOCK();
898 	et = timer;
899 	if (error != 0 || req->newptr == NULL ||
900 	    strcasecmp(buf, et->et_name) == 0) {
901 		ET_UNLOCK();
902 		return (error);
903 	}
904 	et = et_find(buf, 0, 0);
905 	if (et == NULL) {
906 		ET_UNLOCK();
907 		return (ENOENT);
908 	}
909 	configtimer(0);
910 	et_free(timer);
911 	if (et->et_flags & ET_FLAGS_C3STOP)
912 		cpu_disable_c3_sleep++;
913 	if (timer->et_flags & ET_FLAGS_C3STOP)
914 		cpu_disable_c3_sleep--;
915 	periodic = want_periodic;
916 	timer = et;
917 	et_init(timer, timercb, NULL, NULL);
918 	configtimer(1);
919 	ET_UNLOCK();
920 	return (error);
921 }
922 SYSCTL_PROC(_kern_eventtimer, OID_AUTO, timer,
923     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE,
924     0, 0, sysctl_kern_eventtimer_timer, "A", "Chosen event timer");
925 
926 /*
927  * Report or change the active event timer periodicity.
928  */
929 static int
930 sysctl_kern_eventtimer_periodic(SYSCTL_HANDLER_ARGS)
931 {
932 	int error, val;
933 
934 	val = periodic;
935 	error = sysctl_handle_int(oidp, &val, 0, req);
936 	if (error != 0 || req->newptr == NULL)
937 		return (error);
938 	ET_LOCK();
939 	configtimer(0);
940 	periodic = want_periodic = val;
941 	configtimer(1);
942 	ET_UNLOCK();
943 	return (error);
944 }
945 SYSCTL_PROC(_kern_eventtimer, OID_AUTO, periodic,
946     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
947     0, 0, sysctl_kern_eventtimer_periodic, "I", "Enable event timer periodic mode");
948 
949 #include "opt_ddb.h"
950 
951 #ifdef DDB
952 #include <ddb/ddb.h>
953 
954 DB_SHOW_COMMAND(clocksource, db_show_clocksource)
955 {
956 	struct pcpu_state *st;
957 	int c;
958 
959 	CPU_FOREACH(c) {
960 		st = DPCPU_ID_PTR(c, timerstate);
961 		db_printf(
962 		    "CPU %2d: action %d handle %d  ipi %d idle %d\n"
963 		    "        now %#jx nevent %#jx (%jd)\n"
964 		    "        ntick %#jx (%jd) nhard %#jx (%jd)\n"
965 		    "        nstat %#jx (%jd) nprof %#jx (%jd)\n"
966 		    "        ncall %#jx (%jd) ncallopt %#jx (%jd)\n",
967 		    c, st->action, st->handle, st->ipi, st->idle,
968 		    (uintmax_t)st->now,
969 		    (uintmax_t)st->nextevent,
970 		    (uintmax_t)(st->nextevent - st->now) / tick_sbt,
971 		    (uintmax_t)st->nexttick,
972 		    (uintmax_t)(st->nexttick - st->now) / tick_sbt,
973 		    (uintmax_t)st->nexthard,
974 		    (uintmax_t)(st->nexthard - st->now) / tick_sbt,
975 		    (uintmax_t)st->nextstat,
976 		    (uintmax_t)(st->nextstat - st->now) / tick_sbt,
977 		    (uintmax_t)st->nextprof,
978 		    (uintmax_t)(st->nextprof - st->now) / tick_sbt,
979 		    (uintmax_t)st->nextcall,
980 		    (uintmax_t)(st->nextcall - st->now) / tick_sbt,
981 		    (uintmax_t)st->nextcallopt,
982 		    (uintmax_t)(st->nextcallopt - st->now) / tick_sbt);
983 	}
984 }
985 
986 #endif
987