kern_clocksource.c revision 267961
1/*-
2 * Copyright (c) 2010-2013 Alexander Motin <mav@FreeBSD.org>
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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.
14 *
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.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: head/sys/kern/kern_clocksource.c 267961 2014-06-27 16:33:43Z hselasky $");
29
30/*
31 * Common routines to manage event timers hardware.
32 */
33
34#include "opt_device_polling.h"
35
36#include <sys/param.h>
37#include <sys/systm.h>
38#include <sys/bus.h>
39#include <sys/limits.h>
40#include <sys/lock.h>
41#include <sys/kdb.h>
42#include <sys/ktr.h>
43#include <sys/mutex.h>
44#include <sys/proc.h>
45#include <sys/kernel.h>
46#include <sys/sched.h>
47#include <sys/smp.h>
48#include <sys/sysctl.h>
49#include <sys/timeet.h>
50#include <sys/timetc.h>
51
52#include <machine/atomic.h>
53#include <machine/clock.h>
54#include <machine/cpu.h>
55#include <machine/smp.h>
56
57#ifdef KDTRACE_HOOKS
58#include <sys/dtrace_bsd.h>
59cyclic_clock_func_t	cyclic_clock_func = NULL;
60#endif
61
62int			cpu_can_deep_sleep = 0;	/* C3 state is available. */
63int			cpu_disable_deep_sleep = 0; /* Timer dies in C3. */
64
65static void		setuptimer(void);
66static void		loadtimer(sbintime_t now, int first);
67static int		doconfigtimer(void);
68static void		configtimer(int start);
69static int		round_freq(struct eventtimer *et, int freq);
70
71static sbintime_t	getnextcpuevent(int idle);
72static sbintime_t	getnextevent(void);
73static int		handleevents(sbintime_t now, int fake);
74
75static struct mtx	et_hw_mtx;
76
77#define	ET_HW_LOCK(state)						\
78	{								\
79		if (timer->et_flags & ET_FLAGS_PERCPU)			\
80			mtx_lock_spin(&(state)->et_hw_mtx);		\
81		else							\
82			mtx_lock_spin(&et_hw_mtx);			\
83	}
84
85#define	ET_HW_UNLOCK(state)						\
86	{								\
87		if (timer->et_flags & ET_FLAGS_PERCPU)			\
88			mtx_unlock_spin(&(state)->et_hw_mtx);		\
89		else							\
90			mtx_unlock_spin(&et_hw_mtx);			\
91	}
92
93static struct eventtimer *timer = NULL;
94static sbintime_t	timerperiod;	/* Timer period for periodic mode. */
95static sbintime_t	statperiod;	/* statclock() events period. */
96static sbintime_t	profperiod;	/* profclock() events period. */
97static sbintime_t	nexttick;	/* Next global timer tick time. */
98static u_int		busy = 1;	/* Reconfiguration is in progress. */
99static int		profiling;	/* Profiling events enabled. */
100
101static char		timername[32];	/* Wanted timer. */
102TUNABLE_STR("kern.eventtimer.timer", timername, sizeof(timername));
103
104static int		singlemul;	/* Multiplier for periodic mode. */
105SYSCTL_INT(_kern_eventtimer, OID_AUTO, singlemul, CTLFLAG_RWTUN, &singlemul,
106    0, "Multiplier for periodic mode");
107
108static u_int		idletick;	/* Run periodic events when idle. */
109SYSCTL_UINT(_kern_eventtimer, OID_AUTO, idletick, CTLFLAG_RWTUN, &idletick,
110    0, "Run periodic events when idle");
111
112static int		periodic;	/* Periodic or one-shot mode. */
113static int		want_periodic;	/* What mode to prefer. */
114TUNABLE_INT("kern.eventtimer.periodic", &want_periodic);
115
116struct pcpu_state {
117	struct mtx	et_hw_mtx;	/* Per-CPU timer mutex. */
118	u_int		action;		/* Reconfiguration requests. */
119	u_int		handle;		/* Immediate handle resuests. */
120	sbintime_t	now;		/* Last tick time. */
121	sbintime_t	nextevent;	/* Next scheduled event on this CPU. */
122	sbintime_t	nexttick;	/* Next timer tick time. */
123	sbintime_t	nexthard;	/* Next hardlock() event. */
124	sbintime_t	nextstat;	/* Next statclock() event. */
125	sbintime_t	nextprof;	/* Next profclock() event. */
126	sbintime_t	nextcall;	/* Next callout event. */
127	sbintime_t	nextcallopt;	/* Next optional callout event. */
128#ifdef KDTRACE_HOOKS
129	sbintime_t	nextcyc;	/* Next OpenSolaris cyclics event. */
130#endif
131	int		ipi;		/* This CPU needs IPI. */
132	int		idle;		/* This CPU is in idle mode. */
133};
134
135static DPCPU_DEFINE(struct pcpu_state, timerstate);
136DPCPU_DEFINE(sbintime_t, hardclocktime);
137
138/*
139 * Timer broadcast IPI handler.
140 */
141int
142hardclockintr(void)
143{
144	sbintime_t now;
145	struct pcpu_state *state;
146	int done;
147
148	if (doconfigtimer() || busy)
149		return (FILTER_HANDLED);
150	state = DPCPU_PTR(timerstate);
151	now = state->now;
152	CTR3(KTR_SPARE2, "ipi  at %d:    now  %d.%08x",
153	    curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
154	done = handleevents(now, 0);
155	return (done ? FILTER_HANDLED : FILTER_STRAY);
156}
157
158/*
159 * Handle all events for specified time on this CPU
160 */
161static int
162handleevents(sbintime_t now, int fake)
163{
164	sbintime_t t, *hct;
165	struct trapframe *frame;
166	struct pcpu_state *state;
167	int usermode;
168	int done, runs;
169
170	CTR3(KTR_SPARE2, "handle at %d:  now  %d.%08x",
171	    curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
172	done = 0;
173	if (fake) {
174		frame = NULL;
175		usermode = 0;
176	} else {
177		frame = curthread->td_intr_frame;
178		usermode = TRAPF_USERMODE(frame);
179	}
180
181	state = DPCPU_PTR(timerstate);
182
183	runs = 0;
184	while (now >= state->nexthard) {
185		state->nexthard += tick_sbt;
186		runs++;
187	}
188	if (runs) {
189		hct = DPCPU_PTR(hardclocktime);
190		*hct = state->nexthard - tick_sbt;
191		if (fake < 2) {
192			hardclock_cnt(runs, usermode);
193			done = 1;
194		}
195	}
196	runs = 0;
197	while (now >= state->nextstat) {
198		state->nextstat += statperiod;
199		runs++;
200	}
201	if (runs && fake < 2) {
202		statclock_cnt(runs, usermode);
203		done = 1;
204	}
205	if (profiling) {
206		runs = 0;
207		while (now >= state->nextprof) {
208			state->nextprof += profperiod;
209			runs++;
210		}
211		if (runs && !fake) {
212			profclock_cnt(runs, usermode, TRAPF_PC(frame));
213			done = 1;
214		}
215	} else
216		state->nextprof = state->nextstat;
217	if (now >= state->nextcallopt) {
218		state->nextcall = state->nextcallopt = SBT_MAX;
219		callout_process(now);
220	}
221
222#ifdef KDTRACE_HOOKS
223	if (fake == 0 && now >= state->nextcyc && cyclic_clock_func != NULL) {
224		state->nextcyc = SBT_MAX;
225		(*cyclic_clock_func)(frame);
226	}
227#endif
228
229	t = getnextcpuevent(0);
230	ET_HW_LOCK(state);
231	if (!busy) {
232		state->idle = 0;
233		state->nextevent = t;
234		loadtimer(now, (fake == 2) &&
235		    (timer->et_flags & ET_FLAGS_PERCPU));
236	}
237	ET_HW_UNLOCK(state);
238	return (done);
239}
240
241/*
242 * Schedule binuptime of the next event on current CPU.
243 */
244static sbintime_t
245getnextcpuevent(int idle)
246{
247	sbintime_t event;
248	struct pcpu_state *state;
249	u_int hardfreq;
250
251	state = DPCPU_PTR(timerstate);
252	/* Handle hardclock() events, skipping some if CPU is idle. */
253	event = state->nexthard;
254	if (idle) {
255		hardfreq = (u_int)hz / 2;
256		if (tc_min_ticktock_freq > 2
257#ifdef SMP
258		    && curcpu == CPU_FIRST()
259#endif
260		    )
261			hardfreq = hz / tc_min_ticktock_freq;
262		if (hardfreq > 1)
263			event += tick_sbt * (hardfreq - 1);
264	}
265	/* Handle callout events. */
266	if (event > state->nextcall)
267		event = state->nextcall;
268	if (!idle) { /* If CPU is active - handle other types of events. */
269		if (event > state->nextstat)
270			event = state->nextstat;
271		if (profiling && event > state->nextprof)
272			event = state->nextprof;
273	}
274#ifdef KDTRACE_HOOKS
275	if (event > state->nextcyc)
276		event = state->nextcyc;
277#endif
278	return (event);
279}
280
281/*
282 * Schedule binuptime of the next event on all CPUs.
283 */
284static sbintime_t
285getnextevent(void)
286{
287	struct pcpu_state *state;
288	sbintime_t event;
289#ifdef SMP
290	int	cpu;
291#endif
292	int	c;
293
294	state = DPCPU_PTR(timerstate);
295	event = state->nextevent;
296	c = -1;
297#ifdef SMP
298	if ((timer->et_flags & ET_FLAGS_PERCPU) == 0) {
299		CPU_FOREACH(cpu) {
300			state = DPCPU_ID_PTR(cpu, timerstate);
301			if (event > state->nextevent) {
302				event = state->nextevent;
303				c = cpu;
304			}
305		}
306	}
307#endif
308	CTR4(KTR_SPARE2, "next at %d:    next %d.%08x by %d",
309	    curcpu, (int)(event >> 32), (u_int)(event & 0xffffffff), c);
310	return (event);
311}
312
313/* Hardware timer callback function. */
314static void
315timercb(struct eventtimer *et, void *arg)
316{
317	sbintime_t now;
318	sbintime_t *next;
319	struct pcpu_state *state;
320#ifdef SMP
321	int cpu, bcast;
322#endif
323
324	/* Do not touch anything if somebody reconfiguring timers. */
325	if (busy)
326		return;
327	/* Update present and next tick times. */
328	state = DPCPU_PTR(timerstate);
329	if (et->et_flags & ET_FLAGS_PERCPU) {
330		next = &state->nexttick;
331	} else
332		next = &nexttick;
333	now = sbinuptime();
334	if (periodic)
335		*next = now + timerperiod;
336	else
337		*next = -1;	/* Next tick is not scheduled yet. */
338	state->now = now;
339	CTR3(KTR_SPARE2, "intr at %d:    now  %d.%08x",
340	    curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
341
342#ifdef SMP
343	/* Prepare broadcasting to other CPUs for non-per-CPU timers. */
344	bcast = 0;
345	if ((et->et_flags & ET_FLAGS_PERCPU) == 0 && smp_started) {
346		CPU_FOREACH(cpu) {
347			state = DPCPU_ID_PTR(cpu, timerstate);
348			ET_HW_LOCK(state);
349			state->now = now;
350			if (now >= state->nextevent) {
351				state->nextevent += SBT_1S;
352				if (curcpu != cpu) {
353					state->ipi = 1;
354					bcast = 1;
355				}
356			}
357			ET_HW_UNLOCK(state);
358		}
359	}
360#endif
361
362	/* Handle events for this time on this CPU. */
363	handleevents(now, 0);
364
365#ifdef SMP
366	/* Broadcast interrupt to other CPUs for non-per-CPU timers. */
367	if (bcast) {
368		CPU_FOREACH(cpu) {
369			if (curcpu == cpu)
370				continue;
371			state = DPCPU_ID_PTR(cpu, timerstate);
372			if (state->ipi) {
373				state->ipi = 0;
374				ipi_cpu(cpu, IPI_HARDCLOCK);
375			}
376		}
377	}
378#endif
379}
380
381/*
382 * Load new value into hardware timer.
383 */
384static void
385loadtimer(sbintime_t now, int start)
386{
387	struct pcpu_state *state;
388	sbintime_t new;
389	sbintime_t *next;
390	uint64_t tmp;
391	int eq;
392
393	if (timer->et_flags & ET_FLAGS_PERCPU) {
394		state = DPCPU_PTR(timerstate);
395		next = &state->nexttick;
396	} else
397		next = &nexttick;
398	if (periodic) {
399		if (start) {
400			/*
401			 * Try to start all periodic timers aligned
402			 * to period to make events synchronous.
403			 */
404			tmp = now % timerperiod;
405			new = timerperiod - tmp;
406			if (new < tmp)		/* Left less then passed. */
407				new += timerperiod;
408			CTR5(KTR_SPARE2, "load p at %d:   now %d.%08x first in %d.%08x",
409			    curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff),
410			    (int)(new >> 32), (u_int)(new & 0xffffffff));
411			*next = new + now;
412			et_start(timer, new, timerperiod);
413		}
414	} else {
415		new = getnextevent();
416		eq = (new == *next);
417		CTR4(KTR_SPARE2, "load at %d:    next %d.%08x eq %d",
418		    curcpu, (int)(new >> 32), (u_int)(new & 0xffffffff), eq);
419		if (!eq) {
420			*next = new;
421			et_start(timer, new - now, 0);
422		}
423	}
424}
425
426/*
427 * Prepare event timer parameters after configuration changes.
428 */
429static void
430setuptimer(void)
431{
432	int freq;
433
434	if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
435		periodic = 0;
436	else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
437		periodic = 1;
438	singlemul = MIN(MAX(singlemul, 1), 20);
439	freq = hz * singlemul;
440	while (freq < (profiling ? profhz : stathz))
441		freq += hz;
442	freq = round_freq(timer, freq);
443	timerperiod = SBT_1S / freq;
444}
445
446/*
447 * Reconfigure specified per-CPU timer on other CPU. Called from IPI handler.
448 */
449static int
450doconfigtimer(void)
451{
452	sbintime_t now;
453	struct pcpu_state *state;
454
455	state = DPCPU_PTR(timerstate);
456	switch (atomic_load_acq_int(&state->action)) {
457	case 1:
458		now = sbinuptime();
459		ET_HW_LOCK(state);
460		loadtimer(now, 1);
461		ET_HW_UNLOCK(state);
462		state->handle = 0;
463		atomic_store_rel_int(&state->action, 0);
464		return (1);
465	case 2:
466		ET_HW_LOCK(state);
467		et_stop(timer);
468		ET_HW_UNLOCK(state);
469		state->handle = 0;
470		atomic_store_rel_int(&state->action, 0);
471		return (1);
472	}
473	if (atomic_readandclear_int(&state->handle) && !busy) {
474		now = sbinuptime();
475		handleevents(now, 0);
476		return (1);
477	}
478	return (0);
479}
480
481/*
482 * Reconfigure specified timer.
483 * For per-CPU timers use IPI to make other CPUs to reconfigure.
484 */
485static void
486configtimer(int start)
487{
488	sbintime_t now, next;
489	struct pcpu_state *state;
490	int cpu;
491
492	if (start) {
493		setuptimer();
494		now = sbinuptime();
495	} else
496		now = 0;
497	critical_enter();
498	ET_HW_LOCK(DPCPU_PTR(timerstate));
499	if (start) {
500		/* Initialize time machine parameters. */
501		next = now + timerperiod;
502		if (periodic)
503			nexttick = next;
504		else
505			nexttick = -1;
506		CPU_FOREACH(cpu) {
507			state = DPCPU_ID_PTR(cpu, timerstate);
508			state->now = now;
509			if (!smp_started && cpu != CPU_FIRST())
510				state->nextevent = SBT_MAX;
511			else
512				state->nextevent = next;
513			if (periodic)
514				state->nexttick = next;
515			else
516				state->nexttick = -1;
517			state->nexthard = next;
518			state->nextstat = next;
519			state->nextprof = next;
520			state->nextcall = next;
521			state->nextcallopt = next;
522			hardclock_sync(cpu);
523		}
524		busy = 0;
525		/* Start global timer or per-CPU timer of this CPU. */
526		loadtimer(now, 1);
527	} else {
528		busy = 1;
529		/* Stop global timer or per-CPU timer of this CPU. */
530		et_stop(timer);
531	}
532	ET_HW_UNLOCK(DPCPU_PTR(timerstate));
533#ifdef SMP
534	/* If timer is global or there is no other CPUs yet - we are done. */
535	if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 || !smp_started) {
536		critical_exit();
537		return;
538	}
539	/* Set reconfigure flags for other CPUs. */
540	CPU_FOREACH(cpu) {
541		state = DPCPU_ID_PTR(cpu, timerstate);
542		atomic_store_rel_int(&state->action,
543		    (cpu == curcpu) ? 0 : ( start ? 1 : 2));
544	}
545	/* Broadcast reconfigure IPI. */
546	ipi_all_but_self(IPI_HARDCLOCK);
547	/* Wait for reconfiguration completed. */
548restart:
549	cpu_spinwait();
550	CPU_FOREACH(cpu) {
551		if (cpu == curcpu)
552			continue;
553		state = DPCPU_ID_PTR(cpu, timerstate);
554		if (atomic_load_acq_int(&state->action))
555			goto restart;
556	}
557#endif
558	critical_exit();
559}
560
561/*
562 * Calculate nearest frequency supported by hardware timer.
563 */
564static int
565round_freq(struct eventtimer *et, int freq)
566{
567	uint64_t div;
568
569	if (et->et_frequency != 0) {
570		div = lmax((et->et_frequency + freq / 2) / freq, 1);
571		if (et->et_flags & ET_FLAGS_POW2DIV)
572			div = 1 << (flsl(div + div / 2) - 1);
573		freq = (et->et_frequency + div / 2) / div;
574	}
575	if (et->et_min_period > SBT_1S)
576		panic("Event timer \"%s\" doesn't support sub-second periods!",
577		    et->et_name);
578	else if (et->et_min_period != 0)
579		freq = min(freq, SBT2FREQ(et->et_min_period));
580	if (et->et_max_period < SBT_1S && et->et_max_period != 0)
581		freq = max(freq, SBT2FREQ(et->et_max_period));
582	return (freq);
583}
584
585/*
586 * Configure and start event timers (BSP part).
587 */
588void
589cpu_initclocks_bsp(void)
590{
591	struct pcpu_state *state;
592	int base, div, cpu;
593
594	mtx_init(&et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
595	CPU_FOREACH(cpu) {
596		state = DPCPU_ID_PTR(cpu, timerstate);
597		mtx_init(&state->et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
598#ifdef KDTRACE_HOOKS
599		state->nextcyc = SBT_MAX;
600#endif
601		state->nextcall = SBT_MAX;
602		state->nextcallopt = SBT_MAX;
603	}
604	periodic = want_periodic;
605	/* Grab requested timer or the best of present. */
606	if (timername[0])
607		timer = et_find(timername, 0, 0);
608	if (timer == NULL && periodic) {
609		timer = et_find(NULL,
610		    ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
611	}
612	if (timer == NULL) {
613		timer = et_find(NULL,
614		    ET_FLAGS_ONESHOT, ET_FLAGS_ONESHOT);
615	}
616	if (timer == NULL && !periodic) {
617		timer = et_find(NULL,
618		    ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
619	}
620	if (timer == NULL)
621		panic("No usable event timer found!");
622	et_init(timer, timercb, NULL, NULL);
623
624	/* Adapt to timer capabilities. */
625	if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
626		periodic = 0;
627	else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
628		periodic = 1;
629	if (timer->et_flags & ET_FLAGS_C3STOP)
630		cpu_disable_deep_sleep++;
631
632	/*
633	 * We honor the requested 'hz' value.
634	 * We want to run stathz in the neighborhood of 128hz.
635	 * We would like profhz to run as often as possible.
636	 */
637	if (singlemul <= 0 || singlemul > 20) {
638		if (hz >= 1500 || (hz % 128) == 0)
639			singlemul = 1;
640		else if (hz >= 750)
641			singlemul = 2;
642		else
643			singlemul = 4;
644	}
645	if (periodic) {
646		base = round_freq(timer, hz * singlemul);
647		singlemul = max((base + hz / 2) / hz, 1);
648		hz = (base + singlemul / 2) / singlemul;
649		if (base <= 128)
650			stathz = base;
651		else {
652			div = base / 128;
653			if (div >= singlemul && (div % singlemul) == 0)
654				div++;
655			stathz = base / div;
656		}
657		profhz = stathz;
658		while ((profhz + stathz) <= 128 * 64)
659			profhz += stathz;
660		profhz = round_freq(timer, profhz);
661	} else {
662		hz = round_freq(timer, hz);
663		stathz = round_freq(timer, 127);
664		profhz = round_freq(timer, stathz * 64);
665	}
666	tick = 1000000 / hz;
667	tick_sbt = SBT_1S / hz;
668	tick_bt = sbttobt(tick_sbt);
669	statperiod = SBT_1S / stathz;
670	profperiod = SBT_1S / profhz;
671	ET_LOCK();
672	configtimer(1);
673	ET_UNLOCK();
674}
675
676/*
677 * Start per-CPU event timers on APs.
678 */
679void
680cpu_initclocks_ap(void)
681{
682	sbintime_t now;
683	struct pcpu_state *state;
684	struct thread *td;
685
686	state = DPCPU_PTR(timerstate);
687	now = sbinuptime();
688	ET_HW_LOCK(state);
689	state->now = now;
690	hardclock_sync(curcpu);
691	spinlock_enter();
692	ET_HW_UNLOCK(state);
693	td = curthread;
694	td->td_intr_nesting_level++;
695	handleevents(state->now, 2);
696	td->td_intr_nesting_level--;
697	spinlock_exit();
698}
699
700/*
701 * Switch to profiling clock rates.
702 */
703void
704cpu_startprofclock(void)
705{
706
707	ET_LOCK();
708	if (profiling == 0) {
709		if (periodic) {
710			configtimer(0);
711			profiling = 1;
712			configtimer(1);
713		} else
714			profiling = 1;
715	} else
716		profiling++;
717	ET_UNLOCK();
718}
719
720/*
721 * Switch to regular clock rates.
722 */
723void
724cpu_stopprofclock(void)
725{
726
727	ET_LOCK();
728	if (profiling == 1) {
729		if (periodic) {
730			configtimer(0);
731			profiling = 0;
732			configtimer(1);
733		} else
734		profiling = 0;
735	} else
736		profiling--;
737	ET_UNLOCK();
738}
739
740/*
741 * Switch to idle mode (all ticks handled).
742 */
743sbintime_t
744cpu_idleclock(void)
745{
746	sbintime_t now, t;
747	struct pcpu_state *state;
748
749	if (idletick || busy ||
750	    (periodic && (timer->et_flags & ET_FLAGS_PERCPU))
751#ifdef DEVICE_POLLING
752	    || curcpu == CPU_FIRST()
753#endif
754	    )
755		return (-1);
756	state = DPCPU_PTR(timerstate);
757	if (periodic)
758		now = state->now;
759	else
760		now = sbinuptime();
761	CTR3(KTR_SPARE2, "idle at %d:    now  %d.%08x",
762	    curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
763	t = getnextcpuevent(1);
764	ET_HW_LOCK(state);
765	state->idle = 1;
766	state->nextevent = t;
767	if (!periodic)
768		loadtimer(now, 0);
769	ET_HW_UNLOCK(state);
770	return (MAX(t - now, 0));
771}
772
773/*
774 * Switch to active mode (skip empty ticks).
775 */
776void
777cpu_activeclock(void)
778{
779	sbintime_t now;
780	struct pcpu_state *state;
781	struct thread *td;
782
783	state = DPCPU_PTR(timerstate);
784	if (state->idle == 0 || busy)
785		return;
786	if (periodic)
787		now = state->now;
788	else
789		now = sbinuptime();
790	CTR3(KTR_SPARE2, "active at %d:  now  %d.%08x",
791	    curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
792	spinlock_enter();
793	td = curthread;
794	td->td_intr_nesting_level++;
795	handleevents(now, 1);
796	td->td_intr_nesting_level--;
797	spinlock_exit();
798}
799
800/*
801 * Change the frequency of the given timer.  This changes et->et_frequency and
802 * if et is the active timer it reconfigures the timer on all CPUs.  This is
803 * intended to be a private interface for the use of et_change_frequency() only.
804 */
805void
806cpu_et_frequency(struct eventtimer *et, uint64_t newfreq)
807{
808
809	ET_LOCK();
810	if (et == timer) {
811		configtimer(0);
812		et->et_frequency = newfreq;
813		configtimer(1);
814	} else
815		et->et_frequency = newfreq;
816	ET_UNLOCK();
817}
818
819#ifdef KDTRACE_HOOKS
820void
821clocksource_cyc_set(const struct bintime *bt)
822{
823	sbintime_t now, t;
824	struct pcpu_state *state;
825
826	/* Do not touch anything if somebody reconfiguring timers. */
827	if (busy)
828		return;
829	t = bttosbt(*bt);
830	state = DPCPU_PTR(timerstate);
831	if (periodic)
832		now = state->now;
833	else
834		now = sbinuptime();
835
836	CTR5(KTR_SPARE2, "set_cyc at %d:  now  %d.%08x  t  %d.%08x",
837	    curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff),
838	    (int)(t >> 32), (u_int)(t & 0xffffffff));
839
840	ET_HW_LOCK(state);
841	if (t == state->nextcyc)
842		goto done;
843	state->nextcyc = t;
844	if (t >= state->nextevent)
845		goto done;
846	state->nextevent = t;
847	if (!periodic)
848		loadtimer(now, 0);
849done:
850	ET_HW_UNLOCK(state);
851}
852#endif
853
854void
855cpu_new_callout(int cpu, sbintime_t bt, sbintime_t bt_opt)
856{
857	struct pcpu_state *state;
858
859	/* Do not touch anything if somebody reconfiguring timers. */
860	if (busy)
861		return;
862	CTR6(KTR_SPARE2, "new co at %d:    on %d at %d.%08x - %d.%08x",
863	    curcpu, cpu, (int)(bt_opt >> 32), (u_int)(bt_opt & 0xffffffff),
864	    (int)(bt >> 32), (u_int)(bt & 0xffffffff));
865	state = DPCPU_ID_PTR(cpu, timerstate);
866	ET_HW_LOCK(state);
867
868	/*
869	 * If there is callout time already set earlier -- do nothing.
870	 * This check may appear redundant because we check already in
871	 * callout_process() but this double check guarantees we're safe
872	 * with respect to race conditions between interrupts execution
873	 * and scheduling.
874	 */
875	state->nextcallopt = bt_opt;
876	if (bt >= state->nextcall)
877		goto done;
878	state->nextcall = bt;
879	/* If there is some other event set earlier -- do nothing. */
880	if (bt >= state->nextevent)
881		goto done;
882	state->nextevent = bt;
883	/* If timer is periodic -- there is nothing to reprogram. */
884	if (periodic)
885		goto done;
886	/* If timer is global or of the current CPU -- reprogram it. */
887	if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 || cpu == curcpu) {
888		loadtimer(sbinuptime(), 0);
889done:
890		ET_HW_UNLOCK(state);
891		return;
892	}
893	/* Otherwise make other CPU to reprogram it. */
894	state->handle = 1;
895	ET_HW_UNLOCK(state);
896#ifdef SMP
897	ipi_cpu(cpu, IPI_HARDCLOCK);
898#endif
899}
900
901/*
902 * Report or change the active event timers hardware.
903 */
904static int
905sysctl_kern_eventtimer_timer(SYSCTL_HANDLER_ARGS)
906{
907	char buf[32];
908	struct eventtimer *et;
909	int error;
910
911	ET_LOCK();
912	et = timer;
913	snprintf(buf, sizeof(buf), "%s", et->et_name);
914	ET_UNLOCK();
915	error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
916	ET_LOCK();
917	et = timer;
918	if (error != 0 || req->newptr == NULL ||
919	    strcasecmp(buf, et->et_name) == 0) {
920		ET_UNLOCK();
921		return (error);
922	}
923	et = et_find(buf, 0, 0);
924	if (et == NULL) {
925		ET_UNLOCK();
926		return (ENOENT);
927	}
928	configtimer(0);
929	et_free(timer);
930	if (et->et_flags & ET_FLAGS_C3STOP)
931		cpu_disable_deep_sleep++;
932	if (timer->et_flags & ET_FLAGS_C3STOP)
933		cpu_disable_deep_sleep--;
934	periodic = want_periodic;
935	timer = et;
936	et_init(timer, timercb, NULL, NULL);
937	configtimer(1);
938	ET_UNLOCK();
939	return (error);
940}
941SYSCTL_PROC(_kern_eventtimer, OID_AUTO, timer,
942    CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE,
943    0, 0, sysctl_kern_eventtimer_timer, "A", "Chosen event timer");
944
945/*
946 * Report or change the active event timer periodicity.
947 */
948static int
949sysctl_kern_eventtimer_periodic(SYSCTL_HANDLER_ARGS)
950{
951	int error, val;
952
953	val = periodic;
954	error = sysctl_handle_int(oidp, &val, 0, req);
955	if (error != 0 || req->newptr == NULL)
956		return (error);
957	ET_LOCK();
958	configtimer(0);
959	periodic = want_periodic = val;
960	configtimer(1);
961	ET_UNLOCK();
962	return (error);
963}
964SYSCTL_PROC(_kern_eventtimer, OID_AUTO, periodic,
965    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
966    0, 0, sysctl_kern_eventtimer_periodic, "I", "Enable event timer periodic mode");
967