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