]> git.saurik.com Git - apple/xnu.git/blob - osfmk/kern/sched_prim.c
projects / apple / xnu.git / blob
? search:


60191650c365406ed305e2448ae30bb843331f03
[apple/xnu.git] / osfmk / kern / sched_prim.c
1 /*
2 * Copyright (c) 2000-2009 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /*
29 * @OSF_FREE_COPYRIGHT@
30 */
31 /*
32 * Mach Operating System
33 * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
34 * All Rights Reserved.
35 *
36 * Permission to use, copy, modify and distribute this software and its
37 * documentation is hereby granted, provided that both the copyright
38 * notice and this permission notice appear in all copies of the
39 * software, derivative works or modified versions, and any portions
40 * thereof, and that both notices appear in supporting documentation.
41 *
42 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
43 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
44 * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
45 *
46 * Carnegie Mellon requests users of this software to return to
47 *
48 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
49 * School of Computer Science
50 * Carnegie Mellon University
51 * Pittsburgh PA 15213-3890
52 *
53 * any improvements or extensions that they make and grant Carnegie Mellon
54 * the rights to redistribute these changes.
55 */
56 /*
57 */
58 /*
59 * File: sched_prim.c
60 * Author: Avadis Tevanian, Jr.
61 * Date: 1986
62 *
63 * Scheduling primitives
64 *
65 */
66
67 #include <debug.h>
68 #include <mach_kdb.h>
69
70 #include <ddb/db_output.h>
71
72 #include <mach/mach_types.h>
73 #include <mach/machine.h>
74 #include <mach/policy.h>
75 #include <mach/sync_policy.h>
76
77 #include <machine/machine_routines.h>
78 #include <machine/sched_param.h>
79 #include <machine/machine_cpu.h>
80
81 #include <kern/kern_types.h>
82 #include <kern/clock.h>
83 #include <kern/counters.h>
84 #include <kern/cpu_number.h>
85 #include <kern/cpu_data.h>
86 #include <kern/debug.h>
87 #include <kern/lock.h>
88 #include <kern/macro_help.h>
89 #include <kern/machine.h>
90 #include <kern/misc_protos.h>
91 #include <kern/processor.h>
92 #include <kern/queue.h>
93 #include <kern/sched.h>
94 #include <kern/sched_prim.h>
95 #include <kern/syscall_subr.h>
96 #include <kern/task.h>
97 #include <kern/thread.h>
98 #include <kern/wait_queue.h>
99
100 #include <vm/pmap.h>
101 #include <vm/vm_kern.h>
102 #include <vm/vm_map.h>
103
104 #include <mach/sdt.h>
105
106 #include <sys/kdebug.h>
107
108 #include <kern/pms.h>
109
110 struct run_queue rt_runq;
111 #define RT_RUNQ ((processor_t)-1)
112 decl_simple_lock_data(static,rt_lock);
113
114 #define DEFAULT_PREEMPTION_RATE 100 /* (1/s) */
115 int default_preemption_rate = DEFAULT_PREEMPTION_RATE;
116
117 #define MAX_UNSAFE_QUANTA 800
118 int max_unsafe_quanta = MAX_UNSAFE_QUANTA;
119
120 #define MAX_POLL_QUANTA 2
121 int max_poll_quanta = MAX_POLL_QUANTA;
122
123 #define SCHED_POLL_YIELD_SHIFT 4 /* 1/16 */
124 int sched_poll_yield_shift = SCHED_POLL_YIELD_SHIFT;
125
126 uint64_t max_unsafe_computation;
127 uint32_t sched_safe_duration;
128 uint64_t max_poll_computation;
129
130 uint32_t std_quantum;
131 uint32_t min_std_quantum;
132
133 uint32_t std_quantum_us;
134
135 uint32_t max_rt_quantum;
136 uint32_t min_rt_quantum;
137
138 uint32_t sched_cswtime;
139
140 unsigned sched_tick;
141 uint32_t sched_tick_interval;
142
143 uint32_t sched_pri_shift = INT8_MAX;
144 uint32_t sched_fixed_shift;
145
146 uint32_t sched_run_count, sched_share_count;
147 uint32_t sched_load_average, sched_mach_factor;
148
149 /* Forwards */
150 static void load_shift_init(void) __attribute__((section("__TEXT, initcode")));
151 static void preempt_pri_init(void) __attribute__((section("__TEXT, initcode")));
152
153 static thread_t run_queue_dequeue(
154 run_queue_t runq,
155 integer_t options);
156
157 static thread_t choose_thread(
158 processor_t processor,
159 int priority);
160
161 static thread_t thread_select_idle(
162 thread_t thread,
163 processor_t processor);
164
165 static thread_t processor_idle(
166 thread_t thread,
167 processor_t processor);
168
169 static thread_t steal_thread(
170 processor_set_t pset);
171
172 static thread_t steal_processor_thread(
173 processor_t processor);
174
175 static void thread_update_scan(void);
176
177 #if DEBUG
178 extern int debug_task;
179 #define TLOG(a, fmt, args...) if(debug_task & a) kprintf(fmt, ## args)
180 #else
181 #define TLOG(a, fmt, args...) do {} while (0)
182 #endif
183
184 #if DEBUG
185 static
186 boolean_t thread_runnable(
187 thread_t thread);
188
189 #endif /*DEBUG*/
190
191 /*
192 * State machine
193 *
194 * states are combinations of:
195 * R running
196 * W waiting (or on wait queue)
197 * N non-interruptible
198 * O swapped out
199 * I being swapped in
200 *
201 * init action
202 * assert_wait thread_block clear_wait swapout swapin
203 *
204 * R RW, RWN R; setrun - -
205 * RN RWN RN; setrun - -
206 *
207 * RW W R -
208 * RWN WN RN -
209 *
210 * W R; setrun WO
211 * WN RN; setrun -
212 *
213 * RO - - R
214 *
215 */
216
217 int8_t sched_load_shifts[NRQS];
218 int sched_preempt_pri[NRQBM];
219
220 void
221 sched_init(void)
222 {
223 /*
224 * Calculate the timeslicing quantum
225 * in us.
226 */
227 if (default_preemption_rate < 1)
228 default_preemption_rate = DEFAULT_PREEMPTION_RATE;
229 std_quantum_us = (1000 * 1000) / default_preemption_rate;
230
231 printf("standard timeslicing quantum is %d us\n", std_quantum_us);
232
233 sched_safe_duration = (2 * max_unsafe_quanta / default_preemption_rate) *
234 (1 << SCHED_TICK_SHIFT);
235
236 load_shift_init();
237 preempt_pri_init();
238 simple_lock_init(&rt_lock, 0);
239 run_queue_init(&rt_runq);
240 sched_tick = 0;
241 ast_init();
242 }
243
244 void
245 sched_timebase_init(void)
246 {
247 uint64_t abstime;
248 uint32_t shift;
249
250 /* standard timeslicing quantum */
251 clock_interval_to_absolutetime_interval(
252 std_quantum_us, NSEC_PER_USEC, &abstime);
253 assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
254 std_quantum = (uint32_t)abstime;
255
256 /* smallest remaining quantum (250 us) */
257 clock_interval_to_absolutetime_interval(250, NSEC_PER_USEC, &abstime);
258 assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
259 min_std_quantum = (uint32_t)abstime;
260
261 /* smallest rt computaton (50 us) */
262 clock_interval_to_absolutetime_interval(50, NSEC_PER_USEC, &abstime);
263 assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
264 min_rt_quantum = (uint32_t)abstime;
265
266 /* maximum rt computation (50 ms) */
267 clock_interval_to_absolutetime_interval(
268 50, 1000*NSEC_PER_USEC, &abstime);
269 assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
270 max_rt_quantum = (uint32_t)abstime;
271
272 /* scheduler tick interval */
273 clock_interval_to_absolutetime_interval(USEC_PER_SEC >> SCHED_TICK_SHIFT,
274 NSEC_PER_USEC, &abstime);
275 assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
276 sched_tick_interval = (uint32_t)abstime;
277
278 /*
279 * Compute conversion factor from usage to
280 * timesharing priorities with 5/8 ** n aging.
281 */
282 abstime = (abstime * 5) / 3;
283 for (shift = 0; abstime > BASEPRI_DEFAULT; ++shift)
284 abstime >>= 1;
285 sched_fixed_shift = shift;
286
287 max_unsafe_computation = max_unsafe_quanta * std_quantum;
288 max_poll_computation = max_poll_quanta * std_quantum;
289 }
290
291 /*
292 * Set up values for timeshare
293 * loading factors.
294 */
295 static void
296 load_shift_init(void)
297 {
298 int8_t k, *p = sched_load_shifts;
299 uint32_t i, j;
300
301 *p++ = INT8_MIN; *p++ = 0;
302
303 for (i = j = 2, k = 1; i < NRQS; ++k) {
304 for (j <<= 1; i < j; ++i)
305 *p++ = k;
306 }
307 }
308
309 static void
310 preempt_pri_init(void)
311 {
312 int i, *p = sched_preempt_pri;
313
314 for (i = BASEPRI_FOREGROUND + 1; i < MINPRI_KERNEL; ++i)
315 setbit(i, p);
316
317 for (i = BASEPRI_PREEMPT; i <= MAXPRI; ++i)
318 setbit(i, p);
319 }
320
321 /*
322 * Thread wait timer expiration.
323 */
324 void
325 thread_timer_expire(
326 void *p0,
327 __unused void *p1)
328 {
329 thread_t thread = p0;
330 spl_t s;
331
332 s = splsched();
333 thread_lock(thread);
334 if (--thread->wait_timer_active == 0) {
335 if (thread->wait_timer_is_set) {
336 thread->wait_timer_is_set = FALSE;
337 clear_wait_internal(thread, THREAD_TIMED_OUT);
338 }
339 }
340 thread_unlock(thread);
341 splx(s);
342 }
343
344 #ifndef __LP64__
345
346 /*
347 * thread_set_timer:
348 *
349 * Set a timer for the current thread, if the thread
350 * is ready to wait. Must be called between assert_wait()
351 * and thread_block().
352 */
353 void
354 thread_set_timer(
355 uint32_t interval,
356 uint32_t scale_factor)
357 {
358 thread_t thread = current_thread();
359 uint64_t deadline;
360 spl_t s;
361
362 s = splsched();
363 thread_lock(thread);
364 if ((thread->state & TH_WAIT) != 0) {
365 clock_interval_to_deadline(interval, scale_factor, &deadline);
366 if (!timer_call_enter(&thread->wait_timer, deadline))
367 thread->wait_timer_active++;
368 thread->wait_timer_is_set = TRUE;
369 }
370 thread_unlock(thread);
371 splx(s);
372 }
373
374 void
375 thread_set_timer_deadline(
376 uint64_t deadline)
377 {
378 thread_t thread = current_thread();
379 spl_t s;
380
381 s = splsched();
382 thread_lock(thread);
383 if ((thread->state & TH_WAIT) != 0) {
384 if (!timer_call_enter(&thread->wait_timer, deadline))
385 thread->wait_timer_active++;
386 thread->wait_timer_is_set = TRUE;
387 }
388 thread_unlock(thread);
389 splx(s);
390 }
391
392 void
393 thread_cancel_timer(void)
394 {
395 thread_t thread = current_thread();
396 spl_t s;
397
398 s = splsched();
399 thread_lock(thread);
400 if (thread->wait_timer_is_set) {
401 if (timer_call_cancel(&thread->wait_timer))
402 thread->wait_timer_active--;
403 thread->wait_timer_is_set = FALSE;
404 }
405 thread_unlock(thread);
406 splx(s);
407 }
408
409 #endif /* __LP64__ */
410
411 /*
412 * thread_unblock:
413 *
414 * Unblock thread on wake up.
415 *
416 * Returns TRUE if the thread is still running.
417 *
418 * Thread must be locked.
419 */
420 boolean_t
421 thread_unblock(
422 thread_t thread,
423 wait_result_t wresult)
424 {
425 boolean_t result = FALSE;
426
427 /*
428 * Set wait_result.
429 */
430 thread->wait_result = wresult;
431
432 /*
433 * Cancel pending wait timer.
434 */
435 if (thread->wait_timer_is_set) {
436 if (timer_call_cancel(&thread->wait_timer))
437 thread->wait_timer_active--;
438 thread->wait_timer_is_set = FALSE;
439 }
440
441 /*
442 * Update scheduling state: not waiting,
443 * set running.
444 */
445 thread->state &= ~(TH_WAIT|TH_UNINT);
446
447 if (!(thread->state & TH_RUN)) {
448 thread->state |= TH_RUN;
449
450 (*thread->sched_call)(SCHED_CALL_UNBLOCK, thread);
451
452 /*
453 * Update run counts.
454 */
455 sched_run_incr();
456 if (thread->sched_mode & TH_MODE_TIMESHARE)
457 sched_share_incr();
458 }
459 else {
460 /*
461 * Signal if idling on another processor.
462 */
463 if (thread->state & TH_IDLE) {
464 processor_t processor = thread->last_processor;
465
466 if (processor != current_processor())
467 machine_signal_idle(processor);
468 }
469
470 result = TRUE;
471 }
472
473 /*
474 * Calculate deadline for real-time threads.
475 */
476 if (thread->sched_mode & TH_MODE_REALTIME) {
477 thread->realtime.deadline = mach_absolute_time();
478 thread->realtime.deadline += thread->realtime.constraint;
479 }
480
481 /*
482 * Clear old quantum, fail-safe computation, etc.
483 */
484 thread->current_quantum = 0;
485 thread->computation_metered = 0;
486 thread->reason = AST_NONE;
487
488 KERNEL_DEBUG_CONSTANT(
489 MACHDBG_CODE(DBG_MACH_SCHED,MACH_MAKE_RUNNABLE) | DBG_FUNC_NONE,
490 (uintptr_t)thread_tid(thread), thread->sched_pri, 0, 0, 0);
491
492 DTRACE_SCHED2(wakeup, struct thread *, thread, struct proc *, thread->task->bsd_info);
493
494 return (result);
495 }
496
497 /*
498 * Routine: thread_go
499 * Purpose:
500 * Unblock and dispatch thread.
501 * Conditions:
502 * thread lock held, IPC locks may be held.
503 * thread must have been pulled from wait queue under same lock hold.
504 * Returns:
505 * KERN_SUCCESS - Thread was set running
506 * KERN_NOT_WAITING - Thread was not waiting
507 */
508 kern_return_t
509 thread_go(
510 thread_t thread,
511 wait_result_t wresult)
512 {
513 assert(thread->at_safe_point == FALSE);
514 assert(thread->wait_event == NO_EVENT64);
515 assert(thread->wait_queue == WAIT_QUEUE_NULL);
516
517 if ((thread->state & (TH_WAIT|TH_TERMINATE)) == TH_WAIT) {
518 if (!thread_unblock(thread, wresult))
519 thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
520
521 return (KERN_SUCCESS);
522 }
523
524 return (KERN_NOT_WAITING);
525 }
526
527 /*
528 * Routine: thread_mark_wait_locked
529 * Purpose:
530 * Mark a thread as waiting. If, given the circumstances,
531 * it doesn't want to wait (i.e. already aborted), then
532 * indicate that in the return value.
533 * Conditions:
534 * at splsched() and thread is locked.
535 */
536 __private_extern__
537 wait_result_t
538 thread_mark_wait_locked(
539 thread_t thread,
540 wait_interrupt_t interruptible)
541 {
542 boolean_t at_safe_point;
543
544 assert(thread == current_thread());
545
546 /*
547 * The thread may have certain types of interrupts/aborts masked
548 * off. Even if the wait location says these types of interrupts
549 * are OK, we have to honor mask settings (outer-scoped code may
550 * not be able to handle aborts at the moment).
551 */
552 if (interruptible > (thread->options & TH_OPT_INTMASK))
553 interruptible = thread->options & TH_OPT_INTMASK;
554
555 at_safe_point = (interruptible == THREAD_ABORTSAFE);
556
557 if ( interruptible == THREAD_UNINT ||
558 !(thread->sched_mode & TH_MODE_ABORT) ||
559 (!at_safe_point &&
560 (thread->sched_mode & TH_MODE_ABORTSAFELY))) {
561
562 DTRACE_SCHED(sleep);
563
564 thread->state |= (interruptible) ? TH_WAIT : (TH_WAIT | TH_UNINT);
565 thread->at_safe_point = at_safe_point;
566 return (thread->wait_result = THREAD_WAITING);
567 }
568 else
569 if (thread->sched_mode & TH_MODE_ABORTSAFELY)
570 thread->sched_mode &= ~TH_MODE_ISABORTED;
571
572 return (thread->wait_result = THREAD_INTERRUPTED);
573 }
574
575 /*
576 * Routine: thread_interrupt_level
577 * Purpose:
578 * Set the maximum interruptible state for the
579 * current thread. The effective value of any
580 * interruptible flag passed into assert_wait
581 * will never exceed this.
582 *
583 * Useful for code that must not be interrupted,
584 * but which calls code that doesn't know that.
585 * Returns:
586 * The old interrupt level for the thread.
587 */
588 __private_extern__
589 wait_interrupt_t
590 thread_interrupt_level(
591 wait_interrupt_t new_level)
592 {
593 thread_t thread = current_thread();
594 wait_interrupt_t result = thread->options & TH_OPT_INTMASK;
595
596 thread->options = (thread->options & ~TH_OPT_INTMASK) | (new_level & TH_OPT_INTMASK);
597
598 return result;
599 }
600
601 /*
602 * Check to see if an assert wait is possible, without actually doing one.
603 * This is used by debug code in locks and elsewhere to verify that it is
604 * always OK to block when trying to take a blocking lock (since waiting
605 * for the actual assert_wait to catch the case may make it hard to detect
606 * this case.
607 */
608 boolean_t
609 assert_wait_possible(void)
610 {
611
612 thread_t thread;
613
614 #if DEBUG
615 if(debug_mode) return TRUE; /* Always succeed in debug mode */
616 #endif
617
618 thread = current_thread();
619
620 return (thread == NULL || wait_queue_assert_possible(thread));
621 }
622
623 /*
624 * assert_wait:
625 *
626 * Assert that the current thread is about to go to
627 * sleep until the specified event occurs.
628 */
629 wait_result_t
630 assert_wait(
631 event_t event,
632 wait_interrupt_t interruptible)
633 {
634 register wait_queue_t wq;
635 register int index;
636
637 assert(event != NO_EVENT);
638
639 index = wait_hash(event);
640 wq = &wait_queues[index];
641 return wait_queue_assert_wait(wq, event, interruptible, 0);
642 }
643
644 wait_result_t
645 assert_wait_timeout(
646 event_t event,
647 wait_interrupt_t interruptible,
648 uint32_t interval,
649 uint32_t scale_factor)
650 {
651 thread_t thread = current_thread();
652 wait_result_t wresult;
653 wait_queue_t wqueue;
654 uint64_t deadline;
655 spl_t s;
656
657 assert(event != NO_EVENT);
658 wqueue = &wait_queues[wait_hash(event)];
659
660 s = splsched();
661 wait_queue_lock(wqueue);
662 thread_lock(thread);
663
664 clock_interval_to_deadline(interval, scale_factor, &deadline);
665 wresult = wait_queue_assert_wait64_locked(wqueue, CAST_DOWN(event64_t, event),
666 interruptible, deadline, thread);
667
668 thread_unlock(thread);
669 wait_queue_unlock(wqueue);
670 splx(s);
671
672 return (wresult);
673 }
674
675 wait_result_t
676 assert_wait_deadline(
677 event_t event,
678 wait_interrupt_t interruptible,
679 uint64_t deadline)
680 {
681 thread_t thread = current_thread();
682 wait_result_t wresult;
683 wait_queue_t wqueue;
684 spl_t s;
685
686 assert(event != NO_EVENT);
687 wqueue = &wait_queues[wait_hash(event)];
688
689 s = splsched();
690 wait_queue_lock(wqueue);
691 thread_lock(thread);
692
693 wresult = wait_queue_assert_wait64_locked(wqueue, CAST_DOWN(event64_t,event),
694 interruptible, deadline, thread);
695
696 thread_unlock(thread);
697 wait_queue_unlock(wqueue);
698 splx(s);
699
700 return (wresult);
701 }
702
703 /*
704 * thread_sleep_fast_usimple_lock:
705 *
706 * Cause the current thread to wait until the specified event
707 * occurs. The specified simple_lock is unlocked before releasing
708 * the cpu and re-acquired as part of waking up.
709 *
710 * This is the simple lock sleep interface for components that use a
711 * faster version of simple_lock() than is provided by usimple_lock().
712 */
713 __private_extern__ wait_result_t
714 thread_sleep_fast_usimple_lock(
715 event_t event,
716 simple_lock_t lock,
717 wait_interrupt_t interruptible)
718 {
719 wait_result_t res;
720
721 res = assert_wait(event, interruptible);
722 if (res == THREAD_WAITING) {
723 simple_unlock(lock);
724 res = thread_block(THREAD_CONTINUE_NULL);
725 simple_lock(lock);
726 }
727 return res;
728 }
729
730
731 /*
732 * thread_sleep_usimple_lock:
733 *
734 * Cause the current thread to wait until the specified event
735 * occurs. The specified usimple_lock is unlocked before releasing
736 * the cpu and re-acquired as part of waking up.
737 *
738 * This is the simple lock sleep interface for components where
739 * simple_lock() is defined in terms of usimple_lock().
740 */
741 wait_result_t
742 thread_sleep_usimple_lock(
743 event_t event,
744 usimple_lock_t lock,
745 wait_interrupt_t interruptible)
746 {
747 wait_result_t res;
748
749 res = assert_wait(event, interruptible);
750 if (res == THREAD_WAITING) {
751 usimple_unlock(lock);
752 res = thread_block(THREAD_CONTINUE_NULL);
753 usimple_lock(lock);
754 }
755 return res;
756 }
757
758 /*
759 * thread_sleep_lock_write:
760 *
761 * Cause the current thread to wait until the specified event
762 * occurs. The specified (write) lock is unlocked before releasing
763 * the cpu. The (write) lock will be re-acquired before returning.
764 */
765 wait_result_t
766 thread_sleep_lock_write(
767 event_t event,
768 lock_t *lock,
769 wait_interrupt_t interruptible)
770 {
771 wait_result_t res;
772
773 res = assert_wait(event, interruptible);
774 if (res == THREAD_WAITING) {
775 lock_write_done(lock);
776 res = thread_block(THREAD_CONTINUE_NULL);
777 lock_write(lock);
778 }
779 return res;
780 }
781
782 /*
783 * thread_stop:
784 *
785 * Force a preemption point for a thread and wait
786 * for it to stop running. Arbitrates access among
787 * multiple stop requests. (released by unstop)
788 *
789 * The thread must enter a wait state and stop via a
790 * separate means.
791 *
792 * Returns FALSE if interrupted.
793 */
794 boolean_t
795 thread_stop(
796 thread_t thread)
797 {
798 wait_result_t wresult;
799 spl_t s = splsched();
800
801 wake_lock(thread);
802 thread_lock(thread);
803
804 while (thread->state & TH_SUSP) {
805 thread->wake_active = TRUE;
806 thread_unlock(thread);
807
808 wresult = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
809 wake_unlock(thread);
810 splx(s);
811
812 if (wresult == THREAD_WAITING)
813 wresult = thread_block(THREAD_CONTINUE_NULL);
814
815 if (wresult != THREAD_AWAKENED)
816 return (FALSE);
817
818 s = splsched();
819 wake_lock(thread);
820 thread_lock(thread);
821 }
822
823 thread->state |= TH_SUSP;
824
825 while (thread->state & TH_RUN) {
826 processor_t processor = thread->last_processor;
827
828 if (processor != PROCESSOR_NULL && processor->active_thread == thread)
829 cause_ast_check(processor);
830
831 thread->wake_active = TRUE;
832 thread_unlock(thread);
833
834 wresult = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
835 wake_unlock(thread);
836 splx(s);
837
838 if (wresult == THREAD_WAITING)
839 wresult = thread_block(THREAD_CONTINUE_NULL);
840
841 if (wresult != THREAD_AWAKENED) {
842 thread_unstop(thread);
843 return (FALSE);
844 }
845
846 s = splsched();
847 wake_lock(thread);
848 thread_lock(thread);
849 }
850
851 thread_unlock(thread);
852 wake_unlock(thread);
853 splx(s);
854
855 return (TRUE);
856 }
857
858 /*
859 * thread_unstop:
860 *
861 * Release a previous stop request and set
862 * the thread running if appropriate.
863 *
864 * Use only after a successful stop operation.
865 */
866 void
867 thread_unstop(
868 thread_t thread)
869 {
870 spl_t s = splsched();
871
872 wake_lock(thread);
873 thread_lock(thread);
874
875 if ((thread->state & (TH_RUN|TH_WAIT|TH_SUSP)) == TH_SUSP) {
876 thread->state &= ~TH_SUSP;
877 thread_unblock(thread, THREAD_AWAKENED);
878
879 thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
880 }
881 else
882 if (thread->state & TH_SUSP) {
883 thread->state &= ~TH_SUSP;
884
885 if (thread->wake_active) {
886 thread->wake_active = FALSE;
887 thread_unlock(thread);
888
889 thread_wakeup(&thread->wake_active);
890 wake_unlock(thread);
891 splx(s);
892
893 return;
894 }
895 }
896
897 thread_unlock(thread);
898 wake_unlock(thread);
899 splx(s);
900 }
901
902 /*
903 * thread_wait:
904 *
905 * Wait for a thread to stop running. (non-interruptible)
906 *
907 */
908 void
909 thread_wait(
910 thread_t thread)
911 {
912 wait_result_t wresult;
913 spl_t s = splsched();
914
915 wake_lock(thread);
916 thread_lock(thread);
917
918 while (thread->state & TH_RUN) {
919 processor_t processor = thread->last_processor;
920
921 if (processor != PROCESSOR_NULL && processor->active_thread == thread)
922 cause_ast_check(processor);
923
924 thread->wake_active = TRUE;
925 thread_unlock(thread);
926
927 wresult = assert_wait(&thread->wake_active, THREAD_UNINT);
928 wake_unlock(thread);
929 splx(s);
930
931 if (wresult == THREAD_WAITING)
932 thread_block(THREAD_CONTINUE_NULL);
933
934 s = splsched();
935 wake_lock(thread);
936 thread_lock(thread);
937 }
938
939 thread_unlock(thread);
940 wake_unlock(thread);
941 splx(s);
942 }
943
944 /*
945 * Routine: clear_wait_internal
946 *
947 * Clear the wait condition for the specified thread.
948 * Start the thread executing if that is appropriate.
949 * Arguments:
950 * thread thread to awaken
951 * result Wakeup result the thread should see
952 * Conditions:
953 * At splsched
954 * the thread is locked.
955 * Returns:
956 * KERN_SUCCESS thread was rousted out a wait
957 * KERN_FAILURE thread was waiting but could not be rousted
958 * KERN_NOT_WAITING thread was not waiting
959 */
960 __private_extern__ kern_return_t
961 clear_wait_internal(
962 thread_t thread,
963 wait_result_t wresult)
964 {
965 wait_queue_t wq = thread->wait_queue;
966 int i = LockTimeOut;
967
968 do {
969 if (wresult == THREAD_INTERRUPTED && (thread->state & TH_UNINT))
970 return (KERN_FAILURE);
971
972 if (wq != WAIT_QUEUE_NULL) {
973 if (wait_queue_lock_try(wq)) {
974 wait_queue_pull_thread_locked(wq, thread, TRUE);
975 /* wait queue unlocked, thread still locked */
976 }
977 else {
978 thread_unlock(thread);
979 delay(1);
980
981 thread_lock(thread);
982 if (wq != thread->wait_queue)
983 return (KERN_NOT_WAITING);
984
985 continue;
986 }
987 }
988
989 return (thread_go(thread, wresult));
990 } while (--i > 0);
991
992 panic("clear_wait_internal: deadlock: thread=%p, wq=%p, cpu=%d\n",
993 thread, wq, cpu_number());
994
995 return (KERN_FAILURE);
996 }
997
998
999 /*
1000 * clear_wait:
1001 *
1002 * Clear the wait condition for the specified thread. Start the thread
1003 * executing if that is appropriate.
1004 *
1005 * parameters:
1006 * thread thread to awaken
1007 * result Wakeup result the thread should see
1008 */
1009 kern_return_t
1010 clear_wait(
1011 thread_t thread,
1012 wait_result_t result)
1013 {
1014 kern_return_t ret;
1015 spl_t s;
1016
1017 s = splsched();
1018 thread_lock(thread);
1019 ret = clear_wait_internal(thread, result);
1020 thread_unlock(thread);
1021 splx(s);
1022 return ret;
1023 }
1024
1025
1026 /*
1027 * thread_wakeup_prim:
1028 *
1029 * Common routine for thread_wakeup, thread_wakeup_with_result,
1030 * and thread_wakeup_one.
1031 *
1032 */
1033 kern_return_t
1034 thread_wakeup_prim(
1035 event_t event,
1036 boolean_t one_thread,
1037 wait_result_t result)
1038 {
1039 register wait_queue_t wq;
1040 register int index;
1041
1042 index = wait_hash(event);
1043 wq = &wait_queues[index];
1044 if (one_thread)
1045 return (wait_queue_wakeup_one(wq, event, result));
1046 else
1047 return (wait_queue_wakeup_all(wq, event, result));
1048 }
1049
1050 /*
1051 * thread_bind:
1052 *
1053 * Force the current thread to execute on the specified processor.
1054 *
1055 * Returns the previous binding. PROCESSOR_NULL means
1056 * not bound.
1057 *
1058 * XXX - DO NOT export this to users - XXX
1059 */
1060 processor_t
1061 thread_bind(
1062 processor_t processor)
1063 {
1064 thread_t self = current_thread();
1065 processor_t prev;
1066 spl_t s;
1067
1068 s = splsched();
1069 thread_lock(self);
1070
1071 prev = self->bound_processor;
1072 self->bound_processor = processor;
1073
1074 thread_unlock(self);
1075 splx(s);
1076
1077 return (prev);
1078 }
1079
1080 /*
1081 * thread_select:
1082 *
1083 * Select a new thread for the current processor to execute.
1084 *
1085 * May select the current thread, which must be locked.
1086 */
1087 static thread_t
1088 thread_select(
1089 thread_t thread,
1090 processor_t processor)
1091 {
1092 processor_set_t pset = processor->processor_set;
1093 thread_t new_thread = THREAD_NULL;
1094 boolean_t inactive_state;
1095
1096 do {
1097 /*
1098 * Update the priority.
1099 */
1100 if (thread->sched_stamp != sched_tick)
1101 update_priority(thread);
1102
1103 processor->current_pri = thread->sched_pri;
1104
1105 pset_lock(pset);
1106
1107 inactive_state = processor->state != PROCESSOR_SHUTDOWN && machine_processor_is_inactive(processor);
1108
1109 simple_lock(&rt_lock);
1110
1111 /*
1112 * Test to see if the current thread should continue
1113 * to run on this processor. Must be runnable, and not
1114 * bound to a different processor, nor be in the wrong
1115 * processor set.
1116 */
1117 if ( thread->state == TH_RUN &&
1118 (thread->sched_pri >= BASEPRI_RTQUEUES ||
1119 processor->processor_meta == PROCESSOR_META_NULL ||
1120 processor->processor_meta->primary == processor) &&
1121 (thread->bound_processor == PROCESSOR_NULL ||
1122 thread->bound_processor == processor) &&
1123 (thread->affinity_set == AFFINITY_SET_NULL ||
1124 thread->affinity_set->aset_pset == pset) ) {
1125 if ( thread->sched_pri >= BASEPRI_RTQUEUES &&
1126 first_timeslice(processor) ) {
1127 if (rt_runq.highq >= BASEPRI_RTQUEUES) {
1128 register run_queue_t runq = &rt_runq;
1129 register queue_t q;
1130
1131 q = runq->queues + runq->highq;
1132 if (((thread_t)q->next)->realtime.deadline <
1133 processor->deadline) {
1134 thread = (thread_t)q->next;
1135 ((queue_entry_t)thread)->next->prev = q;
1136 q->next = ((queue_entry_t)thread)->next;
1137 thread->runq = PROCESSOR_NULL;
1138 runq->count--; runq->urgency--;
1139 assert(runq->urgency >= 0);
1140 if (queue_empty(q)) {
1141 if (runq->highq != IDLEPRI)
1142 clrbit(MAXPRI - runq->highq, runq->bitmap);
1143 runq->highq = MAXPRI - ffsbit(runq->bitmap);
1144 }
1145 }
1146 }
1147
1148 simple_unlock(&rt_lock);
1149
1150 processor->deadline = thread->realtime.deadline;
1151
1152 pset_unlock(pset);
1153
1154 return (thread);
1155 }
1156
1157 if (!inactive_state && rt_runq.highq < thread->sched_pri &&
1158 (new_thread = choose_thread(processor, thread->sched_pri)) == THREAD_NULL) {
1159
1160 simple_unlock(&rt_lock);
1161
1162 /* I am the highest priority runnable (non-idle) thread */
1163
1164 pset_pri_hint(pset, processor, processor->current_pri);
1165
1166 pset_count_hint(pset, processor, processor->runq.count);
1167
1168 processor->deadline = UINT64_MAX;
1169
1170 pset_unlock(pset);
1171
1172 return (thread);
1173 }
1174 }
1175
1176 if (new_thread != THREAD_NULL ||
1177 (processor->runq.highq >= rt_runq.highq &&
1178 (new_thread = choose_thread(processor, MINPRI)) != THREAD_NULL)) {
1179 simple_unlock(&rt_lock);
1180
1181 if (!inactive_state) {
1182 pset_pri_hint(pset, processor, new_thread->sched_pri);
1183
1184 pset_count_hint(pset, processor, processor->runq.count);
1185 }
1186
1187 processor->deadline = UINT64_MAX;
1188 pset_unlock(pset);
1189
1190 return (new_thread);
1191 }
1192
1193 if (rt_runq.count > 0) {
1194 thread = run_queue_dequeue(&rt_runq, SCHED_HEADQ);
1195 simple_unlock(&rt_lock);
1196
1197 processor->deadline = thread->realtime.deadline;
1198 pset_unlock(pset);
1199
1200 return (thread);
1201 }
1202
1203 simple_unlock(&rt_lock);
1204
1205 processor->deadline = UINT64_MAX;
1206
1207 /*
1208 * Set processor inactive based on
1209 * indication from the platform code.
1210 */
1211 if (inactive_state) {
1212 if (processor->state == PROCESSOR_RUNNING)
1213 remqueue(&pset->active_queue, (queue_entry_t)processor);
1214 else
1215 if (processor->state == PROCESSOR_IDLE)
1216 remqueue(&pset->idle_queue, (queue_entry_t)processor);
1217
1218 processor->state = PROCESSOR_INACTIVE;
1219
1220 pset_unlock(pset);
1221
1222 return (processor->idle_thread);
1223 }
1224
1225 /*
1226 * No runnable threads, attempt to steal
1227 * from other processors.
1228 */
1229 new_thread = steal_thread(pset);
1230 if (new_thread != THREAD_NULL)
1231 return (new_thread);
1232
1233 /*
1234 * If other threads have appeared, shortcut
1235 * around again.
1236 */
1237 if (processor->runq.count > 0 || rt_runq.count > 0)
1238 continue;
1239
1240 pset_lock(pset);
1241
1242 /*
1243 * Nothing is runnable, so set this processor idle if it
1244 * was running.
1245 */
1246 if (processor->state == PROCESSOR_RUNNING) {
1247 remqueue(&pset->active_queue, (queue_entry_t)processor);
1248 processor->state = PROCESSOR_IDLE;
1249
1250 if (processor->processor_meta == PROCESSOR_META_NULL || processor->processor_meta->primary == processor) {
1251 enqueue_head(&pset->idle_queue, (queue_entry_t)processor);
1252 pset->low_pri = pset->low_count = processor;
1253 }
1254 else {
1255 enqueue_head(&processor->processor_meta->idle_queue, (queue_entry_t)processor);
1256
1257 if (thread->sched_pri < BASEPRI_RTQUEUES) {
1258 pset_unlock(pset);
1259
1260 return (processor->idle_thread);
1261 }
1262 }
1263 }
1264
1265 pset_unlock(pset);
1266
1267 /*
1268 * Choose idle thread if fast idle is not possible.
1269 */
1270 if ((thread->state & (TH_IDLE|TH_TERMINATE|TH_SUSP)) || !(thread->state & TH_WAIT) || thread->wake_active)
1271 return (processor->idle_thread);
1272
1273 /*
1274 * Perform idling activities directly without a
1275 * context switch. Return dispatched thread,
1276 * else check again for a runnable thread.
1277 */
1278 new_thread = thread_select_idle(thread, processor);
1279
1280 } while (new_thread == THREAD_NULL);
1281
1282 return (new_thread);
1283 }
1284
1285 /*
1286 * thread_select_idle:
1287 *
1288 * Idle the processor using the current thread context.
1289 *
1290 * Called with thread locked, then dropped and relocked.
1291 */
1292 static thread_t
1293 thread_select_idle(
1294 thread_t thread,
1295 processor_t processor)
1296 {
1297 thread_t new_thread;
1298
1299 if (thread->sched_mode & TH_MODE_TIMESHARE)
1300 sched_share_decr();
1301 sched_run_decr();
1302
1303 thread->state |= TH_IDLE;
1304 processor->current_pri = IDLEPRI;
1305
1306 thread_unlock(thread);
1307
1308 /*
1309 * Switch execution timing to processor idle thread.
1310 */
1311 processor->last_dispatch = mach_absolute_time();
1312 thread_timer_event(processor->last_dispatch, &processor->idle_thread->system_timer);
1313 PROCESSOR_DATA(processor, kernel_timer) = &processor->idle_thread->system_timer;
1314
1315 /*
1316 * Cancel the quantum timer while idling.
1317 */
1318 timer_call_cancel(&processor->quantum_timer);
1319 processor->timeslice = 0;
1320
1321 (*thread->sched_call)(SCHED_CALL_BLOCK, thread);
1322
1323 /*
1324 * Enable interrupts and perform idling activities. No
1325 * preemption due to TH_IDLE being set.
1326 */
1327 spllo(); new_thread = processor_idle(thread, processor);
1328
1329 /*
1330 * Return at splsched.
1331 */
1332 (*thread->sched_call)(SCHED_CALL_UNBLOCK, thread);
1333
1334 thread_lock(thread);
1335
1336 /*
1337 * If awakened, switch to thread timer and start a new quantum.
1338 * Otherwise skip; we will context switch to another thread or return here.
1339 */
1340 if (!(thread->state & TH_WAIT)) {
1341 processor->last_dispatch = mach_absolute_time();
1342 thread_timer_event(processor->last_dispatch, &thread->system_timer);
1343 PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
1344
1345 thread_quantum_init(thread);
1346
1347 processor->quantum_end = processor->last_dispatch + thread->current_quantum;
1348 timer_call_enter1(&processor->quantum_timer, thread, processor->quantum_end);
1349 processor->timeslice = 1;
1350
1351 thread->computation_epoch = processor->last_dispatch;
1352 }
1353
1354 thread->state &= ~TH_IDLE;
1355
1356 sched_run_incr();
1357 if (thread->sched_mode & TH_MODE_TIMESHARE)
1358 sched_share_incr();
1359
1360 return (new_thread);
1361 }
1362
1363 /*
1364 * choose_thread:
1365 *
1366 * Locate a thread to execute from the processor run queue
1367 * and return it. Only choose a thread with greater or equal
1368 * priority.
1369 *
1370 * Associated pset must be locked. Returns THREAD_NULL
1371 * on failure.
1372 */
1373 static thread_t
1374 choose_thread(
1375 processor_t processor,
1376 int priority)
1377 {
1378 run_queue_t rq = &processor->runq;
1379 queue_t queue = rq->queues + rq->highq;
1380 int pri = rq->highq, count = rq->count;
1381 thread_t thread;
1382
1383 while (count > 0 && pri >= priority) {
1384 thread = (thread_t)queue_first(queue);
1385 while (!queue_end(queue, (queue_entry_t)thread)) {
1386 if (thread->bound_processor == PROCESSOR_NULL ||
1387 thread->bound_processor == processor) {
1388 remqueue(queue, (queue_entry_t)thread);
1389
1390 thread->runq = PROCESSOR_NULL;
1391 rq->count--;
1392 if (testbit(pri, sched_preempt_pri)) {
1393 rq->urgency--; assert(rq->urgency >= 0);
1394 }
1395 if (queue_empty(queue)) {
1396 if (pri != IDLEPRI)
1397 clrbit(MAXPRI - pri, rq->bitmap);
1398 rq->highq = MAXPRI - ffsbit(rq->bitmap);
1399 }
1400
1401 return (thread);
1402 }
1403 count--;
1404
1405 thread = (thread_t)queue_next((queue_entry_t)thread);
1406 }
1407
1408 queue--; pri--;
1409 }
1410
1411 return (THREAD_NULL);
1412 }
1413
1414 /*
1415 * Perform a context switch and start executing the new thread.
1416 *
1417 * Returns FALSE on failure, and the thread is re-dispatched.
1418 *
1419 * Called at splsched.
1420 */
1421
1422 #define funnel_release_check(thread, debug) \
1423 MACRO_BEGIN \
1424 if ((thread)->funnel_state & TH_FN_OWNED) { \
1425 (thread)->funnel_state = TH_FN_REFUNNEL; \
1426 KERNEL_DEBUG(0x603242c | DBG_FUNC_NONE, \
1427 (thread)->funnel_lock, (debug), 0, 0, 0); \
1428 funnel_unlock((thread)->funnel_lock); \
1429 } \
1430 MACRO_END
1431
1432 #define funnel_refunnel_check(thread, debug) \
1433 MACRO_BEGIN \
1434 if ((thread)->funnel_state & TH_FN_REFUNNEL) { \
1435 kern_return_t result = (thread)->wait_result; \
1436 \
1437 (thread)->funnel_state = 0; \
1438 KERNEL_DEBUG(0x6032428 | DBG_FUNC_NONE, \
1439 (thread)->funnel_lock, (debug), 0, 0, 0); \
1440 funnel_lock((thread)->funnel_lock); \
1441 KERNEL_DEBUG(0x6032430 | DBG_FUNC_NONE, \
1442 (thread)->funnel_lock, (debug), 0, 0, 0); \
1443 (thread)->funnel_state = TH_FN_OWNED; \
1444 (thread)->wait_result = result; \
1445 } \
1446 MACRO_END
1447
1448 static boolean_t
1449 thread_invoke(
1450 register thread_t self,
1451 register thread_t thread,
1452 ast_t reason)
1453 {
1454 thread_continue_t continuation = self->continuation;
1455 void *parameter = self->parameter;
1456 processor_t processor;
1457
1458 if (get_preemption_level() != 0) {
1459 int pl = get_preemption_level();
1460 panic("thread_invoke: preemption_level %d, possible cause: %s",
1461 pl, (pl < 0 ? "unlocking an unlocked mutex or spinlock" :
1462 "blocking while holding a spinlock, or within interrupt context"));
1463 }
1464
1465 assert(self == current_thread());
1466
1467 /*
1468 * Mark thread interruptible.
1469 */
1470 thread_lock(thread);
1471 thread->state &= ~TH_UNINT;
1472
1473 #if DEBUG
1474 assert(thread_runnable(thread));
1475 #endif
1476
1477 /*
1478 * Allow time constraint threads to hang onto
1479 * a stack.
1480 */
1481 if ((self->sched_mode & TH_MODE_REALTIME) && !self->reserved_stack)
1482 self->reserved_stack = self->kernel_stack;
1483
1484 if (continuation != NULL) {
1485 if (!thread->kernel_stack) {
1486 /*
1487 * If we are using a privileged stack,
1488 * check to see whether we can exchange it with
1489 * that of the other thread.
1490 */
1491 if (self->kernel_stack == self->reserved_stack && !thread->reserved_stack)
1492 goto need_stack;
1493
1494 /*
1495 * Context switch by performing a stack handoff.
1496 */
1497 continuation = thread->continuation;
1498 parameter = thread->parameter;
1499
1500 processor = current_processor();
1501 processor->active_thread = thread;
1502 processor->current_pri = thread->sched_pri;
1503 if (thread->last_processor != processor && thread->last_processor != NULL) {
1504 if (thread->last_processor->processor_set != processor->processor_set)
1505 thread->ps_switch++;
1506 thread->p_switch++;
1507 }
1508 thread->last_processor = processor;
1509 thread->c_switch++;
1510 ast_context(thread);
1511 thread_unlock(thread);
1512
1513 self->reason = reason;
1514
1515 processor->last_dispatch = mach_absolute_time();
1516 thread_timer_event(processor->last_dispatch, &thread->system_timer);
1517 PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
1518
1519 KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_STACK_HANDOFF)|DBG_FUNC_NONE,
1520 self->reason, (uintptr_t)thread_tid(thread), self->sched_pri, thread->sched_pri, 0);
1521
1522 DTRACE_SCHED2(off__cpu, struct thread *, thread, struct proc *, thread->task->bsd_info);
1523
1524 TLOG(1, "thread_invoke: calling machine_stack_handoff\n");
1525 machine_stack_handoff(self, thread);
1526
1527 DTRACE_SCHED(on__cpu);
1528
1529 thread_dispatch(self, thread);
1530
1531 thread->continuation = thread->parameter = NULL;
1532
1533 counter(c_thread_invoke_hits++);
1534
1535 funnel_refunnel_check(thread, 2);
1536 (void) spllo();
1537
1538 assert(continuation);
1539 call_continuation(continuation, parameter, thread->wait_result);
1540 /*NOTREACHED*/
1541 }
1542 else if (thread == self) {
1543 /* same thread but with continuation */
1544 ast_context(self);
1545 counter(++c_thread_invoke_same);
1546 thread_unlock(self);
1547
1548 self->continuation = self->parameter = NULL;
1549
1550 funnel_refunnel_check(self, 3);
1551 (void) spllo();
1552
1553 call_continuation(continuation, parameter, self->wait_result);
1554 /*NOTREACHED*/
1555 }
1556 }
1557 else {
1558 /*
1559 * Check that the other thread has a stack
1560 */
1561 if (!thread->kernel_stack) {
1562 need_stack:
1563 if (!stack_alloc_try(thread)) {
1564 counter(c_thread_invoke_misses++);
1565 thread_unlock(thread);
1566 thread_stack_enqueue(thread);
1567 return (FALSE);
1568 }
1569 }
1570 else if (thread == self) {
1571 ast_context(self);
1572 counter(++c_thread_invoke_same);
1573 thread_unlock(self);
1574 return (TRUE);
1575 }
1576 }
1577
1578 /*
1579 * Context switch by full context save.
1580 */
1581 processor = current_processor();
1582 processor->active_thread = thread;
1583 processor->current_pri = thread->sched_pri;
1584 if (thread->last_processor != processor && thread->last_processor != NULL) {
1585 if (thread->last_processor->processor_set != processor->processor_set)
1586 thread->ps_switch++;
1587 thread->p_switch++;
1588 }
1589 thread->last_processor = processor;
1590 thread->c_switch++;
1591 ast_context(thread);
1592 thread_unlock(thread);
1593
1594 counter(c_thread_invoke_csw++);
1595
1596 assert(self->runq == PROCESSOR_NULL);
1597 self->reason = reason;
1598
1599 processor->last_dispatch = mach_absolute_time();
1600 thread_timer_event(processor->last_dispatch, &thread->system_timer);
1601 PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
1602
1603 KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED,MACH_SCHED) | DBG_FUNC_NONE,
1604 self->reason, (uintptr_t)thread_tid(thread), self->sched_pri, thread->sched_pri, 0);
1605
1606 DTRACE_SCHED2(off__cpu, struct thread *, thread, struct proc *, thread->task->bsd_info);
1607
1608 /*
1609 * This is where we actually switch register context,
1610 * and address space if required. We will next run
1611 * as a result of a subsequent context switch.
1612 */
1613 thread = machine_switch_context(self, continuation, thread);
1614 TLOG(1,"thread_invoke: returning machine_switch_context: self %p continuation %p thread %p\n", self, continuation, thread);
1615
1616 DTRACE_SCHED(on__cpu);
1617
1618 /*
1619 * We have been resumed and are set to run.
1620 */
1621 thread_dispatch(thread, self);
1622
1623 if (continuation) {
1624 self->continuation = self->parameter = NULL;
1625
1626 funnel_refunnel_check(self, 3);
1627 (void) spllo();
1628
1629 call_continuation(continuation, parameter, self->wait_result);
1630 /*NOTREACHED*/
1631 }
1632
1633 return (TRUE);
1634 }
1635
1636 /*
1637 * thread_dispatch:
1638 *
1639 * Handle threads at context switch. Re-dispatch other thread
1640 * if still running, otherwise update run state and perform
1641 * special actions. Update quantum for other thread and begin
1642 * the quantum for ourselves.
1643 *
1644 * Called at splsched.
1645 */
1646 void
1647 thread_dispatch(
1648 thread_t thread,
1649 thread_t self)
1650 {
1651 processor_t processor = self->last_processor;
1652
1653 if (thread != THREAD_NULL) {
1654 /*
1655 * If blocked at a continuation, discard
1656 * the stack.
1657 */
1658 if (thread->continuation != NULL && thread->kernel_stack != 0)
1659 stack_free(thread);
1660
1661 if (!(thread->state & TH_IDLE)) {
1662 wake_lock(thread);
1663 thread_lock(thread);
1664
1665 /*
1666 * Compute remainder of current quantum.
1667 */
1668 if ( first_timeslice(processor) &&
1669 processor->quantum_end > processor->last_dispatch )
1670 thread->current_quantum = (uint32_t)(processor->quantum_end - processor->last_dispatch);
1671 else
1672 thread->current_quantum = 0;
1673
1674 if (thread->sched_mode & TH_MODE_REALTIME) {
1675 /*
1676 * Cancel the deadline if the thread has
1677 * consumed the entire quantum.
1678 */
1679 if (thread->current_quantum == 0) {
1680 thread->realtime.deadline = UINT64_MAX;
1681 thread->reason |= AST_QUANTUM;
1682 }
1683 } else {
1684 /*
1685 * For non-realtime threads treat a tiny
1686 * remaining quantum as an expired quantum
1687 * but include what's left next time.
1688 */
1689 if (thread->current_quantum < min_std_quantum) {
1690 thread->reason |= AST_QUANTUM;
1691 thread->current_quantum += std_quantum;
1692 }
1693 }
1694
1695 /*
1696 * If we are doing a direct handoff then
1697 * take the remainder of the quantum.
1698 */
1699 if ((thread->reason & (AST_HANDOFF|AST_QUANTUM)) == AST_HANDOFF) {
1700 self->current_quantum = thread->current_quantum;
1701 thread->reason |= AST_QUANTUM;
1702 thread->current_quantum = 0;
1703 }
1704
1705 thread->computation_metered += (processor->last_dispatch - thread->computation_epoch);
1706
1707 if (!(thread->state & TH_WAIT)) {
1708 /*
1709 * Still running.
1710 */
1711 if (thread->reason & AST_QUANTUM)
1712 thread_setrun(thread, SCHED_TAILQ);
1713 else
1714 if (thread->reason & AST_PREEMPT)
1715 thread_setrun(thread, SCHED_HEADQ);
1716 else
1717 thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
1718
1719 thread->reason = AST_NONE;
1720
1721 thread_unlock(thread);
1722 wake_unlock(thread);
1723 }
1724 else {
1725 /*
1726 * Waiting.
1727 */
1728 boolean_t should_terminate = FALSE;
1729
1730 /* Only the first call to thread_dispatch
1731 * after explicit termination should add
1732 * the thread to the termination queue
1733 */
1734 if ((thread->state & (TH_TERMINATE|TH_TERMINATE2)) == TH_TERMINATE) {
1735 should_terminate = TRUE;
1736 thread->state |= TH_TERMINATE2;
1737 }
1738
1739 thread->state &= ~TH_RUN;
1740
1741 if (thread->sched_mode & TH_MODE_TIMESHARE)
1742 sched_share_decr();
1743 sched_run_decr();
1744
1745 (*thread->sched_call)(SCHED_CALL_BLOCK, thread);
1746
1747 if (thread->wake_active) {
1748 thread->wake_active = FALSE;
1749 thread_unlock(thread);
1750
1751 thread_wakeup(&thread->wake_active);
1752 }
1753 else
1754 thread_unlock(thread);
1755
1756 wake_unlock(thread);
1757
1758 if (should_terminate)
1759 thread_terminate_enqueue(thread);
1760 }
1761 }
1762 }
1763
1764 if (!(self->state & TH_IDLE)) {
1765 /*
1766 * Get a new quantum if none remaining.
1767 */
1768 if (self->current_quantum == 0)
1769 thread_quantum_init(self);
1770
1771 /*
1772 * Set up quantum timer and timeslice.
1773 */
1774 processor->quantum_end = (processor->last_dispatch + self->current_quantum);
1775 timer_call_enter1(&processor->quantum_timer, self, processor->quantum_end);
1776
1777 processor->timeslice = 1;
1778
1779 self->computation_epoch = processor->last_dispatch;
1780 }
1781 else {
1782 timer_call_cancel(&processor->quantum_timer);
1783 processor->timeslice = 0;
1784 }
1785 }
1786
1787 #include <libkern/OSDebug.h>
1788
1789 uint32_t kdebug_thread_block = 0;
1790
1791
1792 /*
1793 * thread_block_reason:
1794 *
1795 * Forces a reschedule, blocking the caller if a wait
1796 * has been asserted.
1797 *
1798 * If a continuation is specified, then thread_invoke will
1799 * attempt to discard the thread's kernel stack. When the
1800 * thread resumes, it will execute the continuation function
1801 * on a new kernel stack.
1802 */
1803 counter(mach_counter_t c_thread_block_calls = 0;)
1804
1805 wait_result_t
1806 thread_block_reason(
1807 thread_continue_t continuation,
1808 void *parameter,
1809 ast_t reason)
1810 {
1811 register thread_t self = current_thread();
1812 register processor_t processor;
1813 register thread_t new_thread;
1814 spl_t s;
1815
1816 counter(++c_thread_block_calls);
1817
1818 s = splsched();
1819
1820 if (!(reason & AST_PREEMPT))
1821 funnel_release_check(self, 2);
1822
1823 processor = current_processor();
1824
1825 /* If we're explicitly yielding, force a subsequent quantum */
1826 if (reason & AST_YIELD)
1827 processor->timeslice = 0;
1828
1829 /* We're handling all scheduling AST's */
1830 ast_off(AST_SCHEDULING);
1831
1832 self->continuation = continuation;
1833 self->parameter = parameter;
1834
1835 if (kdebug_thread_block && kdebug_enable && self->state != TH_RUN) {
1836 uint32_t bt[8];
1837
1838 OSBacktrace((void **)&bt[0], 8);
1839
1840 KERNEL_DEBUG_CONSTANT(0x140004c | DBG_FUNC_START, bt[0], bt[1], bt[2], bt[3], 0);
1841 KERNEL_DEBUG_CONSTANT(0x140004c | DBG_FUNC_END, bt[4], bt[5], bt[6], bt[7], 0);
1842 }
1843
1844 do {
1845 thread_lock(self);
1846 new_thread = thread_select(self, processor);
1847 thread_unlock(self);
1848 } while (!thread_invoke(self, new_thread, reason));
1849
1850 funnel_refunnel_check(self, 5);
1851 splx(s);
1852
1853 return (self->wait_result);
1854 }
1855
1856 /*
1857 * thread_block:
1858 *
1859 * Block the current thread if a wait has been asserted.
1860 */
1861 wait_result_t
1862 thread_block(
1863 thread_continue_t continuation)
1864 {
1865 return thread_block_reason(continuation, NULL, AST_NONE);
1866 }
1867
1868 wait_result_t
1869 thread_block_parameter(
1870 thread_continue_t continuation,
1871 void *parameter)
1872 {
1873 return thread_block_reason(continuation, parameter, AST_NONE);
1874 }
1875
1876 /*
1877 * thread_run:
1878 *
1879 * Switch directly from the current thread to the
1880 * new thread, handing off our quantum if appropriate.
1881 *
1882 * New thread must be runnable, and not on a run queue.
1883 *
1884 * Called at splsched.
1885 */
1886 int
1887 thread_run(
1888 thread_t self,
1889 thread_continue_t continuation,
1890 void *parameter,
1891 thread_t new_thread)
1892 {
1893 ast_t handoff = AST_HANDOFF;
1894
1895 funnel_release_check(self, 3);
1896
1897 self->continuation = continuation;
1898 self->parameter = parameter;
1899
1900 while (!thread_invoke(self, new_thread, handoff)) {
1901 processor_t processor = current_processor();
1902
1903 thread_lock(self);
1904 new_thread = thread_select(self, processor);
1905 thread_unlock(self);
1906 handoff = AST_NONE;
1907 }
1908
1909 funnel_refunnel_check(self, 6);
1910
1911 return (self->wait_result);
1912 }
1913
1914 /*
1915 * thread_continue:
1916 *
1917 * Called at splsched when a thread first receives
1918 * a new stack after a continuation.
1919 */
1920 void
1921 thread_continue(
1922 register thread_t thread)
1923 {
1924 register thread_t self = current_thread();
1925 register thread_continue_t continuation;
1926 register void *parameter;
1927
1928 DTRACE_SCHED(on__cpu);
1929
1930 continuation = self->continuation;
1931 parameter = self->parameter;
1932
1933 thread_dispatch(thread, self);
1934
1935 self->continuation = self->parameter = NULL;
1936
1937 funnel_refunnel_check(self, 4);
1938
1939 if (thread != THREAD_NULL)
1940 (void)spllo();
1941
1942 TLOG(1, "thread_continue: calling call_continuation \n");
1943 call_continuation(continuation, parameter, self->wait_result);
1944 /*NOTREACHED*/
1945 }
1946
1947 /*
1948 * run_queue_init:
1949 *
1950 * Initialize a run queue before first use.
1951 */
1952 void
1953 run_queue_init(
1954 run_queue_t rq)
1955 {
1956 int i;
1957
1958 rq->highq = IDLEPRI;
1959 for (i = 0; i < NRQBM; i++)
1960 rq->bitmap[i] = 0;
1961 setbit(MAXPRI - IDLEPRI, rq->bitmap);
1962 rq->urgency = rq->count = 0;
1963 for (i = 0; i < NRQS; i++)
1964 queue_init(&rq->queues[i]);
1965 }
1966
1967 /*
1968 * run_queue_dequeue:
1969 *
1970 * Perform a dequeue operation on a run queue,
1971 * and return the resulting thread.
1972 *
1973 * The run queue must be locked (see run_queue_remove()
1974 * for more info), and not empty.
1975 */
1976 static thread_t
1977 run_queue_dequeue(
1978 run_queue_t rq,
1979 integer_t options)
1980 {
1981 thread_t thread;
1982 queue_t queue = rq->queues + rq->highq;
1983
1984 if (options & SCHED_HEADQ) {
1985 thread = (thread_t)queue->next;
1986 ((queue_entry_t)thread)->next->prev = queue;
1987 queue->next = ((queue_entry_t)thread)->next;
1988 }
1989 else {
1990 thread = (thread_t)queue->prev;
1991 ((queue_entry_t)thread)->prev->next = queue;
1992 queue->prev = ((queue_entry_t)thread)->prev;
1993 }
1994
1995 thread->runq = PROCESSOR_NULL;
1996 rq->count--;
1997 if (testbit(rq->highq, sched_preempt_pri)) {
1998 rq->urgency--; assert(rq->urgency >= 0);
1999 }
2000 if (queue_empty(queue)) {
2001 if (rq->highq != IDLEPRI)
2002 clrbit(MAXPRI - rq->highq, rq->bitmap);
2003 rq->highq = MAXPRI - ffsbit(rq->bitmap);
2004 }
2005
2006 return (thread);
2007 }
2008
2009 /*
2010 * realtime_queue_insert:
2011 *
2012 * Enqueue a thread for realtime execution.
2013 */
2014 static boolean_t
2015 realtime_queue_insert(
2016 thread_t thread)
2017 {
2018 run_queue_t rq = &rt_runq;
2019 queue_t queue = rq->queues + thread->sched_pri;
2020 uint64_t deadline = thread->realtime.deadline;
2021 boolean_t preempt = FALSE;
2022
2023 simple_lock(&rt_lock);
2024
2025 if (queue_empty(queue)) {
2026 enqueue_tail(queue, (queue_entry_t)thread);
2027
2028 setbit(MAXPRI - thread->sched_pri, rq->bitmap);
2029 if (thread->sched_pri > rq->highq)
2030 rq->highq = thread->sched_pri;
2031 preempt = TRUE;
2032 }
2033 else {
2034 register thread_t entry = (thread_t)queue_first(queue);
2035
2036 while (TRUE) {
2037 if ( queue_end(queue, (queue_entry_t)entry) ||
2038 deadline < entry->realtime.deadline ) {
2039 entry = (thread_t)queue_prev((queue_entry_t)entry);
2040 break;
2041 }
2042
2043 entry = (thread_t)queue_next((queue_entry_t)entry);
2044 }
2045
2046 if ((queue_entry_t)entry == queue)
2047 preempt = TRUE;
2048
2049 insque((queue_entry_t)thread, (queue_entry_t)entry);
2050 }
2051
2052 thread->runq = RT_RUNQ;
2053 rq->count++; rq->urgency++;
2054
2055 simple_unlock(&rt_lock);
2056
2057 return (preempt);
2058 }
2059
2060 /*
2061 * realtime_setrun:
2062 *
2063 * Dispatch a thread for realtime execution.
2064 *
2065 * Thread must be locked. Associated pset must
2066 * be locked, and is returned unlocked.
2067 */
2068 static void
2069 realtime_setrun(
2070 processor_t processor,
2071 thread_t thread)
2072 {
2073 processor_set_t pset = processor->processor_set;
2074
2075 /*
2076 * Dispatch directly onto idle processor.
2077 */
2078 if (processor->state == PROCESSOR_IDLE) {
2079 remqueue(&pset->idle_queue, (queue_entry_t)processor);
2080 enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
2081
2082 processor->next_thread = thread;
2083 processor->deadline = thread->realtime.deadline;
2084 processor->state = PROCESSOR_DISPATCHING;
2085 pset_unlock(pset);
2086
2087 if (processor != current_processor())
2088 machine_signal_idle(processor);
2089 return;
2090 }
2091
2092 if (realtime_queue_insert(thread)) {
2093 if (processor == current_processor())
2094 ast_on(AST_PREEMPT | AST_URGENT);
2095 else
2096 cause_ast_check(processor);
2097 }
2098
2099 pset_unlock(pset);
2100 }
2101
2102 /*
2103 * processor_enqueue:
2104 *
2105 * Enqueue thread on a processor run queue. Thread must be locked,
2106 * and not already be on a run queue.
2107 *
2108 * Returns TRUE if a preemption is indicated based on the state
2109 * of the run queue.
2110 *
2111 * The run queue must be locked (see run_queue_remove()
2112 * for more info).
2113 */
2114 static boolean_t
2115 processor_enqueue(
2116 processor_t processor,
2117 thread_t thread,
2118 integer_t options)
2119 {
2120 run_queue_t rq = &processor->runq;
2121 queue_t queue = rq->queues + thread->sched_pri;
2122 boolean_t result = FALSE;
2123
2124 if (queue_empty(queue)) {
2125 enqueue_tail(queue, (queue_entry_t)thread);
2126
2127 setbit(MAXPRI - thread->sched_pri, rq->bitmap);
2128 if (thread->sched_pri > rq->highq) {
2129 rq->highq = thread->sched_pri;
2130 result = TRUE;
2131 }
2132 }
2133 else
2134 if (options & SCHED_TAILQ)
2135 enqueue_tail(queue, (queue_entry_t)thread);
2136 else
2137 enqueue_head(queue, (queue_entry_t)thread);
2138
2139 thread->runq = processor;
2140 if (testbit(thread->sched_pri, sched_preempt_pri))
2141 rq->urgency++;
2142 rq->count++;
2143
2144 return (result);
2145 }
2146
2147 /*
2148 * processor_setrun:
2149 *
2150 * Dispatch a thread for execution on a
2151 * processor.
2152 *
2153 * Thread must be locked. Associated pset must
2154 * be locked, and is returned unlocked.
2155 */
2156 static void
2157 processor_setrun(
2158 processor_t processor,
2159 thread_t thread,
2160 integer_t options)
2161 {
2162 processor_set_t pset = processor->processor_set;
2163 ast_t preempt;
2164
2165 /*
2166 * Dispatch directly onto idle processor.
2167 */
2168 if (processor->state == PROCESSOR_IDLE) {
2169 remqueue(&pset->idle_queue, (queue_entry_t)processor);
2170 enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
2171
2172 processor->next_thread = thread;
2173 processor->deadline = UINT64_MAX;
2174 processor->state = PROCESSOR_DISPATCHING;
2175 pset_unlock(pset);
2176
2177 if (processor != current_processor())
2178 machine_signal_idle(processor);
2179 return;
2180 }
2181
2182 /*
2183 * Set preemption mode.
2184 */
2185 if (testbit(thread->sched_pri, sched_preempt_pri))
2186 preempt = (AST_PREEMPT | AST_URGENT);
2187 else
2188 if (thread->sched_mode & TH_MODE_TIMESHARE && thread->sched_pri < thread->priority)
2189 preempt = AST_NONE;
2190 else
2191 preempt = (options & SCHED_PREEMPT)? AST_PREEMPT: AST_NONE;
2192
2193 if (!processor_enqueue(processor, thread, options))
2194 preempt = AST_NONE;
2195
2196 if (preempt != AST_NONE) {
2197 if (processor == current_processor()) {
2198 if (csw_check(processor) != AST_NONE)
2199 ast_on(preempt);
2200 }
2201 else
2202 if ( (processor->state == PROCESSOR_RUNNING ||
2203 processor->state == PROCESSOR_SHUTDOWN) &&
2204 thread->sched_pri >= processor->current_pri ) {
2205 cause_ast_check(processor);
2206 }
2207 }
2208 else
2209 if ( processor->state == PROCESSOR_SHUTDOWN &&
2210 thread->sched_pri >= processor->current_pri ) {
2211 cause_ast_check(processor);
2212 }
2213
2214 pset_unlock(pset);
2215 }
2216
2217 #define next_pset(p) (((p)->pset_list != PROCESSOR_SET_NULL)? (p)->pset_list: (p)->node->psets)
2218
2219 /*
2220 * choose_next_pset:
2221 *
2222 * Return the next sibling pset containing
2223 * available processors.
2224 *
2225 * Returns the original pset if none other is
2226 * suitable.
2227 */
2228 static processor_set_t
2229 choose_next_pset(
2230 processor_set_t pset)
2231 {
2232 processor_set_t nset = pset;
2233
2234 do {
2235 nset = next_pset(nset);
2236 } while (nset->processor_count < 1 && nset != pset);
2237
2238 return (nset);
2239 }
2240
2241 /*
2242 * choose_processor:
2243 *
2244 * Choose a processor for the thread, beginning at
2245 * the pset. Accepts an optional processor hint in
2246 * the pset.
2247 *
2248 * Returns a processor, possibly from a different pset.
2249 *
2250 * The thread must be locked. The pset must be locked,
2251 * and the resulting pset is locked on return.
2252 */
2253 static processor_t
2254 choose_processor(
2255 processor_set_t pset,
2256 processor_t processor,
2257 thread_t thread)
2258 {
2259 processor_set_t nset, cset = pset;
2260 processor_meta_t pmeta = PROCESSOR_META_NULL;
2261
2262 /*
2263 * Prefer the hinted processor, when appropriate.
2264 */
2265 if (processor != PROCESSOR_NULL) {
2266 processor_t mprocessor;
2267
2268 if (processor->processor_meta != PROCESSOR_META_NULL)
2269 processor = processor->processor_meta->primary;
2270
2271 mprocessor = machine_choose_processor(pset, processor);
2272 if (mprocessor != PROCESSOR_NULL)
2273 processor = mprocessor;
2274
2275 if (processor->processor_set != pset || processor->state == PROCESSOR_INACTIVE ||
2276 processor->state == PROCESSOR_SHUTDOWN || processor->state == PROCESSOR_OFF_LINE)
2277 processor = PROCESSOR_NULL;
2278 else
2279 if (processor->state == PROCESSOR_IDLE)
2280 return (processor);
2281 }
2282 else {
2283 processor = machine_choose_processor(pset, processor);
2284
2285 if (processor != PROCESSOR_NULL) {
2286 if (processor->processor_set != pset || processor->state == PROCESSOR_INACTIVE ||
2287 processor->state == PROCESSOR_SHUTDOWN || processor->state == PROCESSOR_OFF_LINE)
2288 processor = PROCESSOR_NULL;
2289 else
2290 if (processor->state == PROCESSOR_IDLE)
2291 return (processor);
2292 }
2293 }
2294
2295 /*
2296 * Iterate through the processor sets to locate
2297 * an appropriate processor.
2298 */
2299 do {
2300 /*
2301 * Choose an idle processor.
2302 */
2303 if (!queue_empty(&cset->idle_queue))
2304 return ((processor_t)queue_first(&cset->idle_queue));
2305
2306 if (thread->sched_pri >= BASEPRI_RTQUEUES) {
2307 /*
2308 * For an RT thread, iterate through active processors, first fit.
2309 */
2310 processor = (processor_t)queue_first(&cset->active_queue);
2311 while (!queue_end(&cset->active_queue, (queue_entry_t)processor)) {
2312 if (thread->sched_pri > processor->current_pri ||
2313 thread->realtime.deadline < processor->deadline)
2314 return (processor);
2315
2316 if (pmeta == PROCESSOR_META_NULL) {
2317 if (processor->processor_meta != PROCESSOR_META_NULL &&
2318 !queue_empty(&processor->processor_meta->idle_queue))
2319 pmeta = processor->processor_meta;
2320 }
2321
2322 processor = (processor_t)queue_next((queue_entry_t)processor);
2323 }
2324
2325 if (pmeta != PROCESSOR_META_NULL)
2326 return ((processor_t)queue_first(&pmeta->idle_queue));
2327
2328 processor = PROCESSOR_NULL;
2329 }
2330 else {
2331 /*
2332 * Check any hinted processors in the processor set if available.
2333 */
2334 if (cset->low_pri != PROCESSOR_NULL && cset->low_pri->state != PROCESSOR_INACTIVE &&
2335 cset->low_pri->state != PROCESSOR_SHUTDOWN && cset->low_pri->state != PROCESSOR_OFF_LINE &&
2336 (processor == PROCESSOR_NULL ||
2337 (thread->sched_pri > BASEPRI_DEFAULT && cset->low_pri->current_pri < thread->sched_pri))) {
2338 processor = cset->low_pri;
2339 }
2340 else
2341 if (cset->low_count != PROCESSOR_NULL && cset->low_count->state != PROCESSOR_INACTIVE &&
2342 cset->low_count->state != PROCESSOR_SHUTDOWN && cset->low_count->state != PROCESSOR_OFF_LINE &&
2343 (processor == PROCESSOR_NULL || (thread->sched_pri <= BASEPRI_DEFAULT &&
2344 cset->low_count->runq.count < processor->runq.count))) {
2345 processor = cset->low_count;
2346 }
2347
2348 /*
2349 * Otherwise, choose an available processor in the set.
2350 */
2351 if (processor == PROCESSOR_NULL) {
2352 processor = (processor_t)dequeue_head(&cset->active_queue);
2353 if (processor != PROCESSOR_NULL)
2354 enqueue_tail(&cset->active_queue, (queue_entry_t)processor);
2355 }
2356
2357 if (processor != PROCESSOR_NULL && pmeta == PROCESSOR_META_NULL) {
2358 if (processor->processor_meta != PROCESSOR_META_NULL &&
2359 !queue_empty(&processor->processor_meta->idle_queue))
2360 pmeta = processor->processor_meta;
2361 }
2362 }
2363
2364 /*
2365 * Move onto the next processor set.
2366 */
2367 nset = next_pset(cset);
2368
2369 if (nset != pset) {
2370 pset_unlock(cset);
2371
2372 cset = nset;
2373 pset_lock(cset);
2374 }
2375 } while (nset != pset);
2376
2377 /*
2378 * Make sure that we pick a running processor,
2379 * and that the correct processor set is locked.
2380 */
2381 do {
2382 if (pmeta != PROCESSOR_META_NULL) {
2383 if (cset != pmeta->primary->processor_set) {
2384 pset_unlock(cset);
2385
2386 cset = pmeta->primary->processor_set;
2387 pset_lock(cset);
2388 }
2389
2390 if (!queue_empty(&pmeta->idle_queue))
2391 return ((processor_t)queue_first(&pmeta->idle_queue));
2392
2393 pmeta = PROCESSOR_META_NULL;
2394 }
2395
2396 /*
2397 * If we haven't been able to choose a processor,
2398 * pick the boot processor and return it.
2399 */
2400 if (processor == PROCESSOR_NULL) {
2401 processor = master_processor;
2402
2403 /*
2404 * Check that the correct processor set is
2405 * returned locked.
2406 */
2407 if (cset != processor->processor_set) {
2408 pset_unlock(cset);
2409
2410 cset = processor->processor_set;
2411 pset_lock(cset);
2412 }
2413
2414 return (processor);
2415 }
2416
2417 /*
2418 * Check that the processor set for the chosen
2419 * processor is locked.
2420 */
2421 if (cset != processor->processor_set) {
2422 pset_unlock(cset);
2423
2424 cset = processor->processor_set;
2425 pset_lock(cset);
2426 }
2427
2428 /*
2429 * We must verify that the chosen processor is still available.
2430 */
2431 if (processor->state == PROCESSOR_INACTIVE ||
2432 processor->state == PROCESSOR_SHUTDOWN || processor->state == PROCESSOR_OFF_LINE)
2433 processor = PROCESSOR_NULL;
2434 } while (processor == PROCESSOR_NULL);
2435
2436 return (processor);
2437 }
2438
2439 /*
2440 * thread_setrun:
2441 *
2442 * Dispatch thread for execution, onto an idle
2443 * processor or run queue, and signal a preemption
2444 * as appropriate.
2445 *
2446 * Thread must be locked.
2447 */
2448 void
2449 thread_setrun(
2450 thread_t thread,
2451 integer_t options)
2452 {
2453 processor_t processor;
2454 processor_set_t pset;
2455
2456 #if DEBUG
2457 assert(thread_runnable(thread));
2458 #endif
2459
2460 /*
2461 * Update priority if needed.
2462 */
2463 if (thread->sched_stamp != sched_tick)
2464 update_priority(thread);
2465
2466 assert(thread->runq == PROCESSOR_NULL);
2467
2468 if (thread->bound_processor == PROCESSOR_NULL) {
2469 /*
2470 * Unbound case.
2471 */
2472 if (thread->affinity_set != AFFINITY_SET_NULL) {
2473 /*
2474 * Use affinity set policy hint.
2475 */
2476 pset = thread->affinity_set->aset_pset;
2477 pset_lock(pset);
2478
2479 processor = choose_processor(pset, PROCESSOR_NULL, thread);
2480 }
2481 else
2482 if (thread->last_processor != PROCESSOR_NULL) {
2483 /*
2484 * Simple (last processor) affinity case.
2485 */
2486 processor = thread->last_processor;
2487 pset = processor->processor_set;
2488 pset_lock(pset);
2489
2490 /*
2491 * Choose a different processor in certain cases.
2492 */
2493 if (thread->sched_pri >= BASEPRI_RTQUEUES) {
2494 /*
2495 * If the processor is executing an RT thread with
2496 * an earlier deadline, choose another.
2497 */
2498 if (thread->sched_pri <= processor->current_pri ||
2499 thread->realtime.deadline >= processor->deadline)
2500 processor = choose_processor(pset, PROCESSOR_NULL, thread);
2501 }
2502 else
2503 processor = choose_processor(pset, processor, thread);
2504 }
2505 else {
2506 /*
2507 * No Affinity case:
2508 *
2509 * Utilitize a per task hint to spread threads
2510 * among the available processor sets.
2511 */
2512 task_t task = thread->task;
2513
2514 pset = task->pset_hint;
2515 if (pset == PROCESSOR_SET_NULL)
2516 pset = current_processor()->processor_set;
2517
2518 pset = choose_next_pset(pset);
2519 pset_lock(pset);
2520
2521 processor = choose_processor(pset, PROCESSOR_NULL, thread);
2522 task->pset_hint = processor->processor_set;
2523 }
2524 }
2525 else {
2526 /*
2527 * Bound case:
2528 *
2529 * Unconditionally dispatch on the processor.
2530 */
2531 processor = thread->bound_processor;
2532 pset = processor->processor_set;
2533 pset_lock(pset);
2534 }
2535
2536 /*
2537 * Dispatch the thread on the choosen processor.
2538 */
2539 if (thread->sched_pri >= BASEPRI_RTQUEUES)
2540 realtime_setrun(processor, thread);
2541 else
2542 processor_setrun(processor, thread, options);
2543 }
2544
2545 processor_set_t
2546 task_choose_pset(
2547 task_t task)
2548 {
2549 processor_set_t pset = task->pset_hint;
2550
2551 if (pset != PROCESSOR_SET_NULL)
2552 pset = choose_next_pset(pset);
2553
2554 return (pset);
2555 }
2556
2557 /*
2558 * processor_queue_shutdown:
2559 *
2560 * Shutdown a processor run queue by
2561 * re-dispatching non-bound threads.
2562 *
2563 * Associated pset must be locked, and is
2564 * returned unlocked.
2565 */
2566 void
2567 processor_queue_shutdown(
2568 processor_t processor)
2569 {
2570 processor_set_t pset = processor->processor_set;
2571 run_queue_t rq = &processor->runq;
2572 queue_t queue = rq->queues + rq->highq;
2573 int pri = rq->highq, count = rq->count;
2574 thread_t next, thread;
2575 queue_head_t tqueue;
2576
2577 queue_init(&tqueue);
2578
2579 while (count > 0) {
2580 thread = (thread_t)queue_first(queue);
2581 while (!queue_end(queue, (queue_entry_t)thread)) {
2582 next = (thread_t)queue_next((queue_entry_t)thread);
2583
2584 if (thread->bound_processor == PROCESSOR_NULL) {
2585 remqueue(queue, (queue_entry_t)thread);
2586
2587 thread->runq = PROCESSOR_NULL;
2588 rq->count--;
2589 if (testbit(pri, sched_preempt_pri)) {
2590 rq->urgency--; assert(rq->urgency >= 0);
2591 }
2592 if (queue_empty(queue)) {
2593 if (pri != IDLEPRI)
2594 clrbit(MAXPRI - pri, rq->bitmap);
2595 rq->highq = MAXPRI - ffsbit(rq->bitmap);
2596 }
2597
2598 enqueue_tail(&tqueue, (queue_entry_t)thread);
2599 }
2600 count--;
2601
2602 thread = next;
2603 }
2604
2605 queue--; pri--;
2606 }
2607
2608 pset_unlock(pset);
2609
2610 while ((thread = (thread_t)dequeue_head(&tqueue)) != THREAD_NULL) {
2611 thread_lock(thread);
2612
2613 thread_setrun(thread, SCHED_TAILQ);
2614
2615 thread_unlock(thread);
2616 }
2617 }
2618
2619 /*
2620 * Check for a preemption point in
2621 * the current context.
2622 *
2623 * Called at splsched.
2624 */
2625 ast_t
2626 csw_check(
2627 processor_t processor)
2628 {
2629 ast_t result = AST_NONE;
2630 run_queue_t runq;
2631
2632 if (first_timeslice(processor)) {
2633 runq = &rt_runq;
2634 if (runq->highq >= BASEPRI_RTQUEUES)
2635 return (AST_PREEMPT | AST_URGENT);
2636
2637 if (runq->highq > processor->current_pri) {
2638 if (runq->urgency > 0)
2639 return (AST_PREEMPT | AST_URGENT);
2640
2641 result |= AST_PREEMPT;
2642 }
2643
2644 runq = &processor->runq;
2645 if (runq->highq > processor->current_pri) {
2646 if (runq->urgency > 0)
2647 return (AST_PREEMPT | AST_URGENT);
2648
2649 result |= AST_PREEMPT;
2650 }
2651 }
2652 else {
2653 runq = &rt_runq;
2654 if (runq->highq >= processor->current_pri) {
2655 if (runq->urgency > 0)
2656 return (AST_PREEMPT | AST_URGENT);
2657
2658 result |= AST_PREEMPT;
2659 }
2660
2661 runq = &processor->runq;
2662 if (runq->highq >= processor->current_pri) {
2663 if (runq->urgency > 0)
2664 return (AST_PREEMPT | AST_URGENT);
2665
2666 result |= AST_PREEMPT;
2667 }
2668 }
2669
2670 if (result != AST_NONE)
2671 return (result);
2672
2673 if (processor->current_pri < BASEPRI_RTQUEUES && processor->processor_meta != PROCESSOR_META_NULL &&
2674 processor->processor_meta->primary != processor)
2675 return (AST_PREEMPT);
2676
2677 if (machine_processor_is_inactive(processor))
2678 return (AST_PREEMPT);
2679
2680 if (processor->active_thread->state & TH_SUSP)
2681 return (AST_PREEMPT);
2682
2683 return (AST_NONE);
2684 }
2685
2686 /*
2687 * set_sched_pri:
2688 *
2689 * Set the scheduled priority of the specified thread.
2690 *
2691 * This may cause the thread to change queues.
2692 *
2693 * Thread must be locked.
2694 */
2695 void
2696 set_sched_pri(
2697 thread_t thread,
2698 int priority)
2699 {
2700 boolean_t removed = run_queue_remove(thread);
2701
2702 thread->sched_pri = priority;
2703 if (removed)
2704 thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
2705 else
2706 if (thread->state & TH_RUN) {
2707 processor_t processor = thread->last_processor;
2708
2709 if (thread == current_thread()) {
2710 ast_t preempt;
2711
2712 processor->current_pri = priority;
2713 if ((preempt = csw_check(processor)) != AST_NONE)
2714 ast_on(preempt);
2715 }
2716 else
2717 if ( processor != PROCESSOR_NULL &&
2718 processor->active_thread == thread )
2719 cause_ast_check(processor);
2720 }
2721 }
2722
2723 #if 0
2724
2725 static void
2726 run_queue_check(
2727 run_queue_t rq,
2728 thread_t thread)
2729 {
2730 queue_t q;
2731 queue_entry_t qe;
2732
2733 if (rq != thread->runq)
2734 panic("run_queue_check: thread runq");
2735
2736 if (thread->sched_pri > MAXPRI || thread->sched_pri < MINPRI)
2737 panic("run_queue_check: thread sched_pri");
2738
2739 q = &rq->queues[thread->sched_pri];
2740 qe = queue_first(q);
2741 while (!queue_end(q, qe)) {
2742 if (qe == (queue_entry_t)thread)
2743 return;
2744
2745 qe = queue_next(qe);
2746 }
2747
2748 panic("run_queue_check: end");
2749 }
2750
2751 #endif /* DEBUG */
2752
2753 /*
2754 * run_queue_remove:
2755 *
2756 * Remove a thread from a current run queue and
2757 * return TRUE if successful.
2758 *
2759 * Thread must be locked.
2760 */
2761 boolean_t
2762 run_queue_remove(
2763 thread_t thread)
2764 {
2765 processor_t processor = thread->runq;
2766
2767 /*
2768 * If processor is PROCESSOR_NULL, the thread will stay out of the
2769 * run queues because the caller locked the thread. Otherwise
2770 * the thread is on a run queue, but could be chosen for dispatch
2771 * and removed.
2772 */
2773 if (processor != PROCESSOR_NULL) {
2774 void * rqlock;
2775 run_queue_t rq;
2776
2777 /*
2778 * The processor run queues are locked by the
2779 * processor set. Real-time priorities use a
2780 * global queue with a dedicated lock.
2781 */
2782 if (thread->sched_pri < BASEPRI_RTQUEUES) {
2783 rqlock = &processor->processor_set->sched_lock;
2784 rq = &processor->runq;
2785 }
2786 else {
2787 rqlock = &rt_lock; rq = &rt_runq;
2788 }
2789
2790 simple_lock(rqlock);
2791
2792 if (processor == thread->runq) {
2793 /*
2794 * Thread is on a run queue and we have a lock on
2795 * that run queue.
2796 */
2797 remqueue(&rq->queues[0], (queue_entry_t)thread);
2798 rq->count--;
2799 if (testbit(thread->sched_pri, sched_preempt_pri)) {
2800 rq->urgency--; assert(rq->urgency >= 0);
2801 }
2802
2803 if (queue_empty(rq->queues + thread->sched_pri)) {
2804 /* update run queue status */
2805 if (thread->sched_pri != IDLEPRI)
2806 clrbit(MAXPRI - thread->sched_pri, rq->bitmap);
2807 rq->highq = MAXPRI - ffsbit(rq->bitmap);
2808 }
2809
2810 thread->runq = PROCESSOR_NULL;
2811 }
2812 else {
2813 /*
2814 * The thread left the run queue before we could
2815 * lock the run queue.
2816 */
2817 assert(thread->runq == PROCESSOR_NULL);
2818 processor = PROCESSOR_NULL;
2819 }
2820
2821 simple_unlock(rqlock);
2822 }
2823
2824 return (processor != PROCESSOR_NULL);
2825 }
2826
2827 /*
2828 * steal_processor_thread:
2829 *
2830 * Locate a thread to steal from the processor and
2831 * return it.
2832 *
2833 * Associated pset must be locked. Returns THREAD_NULL
2834 * on failure.
2835 */
2836 static thread_t
2837 steal_processor_thread(
2838 processor_t processor)
2839 {
2840 run_queue_t rq = &processor->runq;
2841 queue_t queue = rq->queues + rq->highq;
2842 int pri = rq->highq, count = rq->count;
2843 thread_t thread;
2844
2845 while (count > 0) {
2846 thread = (thread_t)queue_first(queue);
2847 while (!queue_end(queue, (queue_entry_t)thread)) {
2848 if (thread->bound_processor == PROCESSOR_NULL) {
2849 remqueue(queue, (queue_entry_t)thread);
2850
2851 thread->runq = PROCESSOR_NULL;
2852 rq->count--;
2853 if (testbit(pri, sched_preempt_pri)) {
2854 rq->urgency--; assert(rq->urgency >= 0);
2855 }
2856 if (queue_empty(queue)) {
2857 if (pri != IDLEPRI)
2858 clrbit(MAXPRI - pri, rq->bitmap);
2859 rq->highq = MAXPRI - ffsbit(rq->bitmap);
2860 }
2861
2862 return (thread);
2863 }
2864 count--;
2865
2866 thread = (thread_t)queue_next((queue_entry_t)thread);
2867 }
2868
2869 queue--; pri--;
2870 }
2871
2872 return (THREAD_NULL);
2873 }
2874
2875 /*
2876 * Locate and steal a thread, beginning
2877 * at the pset.
2878 *
2879 * The pset must be locked, and is returned
2880 * unlocked.
2881 *
2882 * Returns the stolen thread, or THREAD_NULL on
2883 * failure.
2884 */
2885 static thread_t
2886 steal_thread(
2887 processor_set_t pset)
2888 {
2889 processor_set_t nset, cset = pset;
2890 processor_t processor;
2891 thread_t thread;
2892
2893 do {
2894 processor = (processor_t)queue_first(&cset->active_queue);
2895 while (!queue_end(&cset->active_queue, (queue_entry_t)processor)) {
2896 if (processor->runq.count > 0) {
2897 thread = steal_processor_thread(processor);
2898 if (thread != THREAD_NULL) {
2899 remqueue(&cset->active_queue, (queue_entry_t)processor);
2900 enqueue_tail(&cset->active_queue, (queue_entry_t)processor);
2901
2902 pset_unlock(cset);
2903
2904 return (thread);
2905 }
2906 }
2907
2908 processor = (processor_t)queue_next((queue_entry_t)processor);
2909 }
2910
2911 nset = next_pset(cset);
2912
2913 if (nset != pset) {
2914 pset_unlock(cset);
2915
2916 cset = nset;
2917 pset_lock(cset);
2918 }
2919 } while (nset != pset);
2920
2921 pset_unlock(cset);
2922
2923 return (THREAD_NULL);
2924 }
2925
2926 /*
2927 * This is the processor idle loop, which just looks for other threads
2928 * to execute. Processor idle threads invoke this without supplying a
2929 * current thread to idle without an asserted wait state.
2930 *
2931 * Returns a the next thread to execute if dispatched directly.
2932 */
2933 static thread_t
2934 processor_idle(
2935 thread_t thread,
2936 processor_t processor)
2937 {
2938 processor_set_t pset = processor->processor_set;
2939 thread_t new_thread;
2940 int state;
2941
2942 (void)splsched();
2943
2944 KERNEL_DEBUG_CONSTANT(
2945 MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_START, (uintptr_t)thread_tid(thread), 0, 0, 0, 0);
2946
2947 timer_switch(&PROCESSOR_DATA(processor, system_state),
2948 mach_absolute_time(), &PROCESSOR_DATA(processor, idle_state));
2949 PROCESSOR_DATA(processor, current_state) = &PROCESSOR_DATA(processor, idle_state);
2950
2951 while (processor->next_thread == THREAD_NULL && processor->runq.count == 0 && rt_runq.count == 0 &&
2952 (thread == THREAD_NULL || ((thread->state & (TH_WAIT|TH_SUSP)) == TH_WAIT && !thread->wake_active))) {
2953 machine_idle();
2954
2955 (void)splsched();
2956
2957 if (processor->state == PROCESSOR_INACTIVE && !machine_processor_is_inactive(processor))
2958 break;
2959 }
2960
2961 timer_switch(&PROCESSOR_DATA(processor, idle_state),
2962 mach_absolute_time(), &PROCESSOR_DATA(processor, system_state));
2963 PROCESSOR_DATA(processor, current_state) = &PROCESSOR_DATA(processor, system_state);
2964
2965 pset_lock(pset);
2966
2967 state = processor->state;
2968 if (state == PROCESSOR_DISPATCHING) {
2969 /*
2970 * Commmon case -- cpu dispatched.
2971 */
2972 new_thread = processor->next_thread;
2973 processor->next_thread = THREAD_NULL;
2974 processor->state = PROCESSOR_RUNNING;
2975
2976 if ( processor->runq.highq > new_thread->sched_pri ||
2977 (rt_runq.highq > 0 && rt_runq.highq >= new_thread->sched_pri) ) {
2978 processor->deadline = UINT64_MAX;
2979
2980 pset_unlock(pset);
2981
2982 thread_lock(new_thread);
2983 thread_setrun(new_thread, SCHED_HEADQ);
2984 thread_unlock(new_thread);
2985
2986 KERNEL_DEBUG_CONSTANT(
2987 MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (uintptr_t)thread_tid(thread), state, 0, 0, 0);
2988
2989 return (THREAD_NULL);
2990 }
2991
2992 pset_unlock(pset);
2993
2994 KERNEL_DEBUG_CONSTANT(
2995 MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (uintptr_t)thread_tid(thread), state, (uintptr_t)thread_tid(new_thread), 0, 0);
2996
2997 return (new_thread);
2998 }
2999 else
3000 if (state == PROCESSOR_IDLE) {
3001 remqueue(&pset->idle_queue, (queue_entry_t)processor);
3002
3003 processor->state = PROCESSOR_RUNNING;
3004 enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
3005 }
3006 else
3007 if (state == PROCESSOR_INACTIVE) {
3008 processor->state = PROCESSOR_RUNNING;
3009 enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
3010 }
3011 else
3012 if (state == PROCESSOR_SHUTDOWN) {
3013 /*
3014 * Going off-line. Force a
3015 * reschedule.
3016 */
3017 if ((new_thread = processor->next_thread) != THREAD_NULL) {
3018 processor->next_thread = THREAD_NULL;
3019 processor->deadline = UINT64_MAX;
3020
3021 pset_unlock(pset);
3022
3023 thread_lock(new_thread);
3024 thread_setrun(new_thread, SCHED_HEADQ);
3025 thread_unlock(new_thread);
3026
3027 KERNEL_DEBUG_CONSTANT(
3028 MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (uintptr_t)thread_tid(thread), state, 0, 0, 0);
3029
3030 return (THREAD_NULL);
3031 }
3032 }
3033
3034 pset_unlock(pset);
3035
3036 KERNEL_DEBUG_CONSTANT(
3037 MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (uintptr_t)thread_tid(thread), state, 0, 0, 0);
3038
3039 return (THREAD_NULL);
3040 }
3041
3042 /*
3043 * Each processor has a dedicated thread which
3044 * executes the idle loop when there is no suitable
3045 * previous context.
3046 */
3047 void
3048 idle_thread(void)
3049 {
3050 processor_t processor = current_processor();
3051 thread_t new_thread;
3052
3053 new_thread = processor_idle(THREAD_NULL, processor);
3054 if (new_thread != THREAD_NULL) {
3055 thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
3056 /*NOTREACHED*/
3057 }
3058
3059 thread_block((thread_continue_t)idle_thread);
3060 /*NOTREACHED*/
3061 }
3062
3063 kern_return_t
3064 idle_thread_create(
3065 processor_t processor)
3066 {
3067 kern_return_t result;
3068 thread_t thread;
3069 spl_t s;
3070
3071 result = kernel_thread_create((thread_continue_t)idle_thread, NULL, MAXPRI_KERNEL, &thread);
3072 if (result != KERN_SUCCESS)
3073 return (result);
3074
3075 s = splsched();
3076 thread_lock(thread);
3077 thread->bound_processor = processor;
3078 processor->idle_thread = thread;
3079 thread->sched_pri = thread->priority = IDLEPRI;
3080 thread->state = (TH_RUN | TH_IDLE);
3081 thread_unlock(thread);
3082 splx(s);
3083
3084 thread_deallocate(thread);
3085
3086 return (KERN_SUCCESS);
3087 }
3088
3089 static uint64_t sched_tick_deadline;
3090
3091 /*
3092 * sched_startup:
3093 *
3094 * Kicks off scheduler services.
3095 *
3096 * Called at splsched.
3097 */
3098 void
3099 sched_startup(void)
3100 {
3101 kern_return_t result;
3102 thread_t thread;
3103
3104 result = kernel_thread_start_priority((thread_continue_t)sched_tick_thread, NULL, MAXPRI_KERNEL, &thread);
3105 if (result != KERN_SUCCESS)
3106 panic("sched_startup");
3107
3108 thread_deallocate(thread);
3109
3110 /*
3111 * Yield to the sched_tick_thread while it times
3112 * a series of context switches back. It stores
3113 * the baseline value in sched_cswtime.
3114 *
3115 * The current thread is the only other thread
3116 * active at this point.
3117 */
3118 while (sched_cswtime == 0)
3119 thread_block(THREAD_CONTINUE_NULL);
3120
3121 thread_daemon_init();
3122
3123 thread_call_initialize();
3124 }
3125
3126 /*
3127 * sched_tick_thread:
3128 *
3129 * Perform periodic bookkeeping functions about ten
3130 * times per second.
3131 */
3132 static void
3133 sched_tick_continue(void)
3134 {
3135 uint64_t abstime = mach_absolute_time();
3136
3137 sched_tick++;
3138
3139 /*
3140 * Compute various averages.
3141 */
3142 compute_averages();
3143
3144 /*
3145 * Scan the run queues for threads which
3146 * may need to be updated.
3147 */
3148 thread_update_scan();
3149
3150 clock_deadline_for_periodic_event(sched_tick_interval, abstime,
3151 &sched_tick_deadline);
3152
3153 assert_wait_deadline((event_t)sched_tick_thread, THREAD_UNINT, sched_tick_deadline);
3154 thread_block((thread_continue_t)sched_tick_continue);
3155 /*NOTREACHED*/
3156 }
3157
3158 /*
3159 * Time a series of context switches to determine
3160 * a baseline. Toss the high and low and return
3161 * the one-way value.
3162 */
3163 static uint32_t
3164 time_cswitch(void)
3165 {
3166 uint32_t new, hi, low, accum;
3167 uint64_t abstime;
3168 int i, tries = 7;
3169
3170 accum = hi = low = 0;
3171 for (i = 0; i < tries; ++i) {
3172 abstime = mach_absolute_time();
3173 thread_block(THREAD_CONTINUE_NULL);
3174
3175 new = (uint32_t)(mach_absolute_time() - abstime);
3176
3177 if (i == 0)
3178 accum = hi = low = new;
3179 else {
3180 if (new < low)
3181 low = new;
3182 else
3183 if (new > hi)
3184 hi = new;
3185 accum += new;
3186 }
3187 }
3188
3189 return ((accum - hi - low) / (2 * (tries - 2)));
3190 }
3191
3192 void
3193 sched_tick_thread(void)
3194 {
3195 sched_cswtime = time_cswitch();
3196
3197 sched_tick_deadline = mach_absolute_time();
3198
3199 sched_tick_continue();
3200 /*NOTREACHED*/
3201 }
3202
3203 /*
3204 * thread_update_scan / runq_scan:
3205 *
3206 * Scan the run queues to account for timesharing threads
3207 * which need to be updated.
3208 *
3209 * Scanner runs in two passes. Pass one squirrels likely
3210 * threads away in an array, pass two does the update.
3211 *
3212 * This is necessary because the run queue is locked for
3213 * the candidate scan, but the thread is locked for the update.
3214 *
3215 * Array should be sized to make forward progress, without
3216 * disabling preemption for long periods.
3217 */
3218
3219 #define THREAD_UPDATE_SIZE 128
3220
3221 static thread_t thread_update_array[THREAD_UPDATE_SIZE];
3222 static int thread_update_count = 0;
3223
3224 /*
3225 * Scan a runq for candidate threads.
3226 *
3227 * Returns TRUE if retry is needed.
3228 */
3229 static boolean_t
3230 runq_scan(
3231 run_queue_t runq)
3232 {
3233 register int count;
3234 register queue_t q;
3235 register thread_t thread;
3236
3237 if ((count = runq->count) > 0) {
3238 q = runq->queues + runq->highq;
3239 while (count > 0) {
3240 queue_iterate(q, thread, thread_t, links) {
3241 if ( thread->sched_stamp != sched_tick &&
3242 (thread->sched_mode & TH_MODE_TIMESHARE) ) {
3243 if (thread_update_count == THREAD_UPDATE_SIZE)
3244 return (TRUE);
3245
3246 thread_update_array[thread_update_count++] = thread;
3247 thread_reference_internal(thread);
3248 }
3249
3250 count--;
3251 }
3252
3253 q--;
3254 }
3255 }
3256
3257 return (FALSE);
3258 }
3259
3260 static void
3261 thread_update_scan(void)
3262 {
3263 boolean_t restart_needed = FALSE;
3264 processor_t processor = processor_list;
3265 processor_set_t pset;
3266 thread_t thread;
3267 spl_t s;
3268
3269 do {
3270 do {
3271 pset = processor->processor_set;
3272
3273 s = splsched();
3274 pset_lock(pset);
3275
3276 restart_needed = runq_scan(&processor->runq);
3277
3278 pset_unlock(pset);
3279 splx(s);
3280
3281 if (restart_needed)
3282 break;
3283
3284 thread = processor->idle_thread;
3285 if (thread != THREAD_NULL && thread->sched_stamp != sched_tick) {
3286 if (thread_update_count == THREAD_UPDATE_SIZE) {
3287 restart_needed = TRUE;
3288 break;
3289 }
3290
3291 thread_update_array[thread_update_count++] = thread;
3292 thread_reference_internal(thread);
3293 }
3294 } while ((processor = processor->processor_list) != NULL);
3295
3296 /*
3297 * Ok, we now have a collection of candidates -- fix them.
3298 */
3299 while (thread_update_count > 0) {
3300 thread = thread_update_array[--thread_update_count];
3301 thread_update_array[thread_update_count] = THREAD_NULL;
3302
3303 s = splsched();
3304 thread_lock(thread);
3305 if ( !(thread->state & (TH_WAIT|TH_SUSP)) &&
3306 thread->sched_stamp != sched_tick )
3307 update_priority(thread);
3308 thread_unlock(thread);
3309 splx(s);
3310
3311 thread_deallocate(thread);
3312 }
3313 } while (restart_needed);
3314 }
3315
3316 /*
3317 * Just in case someone doesn't use the macro
3318 */
3319 #undef thread_wakeup
3320 void
3321 thread_wakeup(
3322 event_t x);
3323
3324 void
3325 thread_wakeup(
3326 event_t x)
3327 {
3328 thread_wakeup_with_result(x, THREAD_AWAKENED);
3329 }
3330
3331 boolean_t
3332 preemption_enabled(void)
3333 {
3334 return (get_preemption_level() == 0 && ml_get_interrupts_enabled());
3335 }
3336
3337 #if DEBUG
3338 static boolean_t
3339 thread_runnable(
3340 thread_t thread)
3341 {
3342 return ((thread->state & (TH_RUN|TH_WAIT)) == TH_RUN);
3343 }
3344 #endif /* DEBUG */
3345
3346 #if MACH_KDB
3347 #include <ddb/db_output.h>
3348 #define printf kdbprintf
3349 void db_sched(void);
3350
3351 void
3352 db_sched(void)
3353 {
3354 iprintf("Scheduling Statistics:\n");
3355 db_indent += 2;
3356 iprintf("Thread invocations: csw %d same %d\n",
3357 c_thread_invoke_csw, c_thread_invoke_same);
3358 #if MACH_COUNTERS
3359 iprintf("Thread block: calls %d\n",
3360 c_thread_block_calls);
3361 iprintf("Idle thread:\n\thandoff %d block %d\n",
3362 c_idle_thread_handoff,
3363 c_idle_thread_block);
3364 iprintf("Sched thread blocks: %d\n", c_sched_thread_block);
3365 #endif /* MACH_COUNTERS */
3366 db_indent -= 2;
3367 }
3368
3369 #include <ddb/db_output.h>
3370 void db_show_thread_log(void);
3371
3372 void
3373 db_show_thread_log(void)
3374 {
3375 }
3376 #endif /* MACH_KDB */
mirror of XNU