xref: /xnu-11215.81.4/osfmk/kern/sched_prim.h (revision d4514f0bc1d3f944c22d92e68b646ac3fb40d452)
1 /*
2  * Copyright (c) 2000-2012 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_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  [email protected]
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.h
60  *	Author:	David Golub
61  *
62  *	Scheduling primitive definitions file
63  *
64  */
65 
66 #ifndef _KERN_SCHED_PRIM_H_
67 #define _KERN_SCHED_PRIM_H_
68 
69 #include <sys/cdefs.h>
70 #include <mach/boolean.h>
71 #include <mach/machine/vm_types.h>
72 #include <mach/kern_return.h>
73 #include <kern/clock.h>
74 #include <kern/kern_types.h>
75 #include <kern/percpu.h>
76 #include <kern/thread.h>
77 #include <kern/block_hint.h>
78 
79 extern int              thread_get_current_cpuid(void);
80 
81 #ifdef  MACH_KERNEL_PRIVATE
82 
83 #include <kern/sched_urgency.h>
84 #include <kern/thread_group.h>
85 #include <kern/waitq.h>
86 
87 /* Initialization */
88 extern void             sched_init(void);
89 
90 extern void             sched_startup(void);
91 
92 extern void             sched_timebase_init(void);
93 
94 extern void             pset_rt_init(processor_set_t pset);
95 
96 extern void             sched_rtlocal_init(processor_set_t pset);
97 
98 extern rt_queue_t       sched_rtlocal_runq(processor_set_t pset);
99 
100 extern void             sched_rtlocal_queue_shutdown(processor_t processor);
101 
102 extern int64_t          sched_rtlocal_runq_count_sum(void);
103 
104 extern thread_t         sched_rtlocal_steal_thread(processor_set_t stealing_pset, uint64_t earliest_deadline);
105 
106 extern thread_t         sched_rt_choose_thread(processor_set_t pset);
107 
108 extern void             sched_check_spill(processor_set_t pset, thread_t thread);
109 
110 extern bool             sched_thread_should_yield(processor_t processor, thread_t thread);
111 
112 extern bool             sched_steal_thread_DISABLED(processor_set_t pset);
113 extern bool             sched_steal_thread_enabled(processor_set_t pset);
114 
115 /* Force a preemption point for a thread and wait for it to stop running */
116 extern boolean_t        thread_stop(
117 	thread_t        thread,
118 	boolean_t       until_not_runnable);
119 
120 /* Release a previous stop request */
121 extern void                     thread_unstop(
122 	thread_t        thread);
123 
124 /* Wait for a thread to stop running */
125 extern void                     thread_wait(
126 	thread_t        thread,
127 	boolean_t       until_not_runnable);
128 
129 /* Unblock thread on wake up */
130 extern boolean_t        thread_unblock(
131 	thread_t                thread,
132 	wait_result_t   wresult);
133 
134 /* Unblock and dispatch thread */
135 extern void thread_go(
136 	thread_t                thread,
137 	wait_result_t           wresult,
138 	bool                    try_handoff);
139 
140 /* Check if direct handoff is allowed */
141 extern boolean_t
142 thread_allowed_for_handoff(
143 	thread_t         thread);
144 
145 /* Handle threads at context switch */
146 extern void                     thread_dispatch(
147 	thread_t                old_thread,
148 	thread_t                new_thread);
149 
150 /* Switch directly to a particular thread */
151 extern int                      thread_run(
152 	thread_t                        self,
153 	thread_continue_t       continuation,
154 	void                            *parameter,
155 	thread_t                        new_thread);
156 
157 /* Resume thread with new stack */
158 extern __dead2 void     thread_continue(thread_t old_thread);
159 
160 /* Invoke continuation */
161 extern __dead2 void     call_continuation(
162 	thread_continue_t       continuation,
163 	void                    *parameter,
164 	wait_result_t           wresult,
165 	boolean_t               enable_interrupts);
166 
167 /*
168  * Flags that can be passed to set_sched_pri
169  * to skip side effects
170  */
171 __options_decl(set_sched_pri_options_t, uint32_t, {
172 	SETPRI_DEFAULT  = 0x0,
173 	SETPRI_LAZY     = 0x1,  /* Avoid setting AST flags or sending IPIs */
174 });
175 
176 /* Set the current scheduled priority */
177 extern void set_sched_pri(
178 	thread_t      thread,
179 	int16_t       priority,
180 	set_sched_pri_options_t options);
181 
182 /* Set base priority of the specified thread */
183 extern void             sched_set_thread_base_priority(
184 	thread_t                thread,
185 	int                             priority);
186 
187 /* Set absolute base priority of the specified thread */
188 extern void             sched_set_kernel_thread_priority(
189 	thread_t                thread,
190 	int                             priority);
191 
192 
193 /* Set the thread's true scheduling mode */
194 extern void             sched_set_thread_mode(thread_t thread,
195     sched_mode_t mode);
196 
197 /*
198  * Set the thread's scheduling mode taking into account that the thread may have
199  * been demoted.
200  * */
201 extern void             sched_set_thread_mode_user(thread_t thread,
202     sched_mode_t mode);
203 
204 /*
205  * Get the thread's scheduling mode taking into account that the thread may have
206  * been demoted.
207  * */
208 extern sched_mode_t     sched_get_thread_mode_user(thread_t thread);
209 
210 
211 /* Demote the true scheduler mode */
212 extern void             sched_thread_mode_demote(thread_t thread,
213     uint32_t reason);
214 /* Un-demote the true scheduler mode */
215 extern void             sched_thread_mode_undemote(thread_t thread,
216     uint32_t reason);
217 /* Check for a specific demotion */
218 extern bool             sched_thread_mode_has_demotion(thread_t thread,
219     uint32_t reason);
220 
221 extern void sched_thread_promote_reason(thread_t thread, uint32_t reason, uintptr_t trace_obj);
222 extern void sched_thread_unpromote_reason(thread_t thread, uint32_t reason, uintptr_t trace_obj);
223 
224 /* Re-evaluate base priority of thread (thread locked) */
225 void thread_recompute_priority(thread_t thread);
226 
227 /* Re-evaluate scheduled priority of thread (thread locked) */
228 extern void thread_recompute_sched_pri(
229 	thread_t thread,
230 	set_sched_pri_options_t options);
231 
232 /* Periodic scheduler activity */
233 extern void             sched_init_thread(void);
234 
235 /* Perform sched_tick housekeeping activities */
236 extern boolean_t                can_update_priority(
237 	thread_t                thread);
238 
239 extern void             update_priority(
240 	thread_t                thread);
241 
242 extern void             lightweight_update_priority(
243 	thread_t                thread);
244 
245 extern void             sched_default_quantum_expire(thread_t thread);
246 
247 /* Idle processor thread continuation */
248 extern void             idle_thread(
249 	void*           parameter,
250 	wait_result_t   result);
251 
252 extern void idle_thread_create(
253 	processor_t             processor,
254 	thread_continue_t       continuation);
255 
256 /* Continuation return from syscall */
257 extern void     thread_syscall_return(
258 	kern_return_t   ret);
259 
260 /* Context switch */
261 extern wait_result_t    thread_block_reason(
262 	thread_continue_t       continuation,
263 	void                            *parameter,
264 	ast_t                           reason);
265 
266 __options_decl(sched_options_t, uint32_t, {
267 	SCHED_NONE      = 0x0,
268 	SCHED_TAILQ     = 0x1,
269 	SCHED_HEADQ     = 0x2,
270 	SCHED_PREEMPT   = 0x4,
271 	SCHED_REBALANCE = 0x8,
272 });
273 
274 /* Reschedule thread for execution */
275 extern void             thread_setrun(
276 	thread_t        thread,
277 	sched_options_t options);
278 
279 extern processor_set_t  task_choose_pset(
280 	task_t                  task);
281 
282 /* Bind the current thread to a particular processor */
283 extern processor_t      thread_bind(
284 	processor_t             processor);
285 
286 extern void             thread_bind_during_wakeup(
287 	thread_t                thread,
288 	processor_t             processor);
289 
290 extern void             thread_unbind_after_queue_shutdown(
291 	thread_t                thread,
292 	processor_t             processor);
293 
294 extern bool pset_has_stealable_threads(
295 	processor_set_t         pset);
296 
297 extern bool pset_has_stealable_rt_threads(
298 	processor_set_t         pset);
299 
300 extern processor_set_t choose_starting_pset(
301 	pset_node_t  node,
302 	thread_t     thread,
303 	processor_t *processor_hint);
304 
305 extern int pset_available_cpu_count(
306 	processor_set_t pset);
307 
308 extern bool pset_is_recommended(
309 	processor_set_t pset);
310 
311 extern bool pset_type_is_recommended(
312 	processor_set_t pset);
313 
314 extern pset_node_t sched_choose_node(
315 	thread_t     thread);
316 
317 /* Choose the best processor to run a thread */
318 extern processor_t      choose_processor(
319 	processor_set_t                pset,
320 	processor_t                    processor,
321 	thread_t                       thread);
322 
323 extern bool sched_SMT_balance(
324 	processor_t processor,
325 	processor_set_t pset);
326 
327 extern void thread_quantum_init(
328 	thread_t thread,
329 	uint64_t now);
330 
331 
332 extern void             run_queue_init(
333 	run_queue_t             runq);
334 
335 extern thread_t run_queue_dequeue(
336 	run_queue_t           runq,
337 	sched_options_t       options);
338 
339 extern boolean_t        run_queue_enqueue(
340 	run_queue_t           runq,
341 	thread_t              thread,
342 	sched_options_t       options);
343 
344 extern void     run_queue_remove(
345 	run_queue_t            runq,
346 	thread_t                       thread);
347 
348 extern thread_t run_queue_peek(
349 	run_queue_t            runq);
350 
351 struct sched_update_scan_context {
352 	uint64_t        earliest_bg_make_runnable_time;
353 	uint64_t        earliest_normal_make_runnable_time;
354 	uint64_t        earliest_rt_make_runnable_time;
355 	uint64_t        sched_tick_last_abstime;
356 };
357 typedef struct sched_update_scan_context *sched_update_scan_context_t;
358 
359 extern void             sched_rtlocal_runq_scan(sched_update_scan_context_t scan_context);
360 
361 extern void sched_pset_made_schedulable(
362 	processor_t processor,
363 	processor_set_t pset,
364 	boolean_t drop_lock);
365 
366 extern void sched_cpu_init_completed(void);
367 
368 /*
369  * Enum to define various events which need IPIs. The IPI policy
370  * engine decides what kind of IPI to use based on destination
371  * processor state, thread and one of the following scheduling events.
372  */
373 typedef enum {
374 	SCHED_IPI_EVENT_BOUND_THR   = 0x1,
375 	SCHED_IPI_EVENT_PREEMPT     = 0x2,
376 	SCHED_IPI_EVENT_SMT_REBAL   = 0x3,
377 	SCHED_IPI_EVENT_SPILL       = 0x4,
378 	SCHED_IPI_EVENT_REBALANCE   = 0x5,
379 	SCHED_IPI_EVENT_RT_PREEMPT  = 0x6,
380 } sched_ipi_event_t;
381 
382 
383 /* Enum to define various IPI types used by the scheduler */
384 typedef enum {
385 	SCHED_IPI_NONE              = 0x0,
386 	SCHED_IPI_IMMEDIATE         = 0x1,
387 	SCHED_IPI_IDLE              = 0x2,
388 	SCHED_IPI_DEFERRED          = 0x3,
389 } sched_ipi_type_t;
390 
391 /* The IPI policy engine behaves in the following manner:
392  * - All scheduler events which need an IPI invoke sched_ipi_action() with
393  *   the appropriate destination processor, thread and event.
394  * - sched_ipi_action() performs basic checks, invokes the scheduler specific
395  *   ipi_policy routine and sets pending_AST bits based on the result.
396  * - Once the pset lock is dropped, the scheduler invokes sched_ipi_perform()
397  *   routine which actually sends the appropriate IPI to the destination core.
398  */
399 extern sched_ipi_type_t sched_ipi_action(processor_t dst, thread_t thread, sched_ipi_event_t event);
400 extern void sched_ipi_perform(processor_t dst, sched_ipi_type_t ipi);
401 
402 /* sched_ipi_policy() is the global default IPI policy for all schedulers */
403 extern sched_ipi_type_t sched_ipi_policy(processor_t dst, thread_t thread,
404     boolean_t dst_idle, sched_ipi_event_t event);
405 
406 /* sched_ipi_deferred_policy() is the global default deferred IPI policy for all schedulers */
407 extern sched_ipi_type_t sched_ipi_deferred_policy(processor_set_t pset,
408     processor_t dst, thread_t thread, sched_ipi_event_t event);
409 
410 #if defined(CONFIG_SCHED_TIMESHARE_CORE)
411 
412 extern boolean_t        thread_update_add_thread(thread_t thread);
413 extern void             thread_update_process_threads(void);
414 extern boolean_t        runq_scan(run_queue_t runq, sched_update_scan_context_t scan_context);
415 
416 #if CONFIG_SCHED_CLUTCH
417 extern boolean_t        sched_clutch_timeshare_scan(queue_t thread_queue, uint16_t count, sched_update_scan_context_t scan_context);
418 #endif /* CONFIG_SCHED_CLUTCH */
419 
420 extern void sched_timeshare_init(void);
421 extern void sched_timeshare_timebase_init(void);
422 extern void sched_timeshare_maintenance_continue(void);
423 
424 extern boolean_t priority_is_urgent(int priority);
425 extern uint32_t sched_timeshare_initial_quantum_size(thread_t thread);
426 
427 extern int sched_compute_timeshare_priority(thread_t thread);
428 
429 #endif /* CONFIG_SCHED_TIMESHARE_CORE */
430 
431 /* Remove thread from its run queue */
432 extern boolean_t        thread_run_queue_remove(thread_t thread);
433 thread_t thread_run_queue_remove_for_handoff(thread_t thread);
434 
435 /* Put a thread back in the run queue after being yanked */
436 extern void thread_run_queue_reinsert(thread_t thread, sched_options_t options);
437 
438 extern void             thread_timer_expire(
439 	void                    *thread,
440 	void                    *p1);
441 
442 extern bool thread_is_eager_preempt(thread_t thread);
443 
444 extern boolean_t sched_generic_direct_dispatch_to_idle_processors;
445 
446 /* Set the maximum interrupt level for the thread */
447 __private_extern__ wait_interrupt_t thread_interrupt_level(
448 	wait_interrupt_t interruptible);
449 
450 __private_extern__ wait_result_t thread_mark_wait_locked(
451 	thread_t                 thread,
452 	wait_interrupt_t interruptible);
453 
454 /* Wake up locked thread directly, passing result */
455 __private_extern__ kern_return_t clear_wait_internal(
456 	thread_t                thread,
457 	wait_result_t   result);
458 
459 struct sched_statistics {
460 	uint32_t        csw_count;
461 	uint32_t        preempt_count;
462 	uint32_t        preempted_rt_count;
463 	uint32_t        preempted_by_rt_count;
464 	uint32_t        rt_sched_count;
465 	uint32_t        interrupt_count;
466 	uint32_t        ipi_count;
467 	uint32_t        timer_pop_count;
468 	uint32_t        idle_transitions;
469 	uint32_t        quantum_timer_expirations;
470 };
471 PERCPU_DECL(struct sched_statistics, sched_stats);
472 extern bool             sched_stats_active;
473 
474 extern void sched_stats_handle_csw(
475 	processor_t processor,
476 	int reasons,
477 	int selfpri,
478 	int otherpri);
479 
480 extern void sched_stats_handle_runq_change(
481 	struct runq_stats *stats,
482 	int old_count);
483 
484 #define SCHED_STATS_INC(field)                                                  \
485 MACRO_BEGIN                                                                     \
486 	if (__improbable(sched_stats_active)) {                                 \
487 	        PERCPU_GET(sched_stats)->field++;                               \
488 	}                                                                       \
489 MACRO_END
490 
491 #if DEBUG
492 
493 #define SCHED_STATS_CSW(processor, reasons, selfpri, otherpri)                  \
494 MACRO_BEGIN                                                                     \
495 	if (__improbable(sched_stats_active)) {                                 \
496 	        sched_stats_handle_csw((processor),                             \
497 	            (reasons), (selfpri), (otherpri));                          \
498 	}                                                                       \
499 MACRO_END
500 
501 
502 #define SCHED_STATS_RUNQ_CHANGE(stats, old_count)                               \
503 MACRO_BEGIN                                                                     \
504 	if (__improbable(sched_stats_active)) {                                 \
505 	        sched_stats_handle_runq_change((stats), (old_count));           \
506 	}                                                                       \
507 MACRO_END
508 
509 #else /* DEBUG */
510 
511 #define SCHED_STATS_CSW(processor, reasons, selfpri, otherpri) do { }while(0)
512 #define SCHED_STATS_RUNQ_CHANGE(stats, old_count) do { }while(0)
513 
514 #endif /* DEBUG */
515 
516 extern uint32_t sched_debug_flags;
517 #define SCHED_DEBUG_FLAG_PLATFORM_TRACEPOINTS           0x00000001
518 #define SCHED_DEBUG_FLAG_CHOOSE_PROCESSOR_TRACEPOINTS   0x00000002
519 #define SCHED_DEBUG_FLAG_AST_CHECK_TRACEPOINTS   0x00000004
520 
521 #define SCHED_DEBUG_PLATFORM_KERNEL_DEBUG_CONSTANT(...)                         \
522 MACRO_BEGIN                                                                     \
523 	if (__improbable(sched_debug_flags &                                    \
524 	    SCHED_DEBUG_FLAG_PLATFORM_TRACEPOINTS)) {                           \
525 	        KERNEL_DEBUG_CONSTANT(__VA_ARGS__);                             \
526 	}                                                                       \
527 MACRO_END
528 
529 #define SCHED_DEBUG_CHOOSE_PROCESSOR_KERNEL_DEBUG_CONSTANT_IST(...)             \
530 MACRO_BEGIN                                                                     \
531 	if (__improbable(sched_debug_flags &                                    \
532 	    SCHED_DEBUG_FLAG_CHOOSE_PROCESSOR_TRACEPOINTS)) {                   \
533 	        KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, __VA_ARGS__);           \
534 	}                                                                       \
535 MACRO_END
536 
537 #define SCHED_DEBUG_AST_CHECK_KDBG_RELEASE(...)                                 \
538 MACRO_BEGIN                                                                     \
539 	if (__improbable(sched_debug_flags &                                    \
540 	    SCHED_DEBUG_FLAG_AST_CHECK_TRACEPOINTS)) {                          \
541 	        KDBG_RELEASE(__VA_ARGS__);                                      \
542 	}                                                                       \
543 MACRO_END
544 
545 
546 /* Tells if there are "active" RT threads in the system (provided by CPU PM) */
547 extern void     active_rt_threads(
548 	boolean_t       active);
549 
550 /* Returns the perfcontrol attribute for the thread */
551 extern perfcontrol_class_t thread_get_perfcontrol_class(
552 	thread_t        thread);
553 
554 /* Generic routine for Non-AMP schedulers to calculate parallelism */
555 extern uint32_t sched_qos_max_parallelism(int qos, uint64_t options);
556 
557 extern void check_monotonic_time(uint64_t ctime);
558 
559 #endif /* MACH_KERNEL_PRIVATE */
560 
561 __BEGIN_DECLS
562 
563 #ifdef  XNU_KERNEL_PRIVATE
564 
565 extern void thread_bind_cluster_type(thread_t, char cluster_type, bool soft_bind);
566 
567 __options_decl(thread_bind_option_t, uint64_t, {
568 	/* Unbind a previously cluster bound thread */
569 	THREAD_UNBIND                   = 0x1,
570 	/*
571 	 * Soft bind the thread to the cluster; soft binding means the thread will be
572 	 * moved to an available cluster if the bound cluster is de-recommended/offline.
573 	 */
574 	THREAD_BIND_SOFT                = 0x2,
575 	/*
576 	 * Bind thread to the cluster only if it is eligible to run on that cluster. If
577 	 * the thread is not eligible to run on the cluster, thread_bind_cluster_id()
578 	 * returns KERN_INVALID_POLICY.
579 	 */
580 	THREAD_BIND_ELIGIBLE_ONLY       = 0x4,
581 });
582 extern kern_return_t thread_bind_cluster_id(thread_t thread, uint32_t cluster_id, thread_bind_option_t options);
583 
584 extern int sched_get_rt_n_backup_processors(void);
585 extern void sched_set_rt_n_backup_processors(int n);
586 
587 extern int sched_get_rt_deadline_epsilon(void);
588 extern void sched_set_rt_deadline_epsilon(int new_epsilon_us);
589 
590 /* Toggles a global override to turn off CPU Throttling */
591 extern void     sys_override_cpu_throttle(boolean_t enable_override);
592 
593 extern int sched_get_powered_cores(void);
594 extern void sched_set_powered_cores(int n);
595 
596 uint64_t sched_sysctl_get_recommended_cores(void);
597 
598 /*
599  ****************** Only exported until BSD stops using ********************
600  */
601 
602 extern void                     thread_vm_bind_group_add(void);
603 
604 /* Wake up thread directly, passing result */
605 extern kern_return_t clear_wait(
606 	thread_t                thread,
607 	wait_result_t   result);
608 
609 /* Start thread running */
610 extern void             thread_bootstrap_return(void) __attribute__((noreturn));
611 
612 /* Return from exception (BSD-visible interface) */
613 extern void             thread_exception_return(void) __dead2;
614 
615 #define SCHED_STRING_MAX_LENGTH (48)
616 /* String declaring the name of the current scheduler */
617 extern char sched_string[SCHED_STRING_MAX_LENGTH];
618 
619 __options_decl(thread_handoff_option_t, uint32_t, {
620 	THREAD_HANDOFF_NONE          = 0,
621 	THREAD_HANDOFF_SETRUN_NEEDED = 0x1,
622 });
623 
624 /* Remove thread from its run queue */
625 thread_t thread_prepare_for_handoff(thread_t thread, thread_handoff_option_t option);
626 
627 /* Attempt to context switch to a specific runnable thread */
628 extern wait_result_t thread_handoff_deallocate(thread_t thread, thread_handoff_option_t option);
629 
630 __attribute__((nonnull(2)))
631 extern void thread_handoff_parameter(thread_t thread,
632     thread_continue_t continuation, void *parameter, thread_handoff_option_t) __dead2;
633 
634 extern struct waitq     *assert_wait_queue(event_t event);
635 
636 extern kern_return_t thread_wakeup_one_with_pri(event_t event, int priority);
637 
638 extern thread_t thread_wakeup_identify(event_t event, int priority);
639 
640 /*
641  * sched_cond_t:
642  *
643  * A atomic condition variable used to synchronize wake/block operations on threads.
644  * Bits defined below are reserved for use by sched_prim. Remaining
645  * bits may be used by caller for additional synchronization semantics.
646  */
647 __options_decl(sched_cond_t, uint32_t, {
648 	SCHED_COND_INIT = 0x0000,    /* initialize all bits to zero (inactive and not awoken) */
649 	SCHED_COND_ACTIVE = 0x0001,  /* target thread is active */
650 	SCHED_COND_WAKEUP = 0x0002   /* wakeup has been issued for target thread */
651 });
652 typedef _Atomic sched_cond_t sched_cond_atomic_t;
653 
654 /*
655  * sched_cond_init:
656  *
657  * Initialize an atomic condition variable. Note that this does not occur atomically and should be
658  * performed during thread initialization, before the condition is observable by other threads.
659  */
660 extern void sched_cond_init(
661 	sched_cond_atomic_t *cond);
662 
663 /*
664  * sched_cond_signal:
665  *
666  * Wakeup the specified thread if it is waiting on this event and it has not already been issued a wakeup.
667  *
668  * parameters:
669  *      thread    thread to awaken
670  *      cond      atomic condition variable
671  */
672 extern kern_return_t sched_cond_signal(
673 	sched_cond_atomic_t *cond,
674 	thread_t thread);
675 
676 /*
677  * sched_cond_wait_parameter:
678  *
679  * Assert wait and block on cond if no wakeup has been issued.
680  * If a wakeup has been issued on cond since the last `sched_cond_ack`, clear_wait and
681  * return `THREAD_AWAKENED`.
682  *
683  * `sched_cond_wait_parameter` must be paired with `sched_cond_ack`.
684  *
685  * NOTE: `continuation` will only be jumped to if a wakeup has not been issued
686  *
687  * parameters:
688  *      cond             atomic condition variable to synchronize on
689  *      interruptible    interruptible value to pass to assert_wait
690  *      continuation     continuation if block succeeds
691  *      parameter
692  */
693 extern wait_result_t sched_cond_wait_parameter(
694 	sched_cond_atomic_t *cond,
695 	wait_interrupt_t interruptible,
696 	thread_continue_t continuation,
697 	void *parameter);
698 
699 /*
700  * sched_cond_wait:
701  *
702  * Assert wait and block on cond if no wakeup has been issued.
703  * If a wakeup has been issued on cond since the last `sched_cond_ack`, clear_wait and
704  * return `THREAD_AWAKENED`.
705  *
706  * `sched_cond_wait` must be paired with `sched_cond_ack`.
707  *
708  * NOTE: `continuation` will only be jumped to if a wakeup has not been issued
709  *
710  * parameters:
711  *      cond             atomic condition variable to synchronize on
712  *      interruptible    interruptible value to pass to assert_wait
713  *      continuation     continuation if block succeeds
714  */
715 extern wait_result_t sched_cond_wait(
716 	sched_cond_atomic_t *cond,
717 	wait_interrupt_t interruptible,
718 	thread_continue_t continuation);
719 
720 /*
721  * sched_cond_ack:
722  *
723  * Acknowledge an issued wakeup by clearing WAKEUP and setting ACTIVE (via XOR).
724  * It is the callers responsibility to ensure that the ACTIVE bit is always low prior to calling
725  * (i.e. by calling `sched_cond_wait` prior to any rerun or block).
726  * Synchronization schemes that allow for WAKEUP bit to be reset prior to wakeup
727  * (e.g. a cancellation mechanism) should check that WAKEUP was indeed cleared.
728  *
729  * e.g.
730  * ```
731  * if (sched_cond_ack(&my_state) & SCHED_THREAD_WAKEUP) {
732  *     // WAKEUP bit was no longer set by the time this thread woke up
733  *     do_cancellation_policy();
734  * }
735  * ```
736  *
737  * parameters:
738  *      cond:    atomic condition variable
739  */
740 extern sched_cond_t sched_cond_ack(
741 	sched_cond_atomic_t *cond);
742 
743 #endif  /* XNU_KERNEL_PRIVATE */
744 
745 #ifdef KERNEL_PRIVATE
746 /* Set pending block hint for a particular object before we go into a wait state */
747 extern void             thread_set_pending_block_hint(
748 	thread_t                        thread,
749 	block_hint_t                    block_hint);
750 
751 #define QOS_PARALLELISM_COUNT_LOGICAL   0x1
752 #define QOS_PARALLELISM_REALTIME        0x2
753 #define QOS_PARALLELISM_CLUSTER_SHARED_RESOURCE              0x4
754 
755 extern uint32_t qos_max_parallelism(int qos, uint64_t options);
756 #endif /* KERNEL_PRIVATE */
757 
758 #if XNU_KERNEL_PRIVATE
759 extern void             thread_yield_with_continuation(
760 	thread_continue_t       continuation,
761 	void                            *parameter) __dead2;
762 #endif
763 
764 /* Context switch */
765 extern wait_result_t    thread_block(
766 	thread_continue_t       continuation);
767 
768 extern wait_result_t    thread_block_parameter(
769 	thread_continue_t       continuation,
770 	void                            *parameter);
771 
772 /* Declare thread will wait on a particular event */
773 extern wait_result_t    assert_wait(
774 	event_t                         event,
775 	wait_interrupt_t        interruptible);
776 
777 /* Assert that the thread intends to wait with a timeout */
778 extern wait_result_t    assert_wait_timeout(
779 	event_t                         event,
780 	wait_interrupt_t        interruptible,
781 	uint32_t                        interval,
782 	uint32_t                        scale_factor);
783 
784 /* Assert that the thread intends to wait with an urgency, timeout and leeway */
785 extern wait_result_t    assert_wait_timeout_with_leeway(
786 	event_t                         event,
787 	wait_interrupt_t        interruptible,
788 	wait_timeout_urgency_t  urgency,
789 	uint32_t                        interval,
790 	uint32_t                        leeway,
791 	uint32_t                        scale_factor);
792 
793 extern wait_result_t    assert_wait_deadline(
794 	event_t                         event,
795 	wait_interrupt_t        interruptible,
796 	uint64_t                        deadline);
797 
798 /* Assert that the thread intends to wait with an urgency, deadline, and leeway */
799 extern wait_result_t    assert_wait_deadline_with_leeway(
800 	event_t                         event,
801 	wait_interrupt_t        interruptible,
802 	wait_timeout_urgency_t  urgency,
803 	uint64_t                        deadline,
804 	uint64_t                        leeway);
805 
806 
807 /* Wake up thread (or threads) waiting on a particular event */
808 extern kern_return_t    thread_wakeup_prim(
809 	event_t                         event,
810 	boolean_t                       one_thread,
811 	wait_result_t                   result);
812 
813 #define thread_wakeup(x)                                        \
814 	                thread_wakeup_prim((x), FALSE, THREAD_AWAKENED)
815 #define thread_wakeup_with_result(x, z)         \
816 	                thread_wakeup_prim((x), FALSE, (z))
817 #define thread_wakeup_one(x)                            \
818 	                thread_wakeup_prim((x), TRUE, THREAD_AWAKENED)
819 
820 /* Wakeup the specified thread if it is waiting on this event */
821 extern kern_return_t thread_wakeup_thread(event_t event, thread_t thread);
822 
823 extern boolean_t preemption_enabled(void);
824 
825 #ifdef MACH_KERNEL_PRIVATE
826 
827 
828 #if   !CONFIG_SCHED_TIMESHARE_CORE && !CONFIG_SCHED_CLUTCH && !CONFIG_SCHED_EDGE
829 #error Enable at least one scheduler algorithm in osfmk/conf/MASTER.XXX
830 #endif
831 
832 /*
833  * The scheduling policy is fixed at compile-time, in order to save the performance
834  * cost of function pointer indirection that we would otherwise pay each time when
835  * making a policy-specific callout.
836  */
837 #if __AMP__
838 
839 #if CONFIG_SCHED_EDGE
840 extern const struct sched_dispatch_table sched_edge_dispatch;
841 #define SCHED(f) (sched_edge_dispatch.f)
842 #else /* CONFIG_SCHED_EDGE */
843 extern const struct sched_dispatch_table sched_amp_dispatch;
844 #define SCHED(f) (sched_amp_dispatch.f)
845 #endif /* CONFIG_SCHED_EDGE */
846 
847 #else /* __AMP__ */
848 
849 #if CONFIG_SCHED_CLUTCH
850 extern const struct sched_dispatch_table sched_clutch_dispatch;
851 #define SCHED(f) (sched_clutch_dispatch.f)
852 #else /* CONFIG_SCHED_CLUTCH */
853 extern const struct sched_dispatch_table sched_dualq_dispatch;
854 #define SCHED(f) (sched_dualq_dispatch.f)
855 #endif /* CONFIG_SCHED_CLUTCH */
856 
857 #endif /* __AMP__ */
858 
859 struct sched_dispatch_table {
860 	const char *sched_name;
861 	void    (*init)(void);                          /* Init global state */
862 	void    (*timebase_init)(void);         /* Timebase-dependent initialization */
863 	void    (*processor_init)(processor_t processor);       /* Per-processor scheduler init */
864 	void    (*pset_init)(processor_set_t pset);     /* Per-processor set scheduler init */
865 
866 	void    (*maintenance_continuation)(void);      /* Function called regularly */
867 
868 	/*
869 	 * Choose a thread of greater or equal priority from the per-processor
870 	 * runqueue for timeshare/fixed threads
871 	 */
872 	thread_t        (*choose_thread)(
873 		processor_t           processor,
874 		int                           priority,
875 		thread_t              prev_thread,
876 		ast_t reason);
877 
878 	/* True if scheduler supports stealing threads for this pset */
879 	bool    (*steal_thread_enabled)(processor_set_t pset);
880 
881 	/*
882 	 * Steal a thread from another processor in the pset so that it can run
883 	 * immediately
884 	 */
885 	thread_t        (*steal_thread)(
886 		processor_set_t         pset);
887 
888 	/*
889 	 * Compute priority for a timeshare thread based on base priority.
890 	 */
891 	int (*compute_timeshare_priority)(thread_t thread);
892 
893 	/*
894 	 * Pick the best node for a thread to run on.
895 	 */
896 	pset_node_t (*choose_node)(
897 		thread_t                      thread);
898 
899 	/*
900 	 * Pick the best processor for a thread (any kind of thread) to run on.
901 	 */
902 	processor_t     (*choose_processor)(
903 		processor_set_t                pset,
904 		processor_t                    processor,
905 		thread_t                       thread);
906 	/*
907 	 * Enqueue a timeshare or fixed priority thread onto the per-processor
908 	 * runqueue
909 	 */
910 	boolean_t (*processor_enqueue)(
911 		processor_t                    processor,
912 		thread_t                       thread,
913 		sched_options_t                options);
914 
915 	/* Migrate threads away in preparation for processor shutdown */
916 	void (*processor_queue_shutdown)(
917 		processor_t                    processor);
918 
919 	/* Remove the specific thread from the per-processor runqueue */
920 	boolean_t       (*processor_queue_remove)(
921 		processor_t             processor,
922 		thread_t                thread);
923 
924 	/*
925 	 * Does the per-processor runqueue have any timeshare or fixed priority
926 	 * threads on it? Called without pset lock held, so should
927 	 * not assume immutability while executing.
928 	 */
929 	boolean_t       (*processor_queue_empty)(processor_t            processor);
930 
931 	/*
932 	 * Would this priority trigger an urgent preemption if it's sitting
933 	 * on the per-processor runqueue?
934 	 */
935 	boolean_t       (*priority_is_urgent)(int priority);
936 
937 	/*
938 	 * Does the per-processor runqueue contain runnable threads that
939 	 * should cause the currently-running thread to be preempted?
940 	 */
941 	ast_t           (*processor_csw_check)(processor_t processor);
942 
943 	/*
944 	 * Does the per-processor runqueue contain a runnable thread
945 	 * of > or >= priority, as a preflight for choose_thread() or other
946 	 * thread selection
947 	 */
948 	boolean_t       (*processor_queue_has_priority)(processor_t             processor,
949 	    int                             priority,
950 	    boolean_t               gte);
951 
952 	/* Quantum size for the specified non-realtime thread. */
953 	uint32_t        (*initial_quantum_size)(thread_t thread);
954 
955 	/* Scheduler mode for a new thread */
956 	sched_mode_t    (*initial_thread_sched_mode)(task_t parent_task);
957 
958 	/*
959 	 * Is it safe to call update_priority, which may change a thread's
960 	 * runqueue or other state. This can be used to throttle changes
961 	 * to dynamic priority.
962 	 */
963 	boolean_t       (*can_update_priority)(thread_t thread);
964 
965 	/*
966 	 * Update both scheduled priority and other persistent state.
967 	 * Side effects may including migration to another processor's runqueue.
968 	 */
969 	void            (*update_priority)(thread_t thread);
970 
971 	/* Lower overhead update to scheduled priority and state. */
972 	void            (*lightweight_update_priority)(thread_t thread);
973 
974 	/* Callback for non-realtime threads when the quantum timer fires */
975 	void            (*quantum_expire)(thread_t thread);
976 
977 	/*
978 	 * Runnable threads on per-processor runqueue. Should only
979 	 * be used for relative comparisons of load between processors.
980 	 */
981 	int                     (*processor_runq_count)(processor_t     processor);
982 
983 	/* Aggregate runcount statistics for per-processor runqueue */
984 	uint64_t    (*processor_runq_stats_count_sum)(processor_t   processor);
985 
986 	boolean_t       (*processor_bound_count)(processor_t processor);
987 
988 	void            (*thread_update_scan)(sched_update_scan_context_t scan_context);
989 
990 	/* Supports more than one pset */
991 	boolean_t   multiple_psets_enabled;
992 
993 	/* Supports avoid-processor */
994 	boolean_t   avoid_processor_enabled;
995 
996 	/* Returns true if this processor should avoid running this thread. */
997 	bool    (*thread_avoid_processor)(processor_t processor, thread_t thread, ast_t reason);
998 
999 	/*
1000 	 * Invoked when a processor is about to choose the idle thread
1001 	 * Used to send IPIs to a processor which would be preferred to be idle instead.
1002 	 * Returns true if the current processor should anticipate a quick IPI reply back
1003 	 * from another core.
1004 	 * Called with pset lock held, returns with pset lock unlocked.
1005 	 */
1006 	bool    (*processor_balance)(processor_t processor, processor_set_t pset);
1007 	rt_queue_t      (*rt_runq)(processor_set_t pset);
1008 	void    (*rt_init)(processor_set_t pset);
1009 	void    (*rt_queue_shutdown)(processor_t processor);
1010 	void    (*rt_runq_scan)(sched_update_scan_context_t scan_context);
1011 	int64_t (*rt_runq_count_sum)(void);
1012 	thread_t (*rt_steal_thread)(processor_set_t pset, uint64_t earliest_deadline);
1013 
1014 	uint32_t (*qos_max_parallelism)(int qos, uint64_t options);
1015 	void    (*check_spill)(processor_set_t pset, thread_t thread);
1016 	sched_ipi_type_t (*ipi_policy)(processor_t dst, thread_t thread, boolean_t dst_idle, sched_ipi_event_t event);
1017 	bool    (*thread_should_yield)(processor_t processor, thread_t thread);
1018 
1019 	/* Routine to update run counts */
1020 	uint32_t (*run_count_incr)(thread_t thread);
1021 	uint32_t (*run_count_decr)(thread_t thread);
1022 
1023 	/* Routine to update scheduling bucket for a thread */
1024 	void (*update_thread_bucket)(thread_t thread);
1025 
1026 	/* Routine to inform the scheduler when a new pset becomes schedulable */
1027 	void (*pset_made_schedulable)(processor_t processor, processor_set_t pset, boolean_t drop_lock);
1028 #if CONFIG_THREAD_GROUPS
1029 	/* Routine to inform the scheduler when CLPC changes a thread group recommendation */
1030 	void (*thread_group_recommendation_change)(struct thread_group *tg, cluster_type_t new_recommendation);
1031 #endif
1032 	/* Routine to inform the scheduler when all CPUs have finished initializing */
1033 	void (*cpu_init_completed)(void);
1034 	/* Routine to check if a thread is eligible to execute on a specific pset */
1035 	bool (*thread_eligible_for_pset)(thread_t thread, processor_set_t pset);
1036 };
1037 
1038 extern const struct sched_dispatch_table sched_dualq_dispatch;
1039 #if __AMP__
1040 extern const struct sched_dispatch_table sched_amp_dispatch;
1041 #endif
1042 
1043 #if defined(CONFIG_SCHED_CLUTCH)
1044 extern const struct sched_dispatch_table sched_clutch_dispatch;
1045 #endif
1046 
1047 #if defined(CONFIG_SCHED_EDGE)
1048 extern const struct sched_dispatch_table sched_edge_dispatch;
1049 #endif
1050 
1051 extern void sched_set_max_unsafe_rt_quanta(int max);
1052 extern void sched_set_max_unsafe_fixed_quanta(int max);
1053 
1054 #endif  /* MACH_KERNEL_PRIVATE */
1055 
1056 __END_DECLS
1057 
1058 #endif  /* _KERN_SCHED_PRIM_H_ */
1059