xref: /xnu-8792.61.2/osfmk/kern/sched_prim.h (revision 42e220869062b56f8d7d0726fd4c88954f87902c)
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 kern_return_t    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 /* Demote the true scheduler mode */
205 extern void             sched_thread_mode_demote(thread_t thread,
206     uint32_t reason);
207 /* Un-demote the true scheduler mode */
208 extern void             sched_thread_mode_undemote(thread_t thread,
209     uint32_t reason);
210 
211 extern void sched_thread_promote_reason(thread_t thread, uint32_t reason, uintptr_t trace_obj);
212 extern void sched_thread_unpromote_reason(thread_t thread, uint32_t reason, uintptr_t trace_obj);
213 
214 /* Re-evaluate base priority of thread (thread locked) */
215 void thread_recompute_priority(thread_t thread);
216 
217 /* Re-evaluate scheduled priority of thread (thread locked) */
218 extern void thread_recompute_sched_pri(
219 	thread_t thread,
220 	set_sched_pri_options_t options);
221 
222 /* Periodic scheduler activity */
223 extern void             sched_init_thread(void);
224 
225 /* Perform sched_tick housekeeping activities */
226 extern boolean_t                can_update_priority(
227 	thread_t                thread);
228 
229 extern void             update_priority(
230 	thread_t                thread);
231 
232 extern void             lightweight_update_priority(
233 	thread_t                thread);
234 
235 extern void             sched_default_quantum_expire(thread_t thread);
236 
237 /* Idle processor thread continuation */
238 extern void             idle_thread(
239 	void*           parameter,
240 	wait_result_t   result);
241 
242 extern kern_return_t    idle_thread_create(
243 	processor_t             processor);
244 
245 /* Continuation return from syscall */
246 extern void     thread_syscall_return(
247 	kern_return_t   ret);
248 
249 /* Context switch */
250 extern wait_result_t    thread_block_reason(
251 	thread_continue_t       continuation,
252 	void                            *parameter,
253 	ast_t                           reason);
254 
255 __options_decl(sched_options_t, uint32_t, {
256 	SCHED_NONE      = 0x0,
257 	SCHED_TAILQ     = 0x1,
258 	SCHED_HEADQ     = 0x2,
259 	SCHED_PREEMPT   = 0x4,
260 	SCHED_REBALANCE = 0x8,
261 });
262 
263 /* Reschedule thread for execution */
264 extern void             thread_setrun(
265 	thread_t        thread,
266 	sched_options_t options);
267 
268 extern processor_set_t  task_choose_pset(
269 	task_t                  task);
270 
271 /* Bind the current thread to a particular processor */
272 extern processor_t              thread_bind(
273 	processor_t             processor);
274 
275 extern bool pset_has_stealable_threads(
276 	processor_set_t         pset);
277 
278 extern bool pset_has_stealable_rt_threads(
279 	processor_set_t         pset);
280 
281 extern processor_set_t choose_starting_pset(
282 	pset_node_t  node,
283 	thread_t     thread,
284 	processor_t *processor_hint);
285 
286 extern int pset_available_cpu_count(
287 	processor_set_t pset);
288 
289 extern bool pset_is_recommended(
290 	processor_set_t pset);
291 
292 extern pset_node_t sched_choose_node(
293 	thread_t     thread);
294 
295 /* Choose the best processor to run a thread */
296 extern processor_t      choose_processor(
297 	processor_set_t                pset,
298 	processor_t                    processor,
299 	thread_t                       thread);
300 
301 extern void sched_SMT_balance(
302 	processor_t processor,
303 	processor_set_t pset);
304 
305 extern void thread_quantum_init(
306 	thread_t thread);
307 
308 extern void             run_queue_init(
309 	run_queue_t             runq);
310 
311 extern thread_t run_queue_dequeue(
312 	run_queue_t           runq,
313 	sched_options_t       options);
314 
315 extern boolean_t        run_queue_enqueue(
316 	run_queue_t           runq,
317 	thread_t              thread,
318 	sched_options_t       options);
319 
320 extern void     run_queue_remove(
321 	run_queue_t            runq,
322 	thread_t                       thread);
323 
324 extern thread_t run_queue_peek(
325 	run_queue_t            runq);
326 
327 struct sched_update_scan_context {
328 	uint64_t        earliest_bg_make_runnable_time;
329 	uint64_t        earliest_normal_make_runnable_time;
330 	uint64_t        earliest_rt_make_runnable_time;
331 	uint64_t        sched_tick_last_abstime;
332 };
333 typedef struct sched_update_scan_context *sched_update_scan_context_t;
334 
335 extern void             sched_rtlocal_runq_scan(sched_update_scan_context_t scan_context);
336 
337 extern void sched_pset_made_schedulable(
338 	processor_t processor,
339 	processor_set_t pset,
340 	boolean_t drop_lock);
341 
342 extern void sched_cpu_init_completed(void);
343 
344 /*
345  * Enum to define various events which need IPIs. The IPI policy
346  * engine decides what kind of IPI to use based on destination
347  * processor state, thread and one of the following scheduling events.
348  */
349 typedef enum {
350 	SCHED_IPI_EVENT_BOUND_THR   = 0x1,
351 	SCHED_IPI_EVENT_PREEMPT     = 0x2,
352 	SCHED_IPI_EVENT_SMT_REBAL   = 0x3,
353 	SCHED_IPI_EVENT_SPILL       = 0x4,
354 	SCHED_IPI_EVENT_REBALANCE   = 0x5,
355 	SCHED_IPI_EVENT_RT_PREEMPT  = 0x6,
356 } sched_ipi_event_t;
357 
358 
359 /* Enum to define various IPI types used by the scheduler */
360 typedef enum {
361 	SCHED_IPI_NONE              = 0x0,
362 	SCHED_IPI_IMMEDIATE         = 0x1,
363 	SCHED_IPI_IDLE              = 0x2,
364 	SCHED_IPI_DEFERRED          = 0x3,
365 } sched_ipi_type_t;
366 
367 /* The IPI policy engine behaves in the following manner:
368  * - All scheduler events which need an IPI invoke sched_ipi_action() with
369  *   the appropriate destination processor, thread and event.
370  * - sched_ipi_action() performs basic checks, invokes the scheduler specific
371  *   ipi_policy routine and sets pending_AST bits based on the result.
372  * - Once the pset lock is dropped, the scheduler invokes sched_ipi_perform()
373  *   routine which actually sends the appropriate IPI to the destination core.
374  */
375 extern sched_ipi_type_t sched_ipi_action(processor_t dst, thread_t thread, sched_ipi_event_t event);
376 extern void sched_ipi_perform(processor_t dst, sched_ipi_type_t ipi);
377 
378 /* sched_ipi_policy() is the global default IPI policy for all schedulers */
379 extern sched_ipi_type_t sched_ipi_policy(processor_t dst, thread_t thread,
380     boolean_t dst_idle, sched_ipi_event_t event);
381 
382 /* sched_ipi_deferred_policy() is the global default deferred IPI policy for all schedulers */
383 extern sched_ipi_type_t sched_ipi_deferred_policy(processor_set_t pset,
384     processor_t dst, thread_t thread, sched_ipi_event_t event);
385 
386 #if defined(CONFIG_SCHED_TIMESHARE_CORE)
387 
388 extern boolean_t        thread_update_add_thread(thread_t thread);
389 extern void             thread_update_process_threads(void);
390 extern boolean_t        runq_scan(run_queue_t runq, sched_update_scan_context_t scan_context);
391 
392 #if CONFIG_SCHED_CLUTCH
393 extern boolean_t        sched_clutch_timeshare_scan(queue_t thread_queue, uint16_t count, sched_update_scan_context_t scan_context);
394 #endif /* CONFIG_SCHED_CLUTCH */
395 
396 extern void sched_timeshare_init(void);
397 extern void sched_timeshare_timebase_init(void);
398 extern void sched_timeshare_maintenance_continue(void);
399 
400 extern boolean_t priority_is_urgent(int priority);
401 extern uint32_t sched_timeshare_initial_quantum_size(thread_t thread);
402 
403 extern int sched_compute_timeshare_priority(thread_t thread);
404 
405 #endif /* CONFIG_SCHED_TIMESHARE_CORE */
406 
407 /* Remove thread from its run queue */
408 extern boolean_t        thread_run_queue_remove(thread_t thread);
409 thread_t thread_run_queue_remove_for_handoff(thread_t thread);
410 
411 /* Put a thread back in the run queue after being yanked */
412 extern void thread_run_queue_reinsert(thread_t thread, sched_options_t options);
413 
414 extern void             thread_timer_expire(
415 	void                    *thread,
416 	void                    *p1);
417 
418 extern bool thread_is_eager_preempt(thread_t thread);
419 
420 extern boolean_t sched_generic_direct_dispatch_to_idle_processors;
421 
422 /* Set the maximum interrupt level for the thread */
423 __private_extern__ wait_interrupt_t thread_interrupt_level(
424 	wait_interrupt_t interruptible);
425 
426 __private_extern__ wait_result_t thread_mark_wait_locked(
427 	thread_t                 thread,
428 	wait_interrupt_t interruptible);
429 
430 /* Wake up locked thread directly, passing result */
431 __private_extern__ kern_return_t clear_wait_internal(
432 	thread_t                thread,
433 	wait_result_t   result);
434 
435 struct sched_statistics {
436 	uint32_t        csw_count;
437 	uint32_t        preempt_count;
438 	uint32_t        preempted_rt_count;
439 	uint32_t        preempted_by_rt_count;
440 	uint32_t        rt_sched_count;
441 	uint32_t        interrupt_count;
442 	uint32_t        ipi_count;
443 	uint32_t        timer_pop_count;
444 	uint32_t        idle_transitions;
445 	uint32_t        quantum_timer_expirations;
446 };
447 PERCPU_DECL(struct sched_statistics, sched_stats);
448 extern bool             sched_stats_active;
449 
450 extern void sched_stats_handle_csw(
451 	processor_t processor,
452 	int reasons,
453 	int selfpri,
454 	int otherpri);
455 
456 extern void sched_stats_handle_runq_change(
457 	struct runq_stats *stats,
458 	int old_count);
459 
460 #define SCHED_STATS_INC(field)                                                  \
461 MACRO_BEGIN                                                                     \
462 	if (__improbable(sched_stats_active)) {                                 \
463 	        PERCPU_GET(sched_stats)->field++;                               \
464 	}                                                                       \
465 MACRO_END
466 
467 #if DEBUG
468 
469 #define SCHED_STATS_CSW(processor, reasons, selfpri, otherpri)                  \
470 MACRO_BEGIN                                                                     \
471 	if (__improbable(sched_stats_active)) {                                 \
472 	        sched_stats_handle_csw((processor),                             \
473 	            (reasons), (selfpri), (otherpri));                          \
474 	}                                                                       \
475 MACRO_END
476 
477 
478 #define SCHED_STATS_RUNQ_CHANGE(stats, old_count)                               \
479 MACRO_BEGIN                                                                     \
480 	if (__improbable(sched_stats_active)) {                                 \
481 	        sched_stats_handle_runq_change((stats), (old_count));           \
482 	}                                                                       \
483 MACRO_END
484 
485 #else /* DEBUG */
486 
487 #define SCHED_STATS_CSW(processor, reasons, selfpri, otherpri) do { }while(0)
488 #define SCHED_STATS_RUNQ_CHANGE(stats, old_count) do { }while(0)
489 
490 #endif /* DEBUG */
491 
492 extern uint32_t sched_debug_flags;
493 #define SCHED_DEBUG_FLAG_PLATFORM_TRACEPOINTS           0x00000001
494 #define SCHED_DEBUG_FLAG_CHOOSE_PROCESSOR_TRACEPOINTS   0x00000002
495 
496 #define SCHED_DEBUG_PLATFORM_KERNEL_DEBUG_CONSTANT(...)                         \
497 MACRO_BEGIN                                                                     \
498 	if (__improbable(sched_debug_flags &                                    \
499 	    SCHED_DEBUG_FLAG_PLATFORM_TRACEPOINTS)) {                           \
500 	        KERNEL_DEBUG_CONSTANT(__VA_ARGS__);                             \
501 	}                                                                       \
502 MACRO_END
503 
504 #define SCHED_DEBUG_CHOOSE_PROCESSOR_KERNEL_DEBUG_CONSTANT(...)                 \
505 MACRO_BEGIN                                                                     \
506 	if (__improbable(sched_debug_flags &                                    \
507 	    SCHED_DEBUG_FLAG_CHOOSE_PROCESSOR_TRACEPOINTS)) {                   \
508 	        KERNEL_DEBUG_CONSTANT(__VA_ARGS__);                             \
509 	}                                                                       \
510 MACRO_END
511 
512 /* Tells if there are "active" RT threads in the system (provided by CPU PM) */
513 extern void     active_rt_threads(
514 	boolean_t       active);
515 
516 /* Returns the perfcontrol attribute for the thread */
517 extern perfcontrol_class_t thread_get_perfcontrol_class(
518 	thread_t        thread);
519 
520 /* Generic routine for Non-AMP schedulers to calculate parallelism */
521 extern uint32_t sched_qos_max_parallelism(int qos, uint64_t options);
522 
523 extern void check_monotonic_time(uint64_t ctime);
524 
525 #endif /* MACH_KERNEL_PRIVATE */
526 
527 __BEGIN_DECLS
528 
529 #ifdef  XNU_KERNEL_PRIVATE
530 
531 extern void thread_bind_cluster_type(thread_t, char cluster_type, bool soft_bind);
532 
533 __options_decl(thread_bind_option_t, uint64_t, {
534 	/* Unbind a previously cluster bound thread */
535 	THREAD_UNBIND                   = 0x1,
536 	/*
537 	 * Soft bind the thread to the cluster; soft binding means the thread will be
538 	 * moved to an available cluster if the bound cluster is de-recommended/offline.
539 	 */
540 	THREAD_BIND_SOFT                = 0x2,
541 	/*
542 	 * Bind thread to the cluster only if it is eligible to run on that cluster. If
543 	 * the thread is not eligible to run on the cluster, thread_bind_cluster_id()
544 	 * returns KERN_INVALID_POLICY.
545 	 */
546 	THREAD_BIND_ELIGIBLE_ONLY       = 0x4,
547 });
548 extern kern_return_t thread_bind_cluster_id(thread_t thread, uint32_t cluster_id, thread_bind_option_t options);
549 
550 extern int sched_get_rt_n_backup_processors(void);
551 extern void sched_set_rt_n_backup_processors(int n);
552 
553 extern int sched_get_rt_deadline_epsilon(void);
554 extern void sched_set_rt_deadline_epsilon(int new_epsilon_us);
555 
556 /* Toggles a global override to turn off CPU Throttling */
557 extern void     sys_override_cpu_throttle(boolean_t enable_override);
558 
559 extern int sched_get_powered_cores(void);
560 extern void sched_set_powered_cores(int n);
561 
562 /*
563  ****************** Only exported until BSD stops using ********************
564  */
565 
566 extern void                     thread_vm_bind_group_add(void);
567 
568 /* Wake up thread directly, passing result */
569 extern kern_return_t clear_wait(
570 	thread_t                thread,
571 	wait_result_t   result);
572 
573 /* Start thread running */
574 extern void             thread_bootstrap_return(void) __attribute__((noreturn));
575 
576 /* Return from exception (BSD-visible interface) */
577 extern void             thread_exception_return(void) __dead2;
578 
579 #define SCHED_STRING_MAX_LENGTH (48)
580 /* String declaring the name of the current scheduler */
581 extern char sched_string[SCHED_STRING_MAX_LENGTH];
582 
583 __options_decl(thread_handoff_option_t, uint32_t, {
584 	THREAD_HANDOFF_NONE          = 0,
585 	THREAD_HANDOFF_SETRUN_NEEDED = 0x1,
586 });
587 
588 /* Remove thread from its run queue */
589 thread_t thread_prepare_for_handoff(thread_t thread, thread_handoff_option_t option);
590 
591 /* Attempt to context switch to a specific runnable thread */
592 extern wait_result_t thread_handoff_deallocate(thread_t thread, thread_handoff_option_t option);
593 
594 __attribute__((nonnull(2)))
595 extern void thread_handoff_parameter(thread_t thread,
596     thread_continue_t continuation, void *parameter, thread_handoff_option_t) __dead2;
597 
598 extern struct waitq     *assert_wait_queue(event_t event);
599 
600 extern kern_return_t thread_wakeup_one_with_pri(event_t event, int priority);
601 
602 extern thread_t thread_wakeup_identify(event_t event, int priority);
603 
604 /*
605  * sched_cond_t:
606  *
607  * A atomic condition variable used to synchronize wake/block operations on threads.
608  * Bits defined below are reserved for use by sched_prim. Remaining
609  * bits may be used by caller for additional synchronization semantics.
610  */
611 __options_decl(sched_cond_t, uint32_t, {
612 	SCHED_COND_INIT = 0x0000,    /* initialize all bits to zero (inactive and not awoken) */
613 	SCHED_COND_ACTIVE = 0x0001,  /* target thread is active */
614 	SCHED_COND_WAKEUP = 0x0002   /* wakeup has been issued for target thread */
615 });
616 typedef _Atomic sched_cond_t sched_cond_atomic_t;
617 
618 /*
619  * sched_cond_init:
620  *
621  * Initialize an atomic condition variable. Note that this does not occur atomically and should be
622  * performed during thread initialization, before the condition is observable by other threads.
623  */
624 extern void sched_cond_init(
625 	sched_cond_atomic_t *cond);
626 
627 /*
628  * sched_cond_signal:
629  *
630  * Wakeup the specified thread if it is waiting on this event and it has not already been issued a wakeup.
631  *
632  * parameters:
633  *      thread    thread to awaken
634  *      cond      atomic condition variable
635  */
636 extern kern_return_t sched_cond_signal(
637 	sched_cond_atomic_t *cond,
638 	thread_t thread);
639 
640 /*
641  * sched_cond_wait:
642  *
643  * Assert wait and block on cond if no wakeup has been issued.
644  * If a wakeup has been issued on cond since the last `sched_cond_ack`, clear_wait and
645  * return `THREAD_AWAKENED`.
646  *
647  * `sched_cond_wait` must be paired with `sched_cond_ack`.
648  *
649  * NOTE: `continuation` will only be jumped to if a wakeup has not been issued
650  *
651  * parameters:
652  *      cond             atomic condition variable to synchronize on
653  *      interruptible    interruptible value to pass to assert_wait
654  *      continuation     continuation if block succeeds
655  */
656 extern wait_result_t sched_cond_wait(
657 	sched_cond_atomic_t *cond,
658 	wait_interrupt_t interruptible,
659 	thread_continue_t continuation);
660 
661 /*
662  * sched_cond_ack:
663  *
664  * Acknowledge an issued wakeup by clearing WAKEUP and setting ACTIVE (via XOR).
665  * It is the callers responsibility to ensure that the ACTIVE bit is always low prior to calling
666  * (i.e. by calling `sched_cond_wait` prior to any rerun or block).
667  * Synchronization schemes that allow for WAKEUP bit to be reset prior to wakeup
668  * (e.g. a cancellation mechanism) should check that WAKEUP was indeed cleared.
669  *
670  * e.g.
671  * ```
672  * if (sched_cond_ack(&my_state) & SCHED_THREAD_WAKEUP) {
673  *     // WAKEUP bit was no longer set by the time this thread woke up
674  *     do_cancellation_policy();
675  * }
676  * ```
677  *
678  * parameters:
679  *      cond:    atomic condition variable
680  */
681 extern sched_cond_t sched_cond_ack(
682 	sched_cond_atomic_t *cond);
683 
684 #endif  /* XNU_KERNEL_PRIVATE */
685 
686 #ifdef KERNEL_PRIVATE
687 /* Set pending block hint for a particular object before we go into a wait state */
688 extern void             thread_set_pending_block_hint(
689 	thread_t                        thread,
690 	block_hint_t                    block_hint);
691 
692 #define QOS_PARALLELISM_COUNT_LOGICAL   0x1
693 #define QOS_PARALLELISM_REALTIME        0x2
694 #define QOS_PARALLELISM_CLUSTER_SHARED_RESOURCE              0x4
695 
696 extern uint32_t qos_max_parallelism(int qos, uint64_t options);
697 #endif /* KERNEL_PRIVATE */
698 
699 #if XNU_KERNEL_PRIVATE
700 extern void             thread_yield_with_continuation(
701 	thread_continue_t       continuation,
702 	void                            *parameter) __dead2;
703 #endif
704 
705 /* Context switch */
706 extern wait_result_t    thread_block(
707 	thread_continue_t       continuation);
708 
709 extern wait_result_t    thread_block_parameter(
710 	thread_continue_t       continuation,
711 	void                            *parameter);
712 
713 /* Declare thread will wait on a particular event */
714 extern wait_result_t    assert_wait(
715 	event_t                         event,
716 	wait_interrupt_t        interruptible);
717 
718 /* Assert that the thread intends to wait with a timeout */
719 extern wait_result_t    assert_wait_timeout(
720 	event_t                         event,
721 	wait_interrupt_t        interruptible,
722 	uint32_t                        interval,
723 	uint32_t                        scale_factor);
724 
725 /* Assert that the thread intends to wait with an urgency, timeout and leeway */
726 extern wait_result_t    assert_wait_timeout_with_leeway(
727 	event_t                         event,
728 	wait_interrupt_t        interruptible,
729 	wait_timeout_urgency_t  urgency,
730 	uint32_t                        interval,
731 	uint32_t                        leeway,
732 	uint32_t                        scale_factor);
733 
734 extern wait_result_t    assert_wait_deadline(
735 	event_t                         event,
736 	wait_interrupt_t        interruptible,
737 	uint64_t                        deadline);
738 
739 /* Assert that the thread intends to wait with an urgency, deadline, and leeway */
740 extern wait_result_t    assert_wait_deadline_with_leeway(
741 	event_t                         event,
742 	wait_interrupt_t        interruptible,
743 	wait_timeout_urgency_t  urgency,
744 	uint64_t                        deadline,
745 	uint64_t                        leeway);
746 
747 
748 /* Wake up thread (or threads) waiting on a particular event */
749 extern kern_return_t    thread_wakeup_prim(
750 	event_t                         event,
751 	boolean_t                       one_thread,
752 	wait_result_t                   result);
753 
754 #define thread_wakeup(x)                                        \
755 	                thread_wakeup_prim((x), FALSE, THREAD_AWAKENED)
756 #define thread_wakeup_with_result(x, z)         \
757 	                thread_wakeup_prim((x), FALSE, (z))
758 #define thread_wakeup_one(x)                            \
759 	                thread_wakeup_prim((x), TRUE, THREAD_AWAKENED)
760 
761 /* Wakeup the specified thread if it is waiting on this event */
762 extern kern_return_t thread_wakeup_thread(event_t event, thread_t thread);
763 
764 extern boolean_t preemption_enabled(void);
765 
766 #ifdef MACH_KERNEL_PRIVATE
767 
768 /*
769  * Scheduler algorithm indirection. If only one algorithm is
770  * enabled at compile-time, a direction function call is used.
771  * If more than one is enabled, calls are dispatched through
772  * a function pointer table.
773  */
774 
775 #if   !defined(CONFIG_SCHED_TRADITIONAL) && !defined(CONFIG_SCHED_PROTO) && !defined(CONFIG_SCHED_GRRR) && !defined(CONFIG_SCHED_MULTIQ) && !defined(CONFIG_SCHED_CLUTCH) && !defined(CONFIG_SCHED_EDGE)
776 #error Enable at least one scheduler algorithm in osfmk/conf/MASTER.XXX
777 #endif
778 
779 #if __AMP__
780 
781 #if CONFIG_SCHED_EDGE
782 extern const struct sched_dispatch_table sched_edge_dispatch;
783 #define SCHED(f) (sched_edge_dispatch.f)
784 #else /* CONFIG_SCHED_EDGE */
785 extern const struct sched_dispatch_table sched_amp_dispatch;
786 #define SCHED(f) (sched_amp_dispatch.f)
787 #endif /* CONFIG_SCHED_EDGE */
788 
789 #else /* __AMP__ */
790 
791 #if CONFIG_SCHED_CLUTCH
792 extern const struct sched_dispatch_table sched_clutch_dispatch;
793 #define SCHED(f) (sched_clutch_dispatch.f)
794 #else /* CONFIG_SCHED_CLUTCH */
795 extern const struct sched_dispatch_table sched_dualq_dispatch;
796 #define SCHED(f) (sched_dualq_dispatch.f)
797 #endif /* CONFIG_SCHED_CLUTCH */
798 
799 #endif /* __AMP__ */
800 
801 struct sched_dispatch_table {
802 	const char *sched_name;
803 	void    (*init)(void);                          /* Init global state */
804 	void    (*timebase_init)(void);         /* Timebase-dependent initialization */
805 	void    (*processor_init)(processor_t processor);       /* Per-processor scheduler init */
806 	void    (*pset_init)(processor_set_t pset);     /* Per-processor set scheduler init */
807 
808 	void    (*maintenance_continuation)(void);      /* Function called regularly */
809 
810 	/*
811 	 * Choose a thread of greater or equal priority from the per-processor
812 	 * runqueue for timeshare/fixed threads
813 	 */
814 	thread_t        (*choose_thread)(
815 		processor_t           processor,
816 		int                           priority,
817 		ast_t reason);
818 
819 	/* True if scheduler supports stealing threads for this pset */
820 	bool    (*steal_thread_enabled)(processor_set_t pset);
821 
822 	/*
823 	 * Steal a thread from another processor in the pset so that it can run
824 	 * immediately
825 	 */
826 	thread_t        (*steal_thread)(
827 		processor_set_t         pset);
828 
829 	/*
830 	 * Compute priority for a timeshare thread based on base priority.
831 	 */
832 	int (*compute_timeshare_priority)(thread_t thread);
833 
834 	/*
835 	 * Pick the best node for a thread to run on.
836 	 */
837 	pset_node_t (*choose_node)(
838 		thread_t                      thread);
839 
840 	/*
841 	 * Pick the best processor for a thread (any kind of thread) to run on.
842 	 */
843 	processor_t     (*choose_processor)(
844 		processor_set_t                pset,
845 		processor_t                    processor,
846 		thread_t                       thread);
847 	/*
848 	 * Enqueue a timeshare or fixed priority thread onto the per-processor
849 	 * runqueue
850 	 */
851 	boolean_t (*processor_enqueue)(
852 		processor_t                    processor,
853 		thread_t                       thread,
854 		sched_options_t                options);
855 
856 	/* Migrate threads away in preparation for processor shutdown */
857 	void (*processor_queue_shutdown)(
858 		processor_t                    processor);
859 
860 	/* Remove the specific thread from the per-processor runqueue */
861 	boolean_t       (*processor_queue_remove)(
862 		processor_t             processor,
863 		thread_t                thread);
864 
865 	/*
866 	 * Does the per-processor runqueue have any timeshare or fixed priority
867 	 * threads on it? Called without pset lock held, so should
868 	 * not assume immutability while executing.
869 	 */
870 	boolean_t       (*processor_queue_empty)(processor_t            processor);
871 
872 	/*
873 	 * Would this priority trigger an urgent preemption if it's sitting
874 	 * on the per-processor runqueue?
875 	 */
876 	boolean_t       (*priority_is_urgent)(int priority);
877 
878 	/*
879 	 * Does the per-processor runqueue contain runnable threads that
880 	 * should cause the currently-running thread to be preempted?
881 	 */
882 	ast_t           (*processor_csw_check)(processor_t processor);
883 
884 	/*
885 	 * Does the per-processor runqueue contain a runnable thread
886 	 * of > or >= priority, as a preflight for choose_thread() or other
887 	 * thread selection
888 	 */
889 	boolean_t       (*processor_queue_has_priority)(processor_t             processor,
890 	    int                             priority,
891 	    boolean_t               gte);
892 
893 	/* Quantum size for the specified non-realtime thread. */
894 	uint32_t        (*initial_quantum_size)(thread_t thread);
895 
896 	/* Scheduler mode for a new thread */
897 	sched_mode_t    (*initial_thread_sched_mode)(task_t parent_task);
898 
899 	/*
900 	 * Is it safe to call update_priority, which may change a thread's
901 	 * runqueue or other state. This can be used to throttle changes
902 	 * to dynamic priority.
903 	 */
904 	boolean_t       (*can_update_priority)(thread_t thread);
905 
906 	/*
907 	 * Update both scheduled priority and other persistent state.
908 	 * Side effects may including migration to another processor's runqueue.
909 	 */
910 	void            (*update_priority)(thread_t thread);
911 
912 	/* Lower overhead update to scheduled priority and state. */
913 	void            (*lightweight_update_priority)(thread_t thread);
914 
915 	/* Callback for non-realtime threads when the quantum timer fires */
916 	void            (*quantum_expire)(thread_t thread);
917 
918 	/*
919 	 * Runnable threads on per-processor runqueue. Should only
920 	 * be used for relative comparisons of load between processors.
921 	 */
922 	int                     (*processor_runq_count)(processor_t     processor);
923 
924 	/* Aggregate runcount statistics for per-processor runqueue */
925 	uint64_t    (*processor_runq_stats_count_sum)(processor_t   processor);
926 
927 	boolean_t       (*processor_bound_count)(processor_t processor);
928 
929 	void            (*thread_update_scan)(sched_update_scan_context_t scan_context);
930 
931 	/* Supports more than one pset */
932 	boolean_t   multiple_psets_enabled;
933 	/* Supports scheduler groups */
934 	boolean_t   sched_groups_enabled;
935 
936 	/* Supports avoid-processor */
937 	boolean_t   avoid_processor_enabled;
938 
939 	/* Returns true if this processor should avoid running this thread. */
940 	bool    (*thread_avoid_processor)(processor_t processor, thread_t thread);
941 
942 	/*
943 	 * Invoked when a processor is about to choose the idle thread
944 	 * Used to send IPIs to a processor which would be preferred to be idle instead.
945 	 * Called with pset lock held, returns pset lock unlocked.
946 	 */
947 	void    (*processor_balance)(processor_t processor, processor_set_t pset);
948 	rt_queue_t      (*rt_runq)(processor_set_t pset);
949 	void    (*rt_init)(processor_set_t pset);
950 	void    (*rt_queue_shutdown)(processor_t processor);
951 	void    (*rt_runq_scan)(sched_update_scan_context_t scan_context);
952 	int64_t (*rt_runq_count_sum)(void);
953 	thread_t (*rt_steal_thread)(processor_set_t pset, uint64_t earliest_deadline);
954 
955 	uint32_t (*qos_max_parallelism)(int qos, uint64_t options);
956 	void    (*check_spill)(processor_set_t pset, thread_t thread);
957 	sched_ipi_type_t (*ipi_policy)(processor_t dst, thread_t thread, boolean_t dst_idle, sched_ipi_event_t event);
958 	bool    (*thread_should_yield)(processor_t processor, thread_t thread);
959 
960 	/* Routine to update run counts */
961 	uint32_t (*run_count_incr)(thread_t thread);
962 	uint32_t (*run_count_decr)(thread_t thread);
963 
964 	/* Routine to update scheduling bucket for a thread */
965 	void (*update_thread_bucket)(thread_t thread);
966 
967 	/* Routine to inform the scheduler when a new pset becomes schedulable */
968 	void (*pset_made_schedulable)(processor_t processor, processor_set_t pset, boolean_t drop_lock);
969 #if CONFIG_THREAD_GROUPS
970 	/* Routine to inform the scheduler when CLPC changes a thread group recommendation */
971 	void (*thread_group_recommendation_change)(struct thread_group *tg, cluster_type_t new_recommendation);
972 #endif
973 	/* Routine to inform the scheduler when all CPUs have finished initializing */
974 	void (*cpu_init_completed)(void);
975 	/* Routine to check if a thread is eligible to execute on a specific pset */
976 	bool (*thread_eligible_for_pset)(thread_t thread, processor_set_t pset);
977 };
978 
979 #if defined(CONFIG_SCHED_TRADITIONAL)
980 extern const struct sched_dispatch_table sched_traditional_dispatch;
981 extern const struct sched_dispatch_table sched_traditional_with_pset_runqueue_dispatch;
982 #endif
983 
984 #if defined(CONFIG_SCHED_MULTIQ)
985 extern const struct sched_dispatch_table sched_multiq_dispatch;
986 extern const struct sched_dispatch_table sched_dualq_dispatch;
987 #if __AMP__
988 extern const struct sched_dispatch_table sched_amp_dispatch;
989 #endif
990 #endif
991 
992 #if defined(CONFIG_SCHED_PROTO)
993 extern const struct sched_dispatch_table sched_proto_dispatch;
994 #endif
995 
996 #if defined(CONFIG_SCHED_GRRR)
997 extern const struct sched_dispatch_table sched_grrr_dispatch;
998 #endif
999 
1000 #if defined(CONFIG_SCHED_CLUTCH)
1001 extern const struct sched_dispatch_table sched_clutch_dispatch;
1002 #endif
1003 
1004 #if defined(CONFIG_SCHED_EDGE)
1005 extern const struct sched_dispatch_table sched_edge_dispatch;
1006 #endif
1007 
1008 extern void sched_set_max_unsafe_rt_quanta(int max);
1009 extern void sched_set_max_unsafe_fixed_quanta(int max);
1010 
1011 #endif  /* MACH_KERNEL_PRIVATE */
1012 
1013 __END_DECLS
1014 
1015 #endif  /* _KERN_SCHED_PRIM_H_ */
1016