[apple/xnu.git] / osfmk / kern / waitq.h

#ifndef _WAITQ_H_
#define _WAITQ_H_
/*
 * Copyright (c) 2014-2015 Apple Computer, Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 *
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 *
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 *
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */
#ifdef	KERNEL_PRIVATE

#include <mach/mach_types.h>
#include <mach/sync_policy.h>
#include <mach/kern_return.h>		/* for kern_return_t */

#include <kern/kern_types.h>		/* for wait_queue_t */
#include <kern/queue.h>
#include <kern/assert.h>

#include <sys/cdefs.h>

/*
 * Constants and types used in the waitq APIs
 */
#define WAITQ_ALL_PRIORITIES   (-1)
#define WAITQ_PROMOTE_PRIORITY (-2)
#define WAITQ_SELECT_MAX_PRI   (-3)

typedef enum e_waitq_lock_state {
	WAITQ_KEEP_LOCKED    = 0x01,
	WAITQ_UNLOCK         = 0x02,
	WAITQ_SHOULD_LOCK    = 0x04,
	WAITQ_ALREADY_LOCKED = 0x08,
	WAITQ_DONT_LOCK      = 0x10,
} waitq_lock_state_t;

/*
 * The Jenkins "one at a time" hash.
 * TBD: There may be some value to unrolling here,
 * depending on the architecture.
 */
static __inline__ uint32_t 
jenkins_hash(char *key, size_t length)
{
	uint32_t hash = 0;
	size_t i;

	for (i = 0; i < length; i++) {
		hash += (uint32_t)key[i];
		hash += (hash << 10);
		hash ^= (hash >> 6);
	}

	hash += (hash << 3);
	hash ^= (hash >> 11);
	hash += (hash << 15);

	return hash;
}

/* Opaque sizes and alignment used for struct verification */
#if   __x86_64__
	#define WQ_OPAQUE_ALIGN   8
	#define WQS_OPAQUE_ALIGN  8
	#define WQ_OPAQUE_SIZE   48
	#define WQS_OPAQUE_SIZE  64
#else
	#error Unknown size requirement
#endif

#ifdef MACH_KERNEL_PRIVATE

#include <kern/spl.h>
#include <kern/simple_lock.h>
#include <mach/branch_predicates.h>

#include <machine/cpu_number.h>
#include <machine/machine_routines.h> /* machine_timeout_suspended() */

/*
 * The event mask is of 59 bits on 64 bit architeture and 27 bits on
 * 32 bit architecture and so we calculate its size using sizeof(long).
 * If the bitfield for wq_type and wq_fifo is changed, then value of
 * EVENT_MASK_BITS will also change.
 *
 * New plan: this is an optimization anyway, so I'm stealing 32bits
 * from the mask to shrink the waitq object even further.
 */
#define _EVENT_MASK_BITS   ((sizeof(uint32_t) * 8) - 6)

#define WAITQ_BOOST_PRIORITY 31

enum waitq_type {
	WQT_INVALID = 0,
	WQT_QUEUE   = 0x2,
	WQT_SET     = 0x3,
};

#if CONFIG_WAITQ_STATS
#define NWAITQ_BTFRAMES 5
struct wq_stats {
	uint64_t waits;
	uint64_t wakeups;
	uint64_t clears;
	uint64_t failed_wakeups;

	uintptr_t last_wait[NWAITQ_BTFRAMES];
	uintptr_t last_wakeup[NWAITQ_BTFRAMES];
	uintptr_t last_failed_wakeup[NWAITQ_BTFRAMES];
};
#endif

/*
 *	struct waitq
 *
 *	This is the definition of the common event wait queue
 *	that the scheduler APIs understand.  It is used
 *	internally by the gerneralized event waiting mechanism
 *	(assert_wait), and also for items that maintain their
 *	own wait queues (such as ports and semaphores).
 *
 *	It is not published to other kernel components.
 *
 *	NOTE:  Hardware locks are used to protect event wait
 *	queues since interrupt code is free to post events to
 *	them.
 */
struct waitq {
	uint32_t /* flags */
		waitq_type:2,    /* only public field */
		waitq_fifo:1,    /* fifo wakeup policy? */
		waitq_prepost:1, /* waitq supports prepost? */
		waitq_irq:1,     /* waitq requires interrupts disabled */
		waitq_isvalid:1, /* waitq structure is valid */
		waitq_eventmask:_EVENT_MASK_BITS;
		/* the wait queue set (set-of-sets) to which this queue belongs */
	hw_lock_data_t	waitq_interlock;	/* interlock */

	uint64_t waitq_set_id;
	uint64_t waitq_prepost_id;
	queue_head_t	waitq_queue;		/* queue of elements */
};

static_assert(sizeof(struct waitq) == WQ_OPAQUE_SIZE, "waitq structure size mismatch");
static_assert(__alignof(struct waitq) == WQ_OPAQUE_ALIGN, "waitq structure alignment mismatch");

/*
 *	struct waitq_set
 *
 *	This is the common definition for a set wait queue.
 */
struct waitq_set {
	struct waitq wqset_q;
	uint64_t     wqset_id;
	union {
		uint64_t     wqset_prepost_id;
		void        *wqset_prepost_hook;
	};
};

static_assert(sizeof(struct waitq_set) == WQS_OPAQUE_SIZE, "waitq_set structure size mismatch");
static_assert(__alignof(struct waitq_set) == WQS_OPAQUE_ALIGN, "waitq_set structure alignment mismatch");

extern void waitq_bootstrap(void);

#define waitq_is_queue(wq) \
	((wq)->waitq_type == WQT_QUEUE)

#define waitq_is_set(wq) \
	((wq)->waitq_type == WQT_SET && ((struct waitq_set *)(wq))->wqset_id != 0)

#define waitqs_is_set(wqs) \
	(((wqs)->wqset_q.waitq_type == WQT_SET) && ((wqs)->wqset_id != 0))

#define waitq_valid(wq) \
	((wq) != NULL && (wq)->waitq_isvalid && ((wq)->waitq_type & ~1) == WQT_QUEUE)

/*
 * Invalidate a waitq. The only valid waitq functions to call after this are:
 * 	waitq_deinit()
 * 	waitq_set_deinit()
 */
extern void waitq_invalidate_locked(struct waitq *wq);

#define waitq_empty(wq) \
	(queue_empty(&(wq)->waitq_queue))


#define waitq_held(wq) \
	(hw_lock_held(&(wq)->waitq_interlock))

#define waitq_lock_try(wq) \
	(hw_lock_try(&(wq)->waitq_interlock))


#define waitq_wait_possible(thread) \
	((thread)->waitq == NULL)

extern void waitq_lock(struct waitq *wq);
extern void waitq_unlock(struct waitq *wq);

#define waitq_set_lock(wqs)		waitq_lock(&(wqs)->wqset_q)
#define waitq_set_unlock(wqs)		waitq_unlock(&(wqs)->wqset_q)
#define waitq_set_lock_try(wqs)		waitq_lock_try(&(wqs)->wqset_q)
#define waitq_set_can_prepost(wqs)	(waitqs_is_set(wqs) && \
					 (wqs)->wqset_q.waitq_prepost)
#define waitq_set_maybe_preposted(wqs)	((wqs)->wqset_q.waitq_prepost && \
					 (wqs)->wqset_prepost_id > 0)
#define waitq_set_has_prepost_hook(wqs)	(waitqs_is_set(wqs) && \
					 !((wqs)->wqset_q.waitq_prepost) && \
					 (wqs)->wqset_prepost_hook)

/* assert intent to wait on a locked wait queue */
extern wait_result_t waitq_assert_wait64_locked(struct waitq *waitq,
						event64_t wait_event,
						wait_interrupt_t interruptible,
						wait_timeout_urgency_t urgency,
						uint64_t deadline,
						uint64_t leeway,
						thread_t thread);

/* pull a thread from its wait queue */
extern int waitq_pull_thread_locked(struct waitq *waitq, thread_t thread);

/* wakeup all threads waiting for a particular event on locked queue */
extern kern_return_t waitq_wakeup64_all_locked(struct waitq *waitq,
					       event64_t wake_event,
					       wait_result_t result,
					       uint64_t *reserved_preposts,
					       int priority,
					       waitq_lock_state_t lock_state);

/* wakeup one thread waiting for a particular event on locked queue */
extern kern_return_t waitq_wakeup64_one_locked(struct waitq *waitq,
					       event64_t wake_event,
					       wait_result_t result,
					       uint64_t *reserved_preposts,
					       int priority,
					       waitq_lock_state_t lock_state);

/* return identity of a thread awakened for a particular <wait_queue,event> */
extern thread_t
waitq_wakeup64_identify_locked(struct waitq     *waitq,
                               event64_t        wake_event,
                               wait_result_t    result,
                               spl_t            *spl,
                               uint64_t         *reserved_preposts,
                               int              priority,
                               waitq_lock_state_t lock_state);

/* wakeup thread iff its still waiting for a particular event on locked queue */
extern kern_return_t waitq_wakeup64_thread_locked(struct waitq *waitq,
						  event64_t wake_event,
						  thread_t thread,
						  wait_result_t result,
						  waitq_lock_state_t lock_state);

/* clear all preposts generated by the given waitq */
extern int waitq_clear_prepost_locked(struct waitq *waitq);

/* clear all preposts from the given wait queue set */
extern void waitq_set_clear_preposts_locked(struct waitq_set *wqset);

/* unlink the given waitq from all sets - returns unlocked */
extern kern_return_t waitq_unlink_all_unlock(struct waitq *waitq);

/* unlink the given waitq set from all waitqs and waitq sets - returns unlocked */
extern kern_return_t waitq_set_unlink_all_unlock(struct waitq_set *wqset);


/*
 * clear a thread's boosted priority
 * (given via WAITQ_PROMOTE_PRIORITY in the wakeup function)
 */
extern void waitq_clear_promotion_locked(struct waitq *waitq,
					 thread_t thread);

/*
 * waitq iteration
 */

enum waitq_iteration_constant {
	WQ_ITERATE_DROPPED             = -4,
	WQ_ITERATE_INVALID             = -3,
	WQ_ITERATE_ABORTED             = -2,
	WQ_ITERATE_FAILURE             = -1,
	WQ_ITERATE_SUCCESS             =  0,
	WQ_ITERATE_CONTINUE            =  1,
	WQ_ITERATE_BREAK               =  2,
	WQ_ITERATE_BREAK_KEEP_LOCKED   =  3,
	WQ_ITERATE_INVALIDATE_CONTINUE =  4,
	WQ_ITERATE_RESTART             =  5,
	WQ_ITERATE_FOUND               =  6,
	WQ_ITERATE_UNLINKED            =  7,
};

/* callback invoked with both 'waitq' and 'wqset' locked */
typedef int (*waitq_iterator_t)(void *ctx, struct waitq *waitq,
				struct waitq_set *wqset);

/* iterate over all sets to which waitq belongs */
extern int waitq_iterate_sets(struct waitq *waitq, void *ctx,
			      waitq_iterator_t it);

/* iterator over all waitqs that have preposted to wqset */
extern int waitq_set_iterate_preposts(struct waitq_set *wqset,
				      void *ctx, waitq_iterator_t it);

/*
 * prepost reservation
 */
extern uint64_t waitq_prepost_reserve(struct waitq *waitq, int extra,
				      waitq_lock_state_t lock_state);

extern void waitq_prepost_release_reserve(uint64_t id);

#else /* !MACH_KERNEL_PRIVATE */

/*
 * The opaque waitq structure is here mostly for AIO and selinfo,
 * but could potentially be used by other BSD subsystems.
 */
struct waitq { char opaque[WQ_OPAQUE_SIZE]; } __attribute__((aligned(WQ_OPAQUE_ALIGN)));
struct waitq_set { char opaque[WQS_OPAQUE_SIZE]; } __attribute__((aligned(WQS_OPAQUE_ALIGN)));

#endif	/* MACH_KERNEL_PRIVATE */


__BEGIN_DECLS

/*
 * waitq init
 */
extern kern_return_t waitq_init(struct waitq *waitq, int policy);
extern void waitq_deinit(struct waitq *waitq);

/*
 * global waitqs
 */
extern struct waitq *_global_eventq(char *event, size_t event_length);
#define global_eventq(event) _global_eventq((char *)&(event), sizeof(event))

extern struct waitq *global_waitq(int index);

/*
 * set alloc/init/free
 */
extern struct waitq_set *waitq_set_alloc(int policy, void *prepost_hook);

extern kern_return_t waitq_set_init(struct waitq_set *wqset,
				    int policy, uint64_t *reserved_link,
				    void *prepost_hook);

extern void waitq_set_deinit(struct waitq_set *wqset);

extern kern_return_t waitq_set_free(struct waitq_set *wqset);

#if defined(DEVELOPMENT) || defined(DEBUG)
#if CONFIG_WAITQ_DEBUG
extern uint64_t wqset_id(struct waitq_set *wqset);

struct waitq *wqset_waitq(struct waitq_set *wqset);
#endif /* CONFIG_WAITQ_DEBUG */
#endif /* DEVELOPMENT || DEBUG */


/*
 * set membership
 */
extern uint64_t waitq_link_reserve(struct waitq *waitq);

extern void waitq_link_release(uint64_t id);

extern boolean_t waitq_member(struct waitq *waitq, struct waitq_set *wqset);

/* returns true if the waitq is in at least 1 set */
extern boolean_t waitq_in_set(struct waitq *waitq);


/* on success, consumes an reserved_link reference */
extern kern_return_t waitq_link(struct waitq *waitq,
				struct waitq_set *wqset,
				waitq_lock_state_t lock_state,
				uint64_t *reserved_link);

extern kern_return_t waitq_unlink(struct waitq *waitq, struct waitq_set *wqset);

extern kern_return_t waitq_unlink_all(struct waitq *waitq);

extern kern_return_t waitq_set_unlink_all(struct waitq_set *wqset);

/*
 * preposts
 */
extern void waitq_clear_prepost(struct waitq *waitq);

extern void waitq_set_clear_preposts(struct waitq_set *wqset);

/*
 * interfaces used primarily by the select/kqueue subsystems
 */
extern uint64_t waitq_get_prepost_id(struct waitq *waitq);
extern void     waitq_unlink_by_prepost_id(uint64_t wqp_id, struct waitq_set *wqset);

/*
 * waitq attributes
 */
extern int waitq_is_valid(struct waitq *waitq);

extern int waitq_set_is_valid(struct waitq_set *wqset);

extern int waitq_is_global(struct waitq *waitq);

extern int waitq_irq_safe(struct waitq *waitq);

#if CONFIG_WAITQ_STATS
/*
 * waitq statistics
 */
#define WAITQ_STATS_VERSION 1
struct wq_table_stats {
	uint32_t version;
	uint32_t table_elements;
	uint32_t table_used_elems;
	uint32_t table_elem_sz;
	uint32_t table_slabs;
	uint32_t table_slab_sz;

	uint64_t table_num_allocs;
	uint64_t table_num_preposts;
	uint64_t table_num_reservations;

	uint64_t table_max_used;
	uint64_t table_avg_used;
	uint64_t table_max_reservations;
	uint64_t table_avg_reservations;
};

extern void waitq_link_stats(struct wq_table_stats *stats);
extern void waitq_prepost_stats(struct wq_table_stats *stats);
#endif /* CONFIG_WAITQ_STATS */

/*
 *
 * higher-level waiting APIs
 *
 */

/* assert intent to wait on <waitq,event64> pair */
extern wait_result_t waitq_assert_wait64(struct waitq *waitq,
					 event64_t wait_event,
					 wait_interrupt_t interruptible,
					 uint64_t deadline);

extern wait_result_t waitq_assert_wait64_leeway(struct waitq *waitq,
						event64_t wait_event,
						wait_interrupt_t interruptible,
						wait_timeout_urgency_t urgency,
						uint64_t deadline,
						uint64_t leeway);

/* wakeup the most appropriate thread waiting on <waitq,event64> pair */
extern kern_return_t waitq_wakeup64_one(struct waitq *waitq,
					event64_t wake_event,
					wait_result_t result,
					int priority);

/* wakeup all the threads waiting on <waitq,event64> pair */
extern kern_return_t waitq_wakeup64_all(struct waitq *waitq,
					event64_t wake_event,
					wait_result_t result,
					int priority);

#ifdef	XNU_KERNEL_PRIVATE

/* wakeup a specified thread iff it's waiting on <waitq,event64> pair */
extern kern_return_t waitq_wakeup64_thread(struct waitq *waitq,
					   event64_t wake_event,
					   thread_t thread,
					   wait_result_t result);

/* return a reference to the thread that was woken up */
extern thread_t
waitq_wakeup64_identify(struct waitq    *waitq,
                        event64_t       wake_event,
                        wait_result_t   result,
                        int             priority);

#endif /* XNU_KERNEL_PRIVATE */

__END_DECLS

#endif	/* KERNEL_PRIVATE */
#endif	/* _WAITQ_H_ */
Commit	Line	Data
3e170ce0 A	1	#ifndef _WAITQ_H_
	2	#define _WAITQ_H_
	3	/*
	4	* Copyright (c) 2014-2015 Apple Computer, Inc. All rights reserved.
	5	*
	6	* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
	7	*
	8	* This file contains Original Code and/or Modifications of Original Code
	9	* as defined in and that are subject to the Apple Public Source License
	10	* Version 2.0 (the 'License'). You may not use this file except in
	11	* compliance with the License. The rights granted to you under the License
	12	* may not be used to create, or enable the creation or redistribution of,
	13	* unlawful or unlicensed copies of an Apple operating system, or to
	14	* circumvent, violate, or enable the circumvention or violation of, any
	15	* terms of an Apple operating system software license agreement.
	16	*
	17	* Please obtain a copy of the License at
	18	* http://www.opensource.apple.com/apsl/ and read it before using this file.
	19	*
	20	* The Original Code and all software distributed under the License are
	21	* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
	22	* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
	23	* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
	24	* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
	25	* Please see the License for the specific language governing rights and
	26	* limitations under the License.
	27	*
	28	* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
	29	*/
	30	#ifdef KERNEL_PRIVATE
	31
	32	#include <mach/mach_types.h>
	33	#include <mach/sync_policy.h>
	34	#include <mach/kern_return.h> /* for kern_return_t */
	35
	36	#include <kern/kern_types.h> /* for wait_queue_t */
	37	#include <kern/queue.h>
	38	#include <kern/assert.h>
	39
	40	#include <sys/cdefs.h>
	41
	42	/*
	43	* Constants and types used in the waitq APIs
	44	*/
	45	#define WAITQ_ALL_PRIORITIES (-1)
	46	#define WAITQ_PROMOTE_PRIORITY (-2)
39037602	47	#define WAITQ_SELECT_MAX_PRI (-3)
3e170ce0 A	48
	49	typedef enum e_waitq_lock_state {
	50	WAITQ_KEEP_LOCKED = 0x01,
	51	WAITQ_UNLOCK = 0x02,
	52	WAITQ_SHOULD_LOCK = 0x04,
	53	WAITQ_ALREADY_LOCKED = 0x08,
	54	WAITQ_DONT_LOCK = 0x10,
	55	} waitq_lock_state_t;
	56
39037602 A	57	/*
	58	* The Jenkins "one at a time" hash.
	59	* TBD: There may be some value to unrolling here,
	60	* depending on the architecture.
	61	*/
	62	static __inline__ uint32_t
	63	jenkins_hash(char *key, size_t length)
	64	{
	65	uint32_t hash = 0;
	66	size_t i;
	67
	68	for (i = 0; i < length; i++) {
	69	hash += (uint32_t)key[i];
	70	hash += (hash << 10);
	71	hash ^= (hash >> 6);
	72	}
	73
	74	hash += (hash << 3);
	75	hash ^= (hash >> 11);
	76	hash += (hash << 15);
	77
	78	return hash;
	79	}
	80
	81	/* Opaque sizes and alignment used for struct verification */
	82	#if __x86_64__
	83	#define WQ_OPAQUE_ALIGN 8
	84	#define WQS_OPAQUE_ALIGN 8
	85	#define WQ_OPAQUE_SIZE 48
	86	#define WQS_OPAQUE_SIZE 64
	87	#else
	88	#error Unknown size requirement
	89	#endif
	90
813fb2f6	91	#ifdef MACH_KERNEL_PRIVATE
3e170ce0 A	92
	93	#include <kern/spl.h>
	94	#include <kern/simple_lock.h>
	95	#include <mach/branch_predicates.h>
	96
	97	#include <machine/cpu_number.h>
	98	#include <machine/machine_routines.h> /* machine_timeout_suspended() */
	99
	100	/*
	101	* The event mask is of 59 bits on 64 bit architeture and 27 bits on
	102	* 32 bit architecture and so we calculate its size using sizeof(long).
	103	* If the bitfield for wq_type and wq_fifo is changed, then value of
	104	* EVENT_MASK_BITS will also change.
	105	*
	106	* New plan: this is an optimization anyway, so I'm stealing 32bits
	107	* from the mask to shrink the waitq object even further.
	108	*/
39037602	109	#define _EVENT_MASK_BITS ((sizeof(uint32_t) * 8) - 6)
3e170ce0 A	110
	111	#define WAITQ_BOOST_PRIORITY 31
	112
	113	enum waitq_type {
	114	WQT_INVALID = 0,
	115	WQT_QUEUE = 0x2,
	116	WQT_SET = 0x3,
	117	};
	118
	119	#if CONFIG_WAITQ_STATS
	120	#define NWAITQ_BTFRAMES 5
	121	struct wq_stats {
	122	uint64_t waits;
	123	uint64_t wakeups;
	124	uint64_t clears;
	125	uint64_t failed_wakeups;
	126
	127	uintptr_t last_wait[NWAITQ_BTFRAMES];
	128	uintptr_t last_wakeup[NWAITQ_BTFRAMES];
	129	uintptr_t last_failed_wakeup[NWAITQ_BTFRAMES];
	130	};
	131	#endif
	132
	133	/*
	134	* struct waitq
	135	*
	136	* This is the definition of the common event wait queue
	137	* that the scheduler APIs understand. It is used
	138	* internally by the gerneralized event waiting mechanism
	139	* (assert_wait), and also for items that maintain their
	140	* own wait queues (such as ports and semaphores).
	141	*
	142	* It is not published to other kernel components.
	143	*
	144	* NOTE: Hardware locks are used to protect event wait
	145	* queues since interrupt code is free to post events to
	146	* them.
	147	*/
	148	struct waitq {
	149	uint32_t /* flags */
	150	waitq_type:2, /* only public field */
	151	waitq_fifo:1, /* fifo wakeup policy? */
	152	waitq_prepost:1, /* waitq supports prepost? */
	153	waitq_irq:1, /* waitq requires interrupts disabled */
39037602	154	waitq_isvalid:1, /* waitq structure is valid */
3e170ce0 A	155	waitq_eventmask:_EVENT_MASK_BITS;
	156	/* the wait queue set (set-of-sets) to which this queue belongs */
	157	hw_lock_data_t waitq_interlock; /* interlock */
	158
	159	uint64_t waitq_set_id;
	160	uint64_t waitq_prepost_id;
	161	queue_head_t waitq_queue; /* queue of elements */
	162	};
	163
39037602 A	164	static_assert(sizeof(struct waitq) == WQ_OPAQUE_SIZE, "waitq structure size mismatch");
	165	static_assert(__alignof(struct waitq) == WQ_OPAQUE_ALIGN, "waitq structure alignment mismatch");
	166
3e170ce0 A	167	/*
	168	* struct waitq_set
	169	*
	170	* This is the common definition for a set wait queue.
	171	*/
	172	struct waitq_set {
	173	struct waitq wqset_q;
	174	uint64_t wqset_id;
39037602 A	175	union {
	176	uint64_t wqset_prepost_id;
	177	void *wqset_prepost_hook;
	178	};
3e170ce0 A	179	};
3e170ce0 A	180
39037602 A	181	static_assert(sizeof(struct waitq_set) == WQS_OPAQUE_SIZE, "waitq_set structure size mismatch");
	182	static_assert(__alignof(struct waitq_set) == WQS_OPAQUE_ALIGN, "waitq_set structure alignment mismatch");
	183
3e170ce0 A	184	extern void waitq_bootstrap(void);
	185
	186	#define waitq_is_queue(wq) \
	187	((wq)->waitq_type == WQT_QUEUE)
	188
	189	#define waitq_is_set(wq) \
	190	((wq)->waitq_type == WQT_SET && ((struct waitq_set *)(wq))->wqset_id != 0)
	191
	192	#define waitqs_is_set(wqs) \
	193	(((wqs)->wqset_q.waitq_type == WQT_SET) && ((wqs)->wqset_id != 0))
	194
	195	#define waitq_valid(wq) \
39037602 A	196	((wq) != NULL && (wq)->waitq_isvalid && ((wq)->waitq_type & ~1) == WQT_QUEUE)
	197
	198	/*
	199	* Invalidate a waitq. The only valid waitq functions to call after this are:
	200	* waitq_deinit()
	201	* waitq_set_deinit()
	202	*/
	203	extern void waitq_invalidate_locked(struct waitq *wq);
3e170ce0 A	204
	205	#define waitq_empty(wq) \
	206	(queue_empty(&(wq)->waitq_queue))
	207
39037602	208
3e170ce0 A	209	#define waitq_held(wq) \
	210	(hw_lock_held(&(wq)->waitq_interlock))
	211
	212	#define waitq_lock_try(wq) \
	213	(hw_lock_try(&(wq)->waitq_interlock))
	214
39037602	215
3e170ce0 A	216	#define waitq_wait_possible(thread) \
	217	((thread)->waitq == NULL)
	218
	219	extern void waitq_lock(struct waitq *wq);
	220	extern void waitq_unlock(struct waitq *wq);
	221
	222	#define waitq_set_lock(wqs) waitq_lock(&(wqs)->wqset_q)
	223	#define waitq_set_unlock(wqs) waitq_unlock(&(wqs)->wqset_q)
	224	#define waitq_set_lock_try(wqs) waitq_lock_try(&(wqs)->wqset_q)
	225	#define waitq_set_can_prepost(wqs) (waitqs_is_set(wqs) && \
	226	(wqs)->wqset_q.waitq_prepost)
	227	#define waitq_set_maybe_preposted(wqs) ((wqs)->wqset_q.waitq_prepost && \
	228	(wqs)->wqset_prepost_id > 0)
39037602 A	229	#define waitq_set_has_prepost_hook(wqs) (waitqs_is_set(wqs) && \
	230	!((wqs)->wqset_q.waitq_prepost) && \
	231	(wqs)->wqset_prepost_hook)
3e170ce0 A	232
	233	/* assert intent to wait on a locked wait queue */
	234	extern wait_result_t waitq_assert_wait64_locked(struct waitq *waitq,
	235	event64_t wait_event,
	236	wait_interrupt_t interruptible,
	237	wait_timeout_urgency_t urgency,
	238	uint64_t deadline,
	239	uint64_t leeway,
	240	thread_t thread);
	241
	242	/* pull a thread from its wait queue */
39037602	243	extern int waitq_pull_thread_locked(struct waitq *waitq, thread_t thread);
3e170ce0 A	244
	245	/* wakeup all threads waiting for a particular event on locked queue */
	246	extern kern_return_t waitq_wakeup64_all_locked(struct waitq *waitq,
	247	event64_t wake_event,
	248	wait_result_t result,
	249	uint64_t *reserved_preposts,
	250	int priority,
	251	waitq_lock_state_t lock_state);
	252
	253	/* wakeup one thread waiting for a particular event on locked queue */
	254	extern kern_return_t waitq_wakeup64_one_locked(struct waitq *waitq,
	255	event64_t wake_event,
	256	wait_result_t result,
	257	uint64_t *reserved_preposts,
	258	int priority,
	259	waitq_lock_state_t lock_state);
	260
	261	/* return identity of a thread awakened for a particular <wait_queue,event> */
39037602 A	262	extern thread_t
	263	waitq_wakeup64_identify_locked(struct waitq *waitq,
	264	event64_t wake_event,
	265	wait_result_t result,
	266	spl_t *spl,
	267	uint64_t *reserved_preposts,
	268	int priority,
	269	waitq_lock_state_t lock_state);
3e170ce0 A	270
	271	/* wakeup thread iff its still waiting for a particular event on locked queue */
	272	extern kern_return_t waitq_wakeup64_thread_locked(struct waitq *waitq,
	273	event64_t wake_event,
	274	thread_t thread,
	275	wait_result_t result,
	276	waitq_lock_state_t lock_state);
	277
	278	/* clear all preposts generated by the given waitq */
39037602	279	extern int waitq_clear_prepost_locked(struct waitq *waitq);
3e170ce0 A	280
	281	/* clear all preposts from the given wait queue set */
	282	extern void waitq_set_clear_preposts_locked(struct waitq_set *wqset);
	283
39037602 A	284	/* unlink the given waitq from all sets - returns unlocked */
	285	extern kern_return_t waitq_unlink_all_unlock(struct waitq *waitq);
	286
	287	/* unlink the given waitq set from all waitqs and waitq sets - returns unlocked */
	288	extern kern_return_t waitq_set_unlink_all_unlock(struct waitq_set *wqset);
	289
	290
3e170ce0 A	291
	292	/*
	293	* clear a thread's boosted priority
	294	* (given via WAITQ_PROMOTE_PRIORITY in the wakeup function)
	295	*/
	296	extern void waitq_clear_promotion_locked(struct waitq *waitq,
	297	thread_t thread);
	298
	299	/*
	300	* waitq iteration
	301	*/
	302
	303	enum waitq_iteration_constant {
	304	WQ_ITERATE_DROPPED = -4,
	305	WQ_ITERATE_INVALID = -3,
	306	WQ_ITERATE_ABORTED = -2,
	307	WQ_ITERATE_FAILURE = -1,
	308	WQ_ITERATE_SUCCESS = 0,
	309	WQ_ITERATE_CONTINUE = 1,
	310	WQ_ITERATE_BREAK = 2,
	311	WQ_ITERATE_BREAK_KEEP_LOCKED = 3,
	312	WQ_ITERATE_INVALIDATE_CONTINUE = 4,
	313	WQ_ITERATE_RESTART = 5,
	314	WQ_ITERATE_FOUND = 6,
	315	WQ_ITERATE_UNLINKED = 7,
	316	};
	317
	318	/* callback invoked with both 'waitq' and 'wqset' locked */
	319	typedef int (waitq_iterator_t)(void ctx, struct waitq *waitq,
	320	struct waitq_set *wqset);
	321
	322	/* iterate over all sets to which waitq belongs */
	323	extern int waitq_iterate_sets(struct waitq waitq, void ctx,
	324	waitq_iterator_t it);
	325
	326	/* iterator over all waitqs that have preposted to wqset */
	327	extern int waitq_set_iterate_preposts(struct waitq_set *wqset,
39037602	328	void *ctx, waitq_iterator_t it);
3e170ce0 A	329
	330	/*
	331	* prepost reservation
	332	*/
	333	extern uint64_t waitq_prepost_reserve(struct waitq *waitq, int extra,
39037602	334	waitq_lock_state_t lock_state);
3e170ce0 A	335
	336	extern void waitq_prepost_release_reserve(uint64_t id);
	337
813fb2f6 A	338	#else /* !MACH_KERNEL_PRIVATE */
	339
	340	/*
	341	* The opaque waitq structure is here mostly for AIO and selinfo,
	342	* but could potentially be used by other BSD subsystems.
	343	*/
	344	struct waitq { char opaque[WQ_OPAQUE_SIZE]; } __attribute__((aligned(WQ_OPAQUE_ALIGN)));
	345	struct waitq_set { char opaque[WQS_OPAQUE_SIZE]; } __attribute__((aligned(WQS_OPAQUE_ALIGN)));
	346
3e170ce0 A	347	#endif /* MACH_KERNEL_PRIVATE */
	348
	349
	350	__BEGIN_DECLS
	351
	352	/*
	353	* waitq init
	354	*/
	355	extern kern_return_t waitq_init(struct waitq *waitq, int policy);
	356	extern void waitq_deinit(struct waitq *waitq);
	357
	358	/*
	359	* global waitqs
	360	*/
	361	extern struct waitq _global_eventq(char event, size_t event_length);
	362	#define global_eventq(event) _global_eventq((char *)&(event), sizeof(event))
	363
	364	extern struct waitq *global_waitq(int index);
	365
	366	/*
	367	* set alloc/init/free
	368	*/
39037602	369	extern struct waitq_set waitq_set_alloc(int policy, void prepost_hook);
3e170ce0 A	370
3e170ce0 A	371	extern kern_return_t waitq_set_init(struct waitq_set *wqset,
39037602 A	372	int policy, uint64_t *reserved_link,
39037602 A	373	void *prepost_hook);
3e170ce0 A	374
	375	extern void waitq_set_deinit(struct waitq_set *wqset);
	376
	377	extern kern_return_t waitq_set_free(struct waitq_set *wqset);
	378
	379	#if defined(DEVELOPMENT) \|\| defined(DEBUG)
	380	#if CONFIG_WAITQ_DEBUG
	381	extern uint64_t wqset_id(struct waitq_set *wqset);
	382
	383	struct waitq wqset_waitq(struct waitq_set wqset);
	384	#endif /* CONFIG_WAITQ_DEBUG */
	385	#endif /* DEVELOPMENT \|\| DEBUG */
	386
	387
	388	/*
	389	* set membership
	390	*/
	391	extern uint64_t waitq_link_reserve(struct waitq *waitq);
	392
	393	extern void waitq_link_release(uint64_t id);
	394
	395	extern boolean_t waitq_member(struct waitq waitq, struct waitq_set wqset);
	396
	397	/* returns true if the waitq is in at least 1 set */
	398	extern boolean_t waitq_in_set(struct waitq *waitq);
	399
	400
	401	/* on success, consumes an reserved_link reference */
	402	extern kern_return_t waitq_link(struct waitq *waitq,
	403	struct waitq_set *wqset,
	404	waitq_lock_state_t lock_state,
	405	uint64_t *reserved_link);
	406
	407	extern kern_return_t waitq_unlink(struct waitq waitq, struct waitq_set wqset);
	408
	409	extern kern_return_t waitq_unlink_all(struct waitq *waitq);
	410
	411	extern kern_return_t waitq_set_unlink_all(struct waitq_set *wqset);
	412
3e170ce0 A	413	/*
	414	* preposts
	415	*/
	416	extern void waitq_clear_prepost(struct waitq *waitq);
	417
	418	extern void waitq_set_clear_preposts(struct waitq_set *wqset);
	419
	420	/*
	421	* interfaces used primarily by the select/kqueue subsystems
	422	*/
	423	extern uint64_t waitq_get_prepost_id(struct waitq *waitq);
	424	extern void waitq_unlink_by_prepost_id(uint64_t wqp_id, struct waitq_set *wqset);
	425
	426	/*
	427	* waitq attributes
	428	*/
	429	extern int waitq_is_valid(struct waitq *waitq);
	430
	431	extern int waitq_set_is_valid(struct waitq_set *wqset);
	432
	433	extern int waitq_is_global(struct waitq *waitq);
	434
	435	extern int waitq_irq_safe(struct waitq *waitq);
	436
	437	#if CONFIG_WAITQ_STATS
	438	/*
	439	* waitq statistics
	440	*/
	441	#define WAITQ_STATS_VERSION 1
	442	struct wq_table_stats {
	443	uint32_t version;
	444	uint32_t table_elements;
	445	uint32_t table_used_elems;
	446	uint32_t table_elem_sz;
	447	uint32_t table_slabs;
	448	uint32_t table_slab_sz;
	449
	450	uint64_t table_num_allocs;
	451	uint64_t table_num_preposts;
	452	uint64_t table_num_reservations;
	453
	454	uint64_t table_max_used;
	455	uint64_t table_avg_used;
	456	uint64_t table_max_reservations;
	457	uint64_t table_avg_reservations;
	458	};
	459
	460	extern void waitq_link_stats(struct wq_table_stats *stats);
	461	extern void waitq_prepost_stats(struct wq_table_stats *stats);
	462	#endif /* CONFIG_WAITQ_STATS */
	463
	464	/*
	465	*
	466	* higher-level waiting APIs
	467	*
	468	*/
	469
	470	/* assert intent to wait on <waitq,event64> pair */
	471	extern wait_result_t waitq_assert_wait64(struct waitq *waitq,
	472	event64_t wait_event,
	473	wait_interrupt_t interruptible,
	474	uint64_t deadline);
	475
	476	extern wait_result_t waitq_assert_wait64_leeway(struct waitq *waitq,
477	event64_t wait_event,
478	wait_interrupt_t interruptible,
479	wait_timeout_urgency_t urgency,
480	uint64_t deadline,
481	uint64_t leeway);
482
483	/* wakeup the most appropriate thread waiting on <waitq,event64> pair */
484	extern kern_return_t waitq_wakeup64_one(struct waitq *waitq,
485	event64_t wake_event,
486	wait_result_t result,
487	int priority);
488
489	/* wakeup all the threads waiting on <waitq,event64> pair */
490	extern kern_return_t waitq_wakeup64_all(struct waitq *waitq,
491	event64_t wake_event,
492	wait_result_t result,
493	int priority);
494
39037602 A	495	#ifdef XNU_KERNEL_PRIVATE
39037602 A	496
3e170ce0 A	497	/* wakeup a specified thread iff it's waiting on <waitq,event64> pair */
	498	extern kern_return_t waitq_wakeup64_thread(struct waitq *waitq,
	499	event64_t wake_event,
	500	thread_t thread,
	501	wait_result_t result);
39037602 A	502
	503	/* return a reference to the thread that was woken up */
	504	extern thread_t
	505	waitq_wakeup64_identify(struct waitq *waitq,
	506	event64_t wake_event,
	507	wait_result_t result,
	508	int priority);
	509
	510	#endif /* XNU_KERNEL_PRIVATE */
	511
3e170ce0 A	512	__END_DECLS
	513
	514	#endif /* KERNEL_PRIVATE */
	515	#endif /* _WAITQ_H_ */