[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>

#ifdef XNU_KERNEL_PRIVATE
/* priority queue static asserts fail for __ARM64_ARCH_8_32__ kext builds */
#include <kern/priority_queue.h>
#endif /* XNU_KERNEL_PRIVATE */

/*
 * 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 __arm__ || __arm64__
	#define WQ_OPAQUE_ALIGN   __BIGGEST_ALIGNMENT__
	#define WQS_OPAQUE_ALIGN  __BIGGEST_ALIGNMENT__
	#if __arm__
		#define WQ_OPAQUE_SIZE   32
		#define WQS_OPAQUE_SIZE  48
	#else
		#define WQ_OPAQUE_SIZE   40
		#define WQS_OPAQUE_SIZE  56
	#endif
#elif __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 <machine/cpu_number.h>
#include <machine/machine_routines.h> /* machine_timeout_suspended() */

/*
 * The event mask is of 57 bits on 64 bit architeture and 25 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) - 7)


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_turnstile_or_port:1, /* waitq is embedded in a turnstile (if irq safe), or port (if not irq safe) */
		waitq_eventmask:_EVENT_MASK_BITS;
		/* the wait queue set (set-of-sets) to which this queue belongs */
#if __arm64__
	hw_lock_bit_t	waitq_interlock;	/* interlock */
#else
	hw_lock_data_t	waitq_interlock;	/* interlock */
#endif /* __arm64__ */

	uint64_t waitq_set_id;
	uint64_t waitq_prepost_id;
	union {
		queue_head_t            waitq_queue;		/* queue of elements */
		struct priority_queue   waitq_prio_queue;	/* priority ordered 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;
	};
};

#define WQSET_NOT_LINKED ((uint64_t)(~0))
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_turnstile_queue(wq) \
	(((wq)->waitq_irq) && (wq)->waitq_turnstile_or_port)

#define waitq_is_port_queue(wq) \
	(!((wq)->waitq_irq) && (wq)->waitq_turnstile_or_port)

#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)

#define waitqs_is_linked(wqs) \
	(((wqs)->wqset_id != WQSET_NOT_LINKED) && ((wqs)->wqset_id != 0))

/*
 * 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);

static inline boolean_t waitq_empty(struct waitq *wq)
{
	if (waitq_is_turnstile_queue(wq)) {
		return priority_queue_empty(&(wq->waitq_prio_queue));
	} else {
		return queue_empty(&(wq->waitq_queue));
	}
}

#if __arm64__

#define waitq_held(wq) \
	(hw_lock_bit_held(&(wq)->waitq_interlock, LCK_ILOCK))

#define waitq_lock_try(wq) \
	(hw_lock_bit_try(&(wq)->waitq_interlock, LCK_ILOCK))

#else

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

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

#endif /* __arm64__ */

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

extern void waitq_lock(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 DEVELOPMENT || DEBUG
extern int sysctl_helper_waitq_set_nelem(void);
#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_set_lazy_init_link(struct waitq_set *wqset);
extern boolean_t waitq_set_should_lazy_init_link(struct waitq_set *wqset);

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);
extern struct waitq *waitq_lock_by_prepost_id(uint64_t wqp_id);

/*
 * 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);

extern struct waitq * waitq_get_safeq(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);

/* take the waitq lock */
extern void waitq_unlock(struct waitq *wq);

#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
d9a64523 A	42	#ifdef XNU_KERNEL_PRIVATE
	43	/* priority queue static asserts fail for __ARM64_ARCH_8_32__ kext builds */
	44	#include <kern/priority_queue.h>
	45	#endif /* XNU_KERNEL_PRIVATE */
	46
3e170ce0 A	47	/*
	48	* Constants and types used in the waitq APIs
	49	*/
	50	#define WAITQ_ALL_PRIORITIES (-1)
	51	#define WAITQ_PROMOTE_PRIORITY (-2)
39037602	52	#define WAITQ_SELECT_MAX_PRI (-3)
3e170ce0 A	53
	54	typedef enum e_waitq_lock_state {
	55	WAITQ_KEEP_LOCKED = 0x01,
	56	WAITQ_UNLOCK = 0x02,
	57	WAITQ_SHOULD_LOCK = 0x04,
	58	WAITQ_ALREADY_LOCKED = 0x08,
	59	WAITQ_DONT_LOCK = 0x10,
	60	} waitq_lock_state_t;
	61
39037602 A	62	/*
	63	* The Jenkins "one at a time" hash.
	64	* TBD: There may be some value to unrolling here,
	65	* depending on the architecture.
	66	*/
	67	static __inline__ uint32_t
	68	jenkins_hash(char *key, size_t length)
	69	{
	70	uint32_t hash = 0;
	71	size_t i;
	72
	73	for (i = 0; i < length; i++) {
	74	hash += (uint32_t)key[i];
	75	hash += (hash << 10);
	76	hash ^= (hash >> 6);
	77	}
	78
	79	hash += (hash << 3);
	80	hash ^= (hash >> 11);
	81	hash += (hash << 15);
	82
	83	return hash;
	84	}
	85
	86	/* Opaque sizes and alignment used for struct verification */
5ba3f43e A	87	#if __arm__ \|\| __arm64__
	88	#define WQ_OPAQUE_ALIGN __BIGGEST_ALIGNMENT__
	89	#define WQS_OPAQUE_ALIGN __BIGGEST_ALIGNMENT__
	90	#if __arm__
	91	#define WQ_OPAQUE_SIZE 32
	92	#define WQS_OPAQUE_SIZE 48
	93	#else
	94	#define WQ_OPAQUE_SIZE 40
	95	#define WQS_OPAQUE_SIZE 56
	96	#endif
	97	#elif __x86_64__
39037602 A	98	#define WQ_OPAQUE_ALIGN 8
	99	#define WQS_OPAQUE_ALIGN 8
	100	#define WQ_OPAQUE_SIZE 48
	101	#define WQS_OPAQUE_SIZE 64
	102	#else
	103	#error Unknown size requirement
	104	#endif
	105
813fb2f6	106	#ifdef MACH_KERNEL_PRIVATE
3e170ce0 A	107
	108	#include <kern/spl.h>
	109	#include <kern/simple_lock.h>
3e170ce0 A	110
	111	#include <machine/cpu_number.h>
	112	#include <machine/machine_routines.h> /* machine_timeout_suspended() */
	113
	114	/*
d9a64523	115	* The event mask is of 57 bits on 64 bit architeture and 25 bits on
3e170ce0 A	116	* 32 bit architecture and so we calculate its size using sizeof(long).
	117	* If the bitfield for wq_type and wq_fifo is changed, then value of
	118	* EVENT_MASK_BITS will also change.
	119	*
	120	* New plan: this is an optimization anyway, so I'm stealing 32bits
	121	* from the mask to shrink the waitq object even further.
	122	*/
d9a64523	123	#define _EVENT_MASK_BITS ((sizeof(uint32_t) * 8) - 7)
3e170ce0	124
3e170ce0 A	125
	126	enum waitq_type {
	127	WQT_INVALID = 0,
	128	WQT_QUEUE = 0x2,
	129	WQT_SET = 0x3,
	130	};
	131
	132	#if CONFIG_WAITQ_STATS
	133	#define NWAITQ_BTFRAMES 5
	134	struct wq_stats {
	135	uint64_t waits;
	136	uint64_t wakeups;
	137	uint64_t clears;
	138	uint64_t failed_wakeups;
	139
	140	uintptr_t last_wait[NWAITQ_BTFRAMES];
	141	uintptr_t last_wakeup[NWAITQ_BTFRAMES];
	142	uintptr_t last_failed_wakeup[NWAITQ_BTFRAMES];
	143	};
	144	#endif
	145
	146	/*
	147	* struct waitq
	148	*
	149	* This is the definition of the common event wait queue
	150	* that the scheduler APIs understand. It is used
	151	* internally by the gerneralized event waiting mechanism
	152	* (assert_wait), and also for items that maintain their
	153	* own wait queues (such as ports and semaphores).
	154	*
	155	* It is not published to other kernel components.
	156	*
	157	* NOTE: Hardware locks are used to protect event wait
	158	* queues since interrupt code is free to post events to
	159	* them.
	160	*/
	161	struct waitq {
	162	uint32_t /* flags */
	163	waitq_type:2, /* only public field */
	164	waitq_fifo:1, /* fifo wakeup policy? */
	165	waitq_prepost:1, /* waitq supports prepost? */
	166	waitq_irq:1, /* waitq requires interrupts disabled */
39037602	167	waitq_isvalid:1, /* waitq structure is valid */
d9a64523	168	waitq_turnstile_or_port:1, /* waitq is embedded in a turnstile (if irq safe), or port (if not irq safe) */
3e170ce0 A	169	waitq_eventmask:_EVENT_MASK_BITS;
3e170ce0 A	170	/* the wait queue set (set-of-sets) to which this queue belongs */
5ba3f43e A	171	#if __arm64__
	172	hw_lock_bit_t waitq_interlock; /* interlock */
	173	#else
3e170ce0	174	hw_lock_data_t waitq_interlock; /* interlock */
5ba3f43e	175	#endif /* __arm64__ */
3e170ce0 A	176
	177	uint64_t waitq_set_id;
	178	uint64_t waitq_prepost_id;
d9a64523 A	179	union {
	180	queue_head_t waitq_queue; /* queue of elements */
	181	struct priority_queue waitq_prio_queue; /* priority ordered queue of elements */
	182	};
3e170ce0 A	183	};
3e170ce0 A	184
39037602 A	185	static_assert(sizeof(struct waitq) == WQ_OPAQUE_SIZE, "waitq structure size mismatch");
	186	static_assert(__alignof(struct waitq) == WQ_OPAQUE_ALIGN, "waitq structure alignment mismatch");
	187
3e170ce0 A	188	/*
	189	* struct waitq_set
	190	*
	191	* This is the common definition for a set wait queue.
	192	*/
	193	struct waitq_set {
	194	struct waitq wqset_q;
	195	uint64_t wqset_id;
39037602 A	196	union {
	197	uint64_t wqset_prepost_id;
	198	void *wqset_prepost_hook;
	199	};
3e170ce0 A	200	};
3e170ce0 A	201
d9a64523	202	#define WQSET_NOT_LINKED ((uint64_t)(~0))
39037602 A	203	static_assert(sizeof(struct waitq_set) == WQS_OPAQUE_SIZE, "waitq_set structure size mismatch");
	204	static_assert(__alignof(struct waitq_set) == WQS_OPAQUE_ALIGN, "waitq_set structure alignment mismatch");
	205
3e170ce0 A	206	extern void waitq_bootstrap(void);
	207
	208	#define waitq_is_queue(wq) \
	209	((wq)->waitq_type == WQT_QUEUE)
	210
d9a64523 A	211	#define waitq_is_turnstile_queue(wq) \
	212	(((wq)->waitq_irq) && (wq)->waitq_turnstile_or_port)
	213
	214	#define waitq_is_port_queue(wq) \
	215	(!((wq)->waitq_irq) && (wq)->waitq_turnstile_or_port)
	216
3e170ce0 A	217	#define waitq_is_set(wq) \
	218	((wq)->waitq_type == WQT_SET && ((struct waitq_set *)(wq))->wqset_id != 0)
	219
	220	#define waitqs_is_set(wqs) \
	221	(((wqs)->wqset_q.waitq_type == WQT_SET) && ((wqs)->wqset_id != 0))
	222
	223	#define waitq_valid(wq) \
d9a64523 A	224	((wq) != NULL && (wq)->waitq_isvalid)
	225
	226	#define waitqs_is_linked(wqs) \
	227	(((wqs)->wqset_id != WQSET_NOT_LINKED) && ((wqs)->wqset_id != 0))
39037602 A	228
	229	/*
	230	* Invalidate a waitq. The only valid waitq functions to call after this are:
	231	* waitq_deinit()
	232	* waitq_set_deinit()
	233	*/
	234	extern void waitq_invalidate_locked(struct waitq *wq);
3e170ce0	235
d9a64523 A	236	static inline boolean_t waitq_empty(struct waitq *wq)
	237	{
	238	if (waitq_is_turnstile_queue(wq)) {
	239	return priority_queue_empty(&(wq->waitq_prio_queue));
	240	} else {
	241	return queue_empty(&(wq->waitq_queue));
	242	}
	243	}
3e170ce0	244
5ba3f43e A	245	#if __arm64__
	246
	247	#define waitq_held(wq) \
	248	(hw_lock_bit_held(&(wq)->waitq_interlock, LCK_ILOCK))
	249
	250	#define waitq_lock_try(wq) \
	251	(hw_lock_bit_try(&(wq)->waitq_interlock, LCK_ILOCK))
	252
	253	#else
39037602	254
3e170ce0 A	255	#define waitq_held(wq) \
	256	(hw_lock_held(&(wq)->waitq_interlock))
	257
	258	#define waitq_lock_try(wq) \
	259	(hw_lock_try(&(wq)->waitq_interlock))
	260
5ba3f43e	261	#endif /* __arm64__ */
39037602	262
3e170ce0 A	263	#define waitq_wait_possible(thread) \
	264	((thread)->waitq == NULL)
	265
	266	extern void waitq_lock(struct waitq *wq);
3e170ce0 A	267
	268	#define waitq_set_lock(wqs) waitq_lock(&(wqs)->wqset_q)
	269	#define waitq_set_unlock(wqs) waitq_unlock(&(wqs)->wqset_q)
	270	#define waitq_set_lock_try(wqs) waitq_lock_try(&(wqs)->wqset_q)
	271	#define waitq_set_can_prepost(wqs) (waitqs_is_set(wqs) && \
	272	(wqs)->wqset_q.waitq_prepost)
	273	#define waitq_set_maybe_preposted(wqs) ((wqs)->wqset_q.waitq_prepost && \
	274	(wqs)->wqset_prepost_id > 0)
39037602 A	275	#define waitq_set_has_prepost_hook(wqs) (waitqs_is_set(wqs) && \
	276	!((wqs)->wqset_q.waitq_prepost) && \
	277	(wqs)->wqset_prepost_hook)
3e170ce0 A	278
	279	/* assert intent to wait on a locked wait queue */
	280	extern wait_result_t waitq_assert_wait64_locked(struct waitq *waitq,
	281	event64_t wait_event,
	282	wait_interrupt_t interruptible,
	283	wait_timeout_urgency_t urgency,
	284	uint64_t deadline,
	285	uint64_t leeway,
	286	thread_t thread);
	287
	288	/* pull a thread from its wait queue */
39037602	289	extern int waitq_pull_thread_locked(struct waitq *waitq, thread_t thread);
3e170ce0 A	290
	291	/* wakeup all threads waiting for a particular event on locked queue */
	292	extern kern_return_t waitq_wakeup64_all_locked(struct waitq *waitq,
	293	event64_t wake_event,
	294	wait_result_t result,
	295	uint64_t *reserved_preposts,
	296	int priority,
	297	waitq_lock_state_t lock_state);
	298
	299	/* wakeup one thread waiting for a particular event on locked queue */
	300	extern kern_return_t waitq_wakeup64_one_locked(struct waitq *waitq,
	301	event64_t wake_event,
	302	wait_result_t result,
	303	uint64_t *reserved_preposts,
	304	int priority,
	305	waitq_lock_state_t lock_state);
	306
	307	/* return identity of a thread awakened for a particular <wait_queue,event> */
39037602 A	308	extern thread_t
	309	waitq_wakeup64_identify_locked(struct waitq *waitq,
	310	event64_t wake_event,
	311	wait_result_t result,
	312	spl_t *spl,
	313	uint64_t *reserved_preposts,
	314	int priority,
	315	waitq_lock_state_t lock_state);
3e170ce0 A	316
	317	/* wakeup thread iff its still waiting for a particular event on locked queue */
	318	extern kern_return_t waitq_wakeup64_thread_locked(struct waitq *waitq,
	319	event64_t wake_event,
	320	thread_t thread,
	321	wait_result_t result,
	322	waitq_lock_state_t lock_state);
	323
	324	/* clear all preposts generated by the given waitq */
39037602	325	extern int waitq_clear_prepost_locked(struct waitq *waitq);
3e170ce0 A	326
	327	/* clear all preposts from the given wait queue set */
	328	extern void waitq_set_clear_preposts_locked(struct waitq_set *wqset);
	329
39037602 A	330	/* unlink the given waitq from all sets - returns unlocked */
	331	extern kern_return_t waitq_unlink_all_unlock(struct waitq *waitq);
	332
	333	/* unlink the given waitq set from all waitqs and waitq sets - returns unlocked */
	334	extern kern_return_t waitq_set_unlink_all_unlock(struct waitq_set *wqset);
	335
	336
3e170ce0 A	337
	338	/*
	339	* clear a thread's boosted priority
	340	* (given via WAITQ_PROMOTE_PRIORITY in the wakeup function)
	341	*/
	342	extern void waitq_clear_promotion_locked(struct waitq *waitq,
	343	thread_t thread);
	344
	345	/*
	346	* waitq iteration
	347	*/
	348
	349	enum waitq_iteration_constant {
	350	WQ_ITERATE_DROPPED = -4,
	351	WQ_ITERATE_INVALID = -3,
	352	WQ_ITERATE_ABORTED = -2,
	353	WQ_ITERATE_FAILURE = -1,
	354	WQ_ITERATE_SUCCESS = 0,
	355	WQ_ITERATE_CONTINUE = 1,
	356	WQ_ITERATE_BREAK = 2,
	357	WQ_ITERATE_BREAK_KEEP_LOCKED = 3,
	358	WQ_ITERATE_INVALIDATE_CONTINUE = 4,
	359	WQ_ITERATE_RESTART = 5,
	360	WQ_ITERATE_FOUND = 6,
	361	WQ_ITERATE_UNLINKED = 7,
	362	};
	363
	364	/* callback invoked with both 'waitq' and 'wqset' locked */
	365	typedef int (waitq_iterator_t)(void ctx, struct waitq *waitq,
	366	struct waitq_set *wqset);
	367
	368	/* iterate over all sets to which waitq belongs */
	369	extern int waitq_iterate_sets(struct waitq waitq, void ctx,
	370	waitq_iterator_t it);
	371
	372	/* iterator over all waitqs that have preposted to wqset */
	373	extern int waitq_set_iterate_preposts(struct waitq_set *wqset,
39037602	374	void *ctx, waitq_iterator_t it);
3e170ce0 A	375
	376	/*
	377	* prepost reservation
	378	*/
	379	extern uint64_t waitq_prepost_reserve(struct waitq *waitq, int extra,
39037602	380	waitq_lock_state_t lock_state);
3e170ce0 A	381
	382	extern void waitq_prepost_release_reserve(uint64_t id);
	383
813fb2f6 A	384	#else /* !MACH_KERNEL_PRIVATE */
	385
	386	/*
	387	* The opaque waitq structure is here mostly for AIO and selinfo,
	388	* but could potentially be used by other BSD subsystems.
	389	*/
	390	struct waitq { char opaque[WQ_OPAQUE_SIZE]; } __attribute__((aligned(WQ_OPAQUE_ALIGN)));
	391	struct waitq_set { char opaque[WQS_OPAQUE_SIZE]; } __attribute__((aligned(WQS_OPAQUE_ALIGN)));
	392
3e170ce0 A	393	#endif /* MACH_KERNEL_PRIVATE */
	394
	395
	396	__BEGIN_DECLS
	397
	398	/*
	399	* waitq init
	400	*/
	401	extern kern_return_t waitq_init(struct waitq *waitq, int policy);
	402	extern void waitq_deinit(struct waitq *waitq);
	403
	404	/*
	405	* global waitqs
	406	*/
	407	extern struct waitq _global_eventq(char event, size_t event_length);
	408	#define global_eventq(event) _global_eventq((char *)&(event), sizeof(event))
	409
	410	extern struct waitq *global_waitq(int index);
	411
	412	/*
	413	* set alloc/init/free
	414	*/
39037602	415	extern struct waitq_set waitq_set_alloc(int policy, void prepost_hook);
3e170ce0 A	416
3e170ce0 A	417	extern kern_return_t waitq_set_init(struct waitq_set *wqset,
39037602 A	418	int policy, uint64_t *reserved_link,
39037602 A	419	void *prepost_hook);
3e170ce0 A	420
	421	extern void waitq_set_deinit(struct waitq_set *wqset);
	422
	423	extern kern_return_t waitq_set_free(struct waitq_set *wqset);
	424
5ba3f43e	425	#if DEVELOPMENT \|\| DEBUG
d9a64523	426	extern int sysctl_helper_waitq_set_nelem(void);
3e170ce0 A	427	#if CONFIG_WAITQ_DEBUG
	428	extern uint64_t wqset_id(struct waitq_set *wqset);
	429
	430	struct waitq wqset_waitq(struct waitq_set wqset);
	431	#endif /* CONFIG_WAITQ_DEBUG */
	432	#endif /* DEVELOPMENT \|\| DEBUG */
	433
	434
	435	/*
	436	* set membership
	437	*/
	438	extern uint64_t waitq_link_reserve(struct waitq *waitq);
d9a64523 A	439	extern void waitq_set_lazy_init_link(struct waitq_set *wqset);
d9a64523 A	440	extern boolean_t waitq_set_should_lazy_init_link(struct waitq_set *wqset);
3e170ce0 A	441
	442	extern void waitq_link_release(uint64_t id);
	443
	444	extern boolean_t waitq_member(struct waitq waitq, struct waitq_set wqset);
	445
	446	/* returns true if the waitq is in at least 1 set */
	447	extern boolean_t waitq_in_set(struct waitq *waitq);
	448
	449
	450	/* on success, consumes an reserved_link reference */
	451	extern kern_return_t waitq_link(struct waitq *waitq,
	452	struct waitq_set *wqset,
	453	waitq_lock_state_t lock_state,
	454	uint64_t *reserved_link);
	455
	456	extern kern_return_t waitq_unlink(struct waitq waitq, struct waitq_set wqset);
	457
	458	extern kern_return_t waitq_unlink_all(struct waitq *waitq);
	459
	460	extern kern_return_t waitq_set_unlink_all(struct waitq_set *wqset);
	461
3e170ce0 A	462	/*
	463	* preposts
	464	*/
	465	extern void waitq_clear_prepost(struct waitq *waitq);
	466
	467	extern void waitq_set_clear_preposts(struct waitq_set *wqset);
	468
	469	/*
	470	* interfaces used primarily by the select/kqueue subsystems
	471	*/
	472	extern uint64_t waitq_get_prepost_id(struct waitq *waitq);
	473	extern void waitq_unlink_by_prepost_id(uint64_t wqp_id, struct waitq_set *wqset);
5ba3f43e	474	extern struct waitq *waitq_lock_by_prepost_id(uint64_t wqp_id);
3e170ce0 A	475
	476	/*
	477	* waitq attributes
	478	*/
	479	extern int waitq_is_valid(struct waitq *waitq);
	480
	481	extern int waitq_set_is_valid(struct waitq_set *wqset);
	482
	483	extern int waitq_is_global(struct waitq *waitq);
	484
	485	extern int waitq_irq_safe(struct waitq *waitq);
	486
d9a64523 A	487	extern struct waitq * waitq_get_safeq(struct waitq *waitq);
d9a64523 A	488
3e170ce0 A	489	#if CONFIG_WAITQ_STATS
	490	/*
	491	* waitq statistics
	492	*/
	493	#define WAITQ_STATS_VERSION 1
	494	struct wq_table_stats {
	495	uint32_t version;
	496	uint32_t table_elements;
	497	uint32_t table_used_elems;
	498	uint32_t table_elem_sz;
	499	uint32_t table_slabs;
	500	uint32_t table_slab_sz;
	501
	502	uint64_t table_num_allocs;
	503	uint64_t table_num_preposts;
	504	uint64_t table_num_reservations;
	505
	506	uint64_t table_max_used;
	507	uint64_t table_avg_used;
	508	uint64_t table_max_reservations;
	509	uint64_t table_avg_reservations;
	510	};
	511
	512	extern void waitq_link_stats(struct wq_table_stats *stats);
	513	extern void waitq_prepost_stats(struct wq_table_stats *stats);
	514	#endif /* CONFIG_WAITQ_STATS */
	515
	516	/*
	517	*
	518	* higher-level waiting APIs
	519	*
	520	*/
	521
	522	/* assert intent to wait on <waitq,event64> pair */
	523	extern wait_result_t waitq_assert_wait64(struct waitq *waitq,
	524	event64_t wait_event,
	525	wait_interrupt_t interruptible,
	526	uint64_t deadline);
	527
	528	extern wait_result_t waitq_assert_wait64_leeway(struct waitq *waitq,
	529	event64_t wait_event,
	530	wait_interrupt_t interruptible,
	531	wait_timeout_urgency_t urgency,
	532	uint64_t deadline,
	533	uint64_t leeway);
	534
	535	/* wakeup the most appropriate thread waiting on <waitq,event64> pair */
	536	extern kern_return_t waitq_wakeup64_one(struct waitq *waitq,
	537	event64_t wake_event,
	538	wait_result_t result,
	539	int priority);
	540
	541	/* wakeup all the threads waiting on <waitq,event64> pair */
	542	extern kern_return_t waitq_wakeup64_all(struct waitq *waitq,
	543	event64_t wake_event,
	544	wait_result_t result,
	545	int priority);
	546
39037602 A	547	#ifdef XNU_KERNEL_PRIVATE
39037602 A	548
3e170ce0 A	549	/* wakeup a specified thread iff it's waiting on <waitq,event64> pair */
	550	extern kern_return_t waitq_wakeup64_thread(struct waitq *waitq,
	551	event64_t wake_event,
	552	thread_t thread,
	553	wait_result_t result);
39037602 A	554
	555	/* return a reference to the thread that was woken up */
	556	extern thread_t
	557	waitq_wakeup64_identify(struct waitq *waitq,
	558	event64_t wake_event,
	559	wait_result_t result,
	560	int priority);
	561
5ba3f43e A	562	/* take the waitq lock */
	563	extern void waitq_unlock(struct waitq *wq);
	564
39037602 A	565	#endif /* XNU_KERNEL_PRIVATE */
39037602 A	566
3e170ce0 A	567	__END_DECLS
	568
	569	#endif /* KERNEL_PRIVATE */
	570	#endif /* _WAITQ_H_ */