[apple/xnu.git] / bsd / sys / mcache.h

/*
 * Copyright (c) 2006-2014 Apple 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@
 */
#ifndef _SYS_MCACHE_H
#define	_SYS_MCACHE_H

#ifdef KERNEL_PRIVATE

#ifdef  __cplusplus
extern "C" {
#endif

#include <sys/types.h>
#include <sys/queue.h>
#include <mach/boolean.h>
#include <kern/locks.h>
#include <libkern/OSAtomic.h>

#ifdef ASSERT
#undef ASSERT
#endif

#ifdef VERIFY
#undef VERIFY
#endif

/*
 * Unlike VERIFY(), ASSERT() is evaluated only in DEBUG build.
 */
#define	VERIFY(EX)	((void)((EX) || assfail(#EX, __FILE__, __LINE__)))
#if DEBUG
#define	ASSERT(EX)	VERIFY(EX)
#else
#define	ASSERT(EX)	((void)0)
#endif

/*
 * Compile time assert; this should be on its own someday.
 */
#define	_CASSERT(x)	\
	switch (0) { case 0: case (x): ; }

/*
 * Atomic macros; these should be on their own someday.
 */
#define	atomic_add_16_ov(a, n)						\
	((u_int16_t) OSAddAtomic16(n, (volatile SInt16 *)a))

#define	atomic_add_16(a, n)						\
	((void) atomic_add_16_ov(a, n))

#define	atomic_add_32_ov(a, n)						\
	((u_int32_t) OSAddAtomic(n, (volatile SInt32 *)a))

#define	atomic_add_32(a, n)						\
	((void) atomic_add_32_ov(a, n))

#define	atomic_add_64_ov(a, n)						\
	((u_int64_t) OSAddAtomic64(n, (volatile SInt64 *)a))

#define	atomic_add_64(a, n)						\
	((void) atomic_add_64_ov(a, n))

#define	atomic_set_64(a, n) do {					\
	while (!OSCompareAndSwap64(*a, n, (volatile UInt64 *)a))	\
		;							\
} while (0)

#if defined(__LP64__)
#define	atomic_get_64(n, a) do {					\
	(n) = *(a);							\
} while (0)
#else
#define	atomic_get_64(n, a) do {					\
	(n) = atomic_add_64_ov(a, 0);					\
} while (0)
#endif /* __LP64__ */

#define	atomic_or_8_ov(a, n)						\
	((u_int8_t) OSBitOrAtomic8(n, (volatile UInt8 *)a))

#define	atomic_or_8(a, n)						\
	((void) atomic_or_8_ov(a, n))

#define	atomic_bitset_8(a, n)						\
	atomic_or_8(a, n)

#define	atomic_or_16_ov(a, n)						\
	((u_int16_t) OSBitOrAtomic16(n, (volatile UInt16 *)a))

#define	atomic_or_16(a, n)						\
	((void) atomic_or_16_ov(a, n))

#define	atomic_bitset_16(a, n)						\
	atomic_or_16(a, n)

#define	atomic_or_32_ov(a, n)						\
	((u_int32_t) OSBitOrAtomic(n, (volatile UInt32 *)a))

#define	atomic_or_32(a, n)						\
	((void) atomic_or_32_ov(a, n))

#define	atomic_bitset_32(a, n)						\
	atomic_or_32(a, n)

#define	atomic_and_8_ov(a, n)						\
	((u_int8_t) OSBitAndAtomic8(n, (volatile UInt8 *)a))

#define	atomic_and_8(a, n)						\
	((void) atomic_and_8_ov(a, n))

#define	atomic_bitclear_8(a, n)						\
	atomic_and_8(a, ~(n))

#define	atomic_and_16_ov(a, n)						\
	((u_int16_t) OSBitAndAtomic16(n, (volatile UInt16 *)a))

#define	atomic_and_16(a, n)						\
	((void) atomic_and_16_ov(a, n))

#define	atomic_bitclear_16(a, n)					\
	atomic_and_16(a, ~(n))

#define	atomic_and_32_ov(a, n)						\
	((u_int32_t) OSBitAndAtomic(n, (volatile UInt32 *)a))

#define	atomic_and_32(a, n)						\
	((void) atomic_and_32_ov(a, n))

#define	atomic_bitclear_32(a, n)					\
	atomic_and_32(a, ~(n))

/*
 * Use CPU_CACHE_LINE_SIZE instead of MAX_CPU_CACHE_LINE_SIZE, unless
 * wasting space is of no concern.
 */
#define	MAX_CPU_CACHE_LINE_SIZE	64
#define	CPU_CACHE_LINE_SIZE	mcache_cache_line_size()

#ifndef IS_P2ALIGNED
#define	IS_P2ALIGNED(v, a) \
	((((uintptr_t)(v)) & ((uintptr_t)(a) - 1)) == 0)
#endif /* IS_P2ALIGNED */

#ifndef P2ROUNDUP
#define	P2ROUNDUP(x, align) \
	(-(-((uintptr_t)(x)) & -((uintptr_t)align)))
#endif /* P2ROUNDUP */

#ifndef P2ROUNDDOWN
#define	P2ROUNDDOWN(x, align) \
	(((uintptr_t)(x)) & ~((uintptr_t)(align) - 1))
#endif /* P2ROUNDDOWN */

#define	MCACHE_FREE_PATTERN		0xdeadbeefdeadbeefULL
#define	MCACHE_UNINITIALIZED_PATTERN	0xbaddcafebaddcafeULL

/*
 * mcache allocation request flags.
 *
 * MCR_NOSLEEP and MCR_FAILOK are mutually exclusive.  The latter is used
 * by the mbuf allocator to handle the implementation of several caches that
 * involve multiple layers of mcache.  It implies a best effort blocking
 * allocation request; if the request cannot be satisfied, the caller will
 * be blocked until further notice, similar to MCR_SLEEP, except that upon
 * a wake up it will return immediately to the caller regardless of whether
 * the request can been fulfilled.
 *
 * MCR_TRYHARD implies a non-blocking allocation request, regardless of
 * whether MCR_NOSLEEP is set.  It informs the allocator that the request
 * should not cause the calling thread to block, and that it must have
 * exhausted all possible schemes to fulfill the request, including doing
 * reclaims and/or purges, before returning to the caller.
 *
 * Regular mcache clients should only use MCR_SLEEP or MCR_NOSLEEP.
 */
#define	MCR_SLEEP	0x0000		/* same as M_WAITOK */
#define	MCR_NOSLEEP	0x0001		/* same as M_NOWAIT */
#define	MCR_FAILOK	0x0100		/* private, for internal use only */
#define	MCR_TRYHARD	0x0200		/* private, for internal use only */
#define	MCR_USR1	0x1000		/* private, for internal use only */

#define	MCR_NONBLOCKING	(MCR_NOSLEEP | MCR_FAILOK | MCR_TRYHARD)

/*
 * Generic one-way linked list element structure.  This is used to handle
 * mcache_alloc_ext() requests in order to chain the allocated objects
 * together before returning them to the caller.
 */
typedef struct mcache_obj {
	struct mcache_obj	*obj_next;
} mcache_obj_t;

typedef struct mcache_bkt {
	void		*bkt_next;	/* next bucket in list */
	void		*bkt_obj[1];	/* one or more objects */
} mcache_bkt_t;

typedef struct mcache_bktlist {
	mcache_bkt_t	*bl_list;	/* bucket list */
	u_int32_t	bl_total;	/* number of buckets */
	u_int32_t	bl_min;		/* min since last update */
	u_int32_t	bl_reaplimit;	/* max reapable buckets */
	u_int64_t	bl_alloc;	/* allocations from this list */
} mcache_bktlist_t;

typedef struct mcache_bkttype {
	int		bt_bktsize;	/* bucket size (number of elements) */
	size_t		bt_minbuf;	/* all smaller buffers qualify */
	size_t		bt_maxbuf;	/* no larger bfufers qualify */
	struct mcache	*bt_cache;	/* bucket cache */
} mcache_bkttype_t;

typedef struct mcache_cpu {
	decl_lck_mtx_data(, cc_lock);
	mcache_bkt_t	*cc_filled;	/* the currently filled bucket */
	mcache_bkt_t	*cc_pfilled;	/* the previously filled bucket */
	u_int64_t	cc_alloc;	/* allocations from this cpu */
	u_int64_t	cc_free;	/* frees to this cpu */
	int		cc_objs;	/* number of objects in filled bkt */
	int		cc_pobjs;	/* number of objects in previous bkt */
	int		cc_bktsize;	/* number of elements in a full bkt */
} __attribute__((aligned(MAX_CPU_CACHE_LINE_SIZE), packed)) mcache_cpu_t;

typedef unsigned int (*mcache_allocfn_t)(void *, mcache_obj_t ***,
    unsigned int, int);
typedef void (*mcache_freefn_t)(void *, mcache_obj_t *, boolean_t);
typedef void (*mcache_auditfn_t)(void *, mcache_obj_t *, boolean_t);
typedef void (*mcache_logfn_t)(u_int32_t, mcache_obj_t *, boolean_t);
typedef void (*mcache_notifyfn_t)(void *, u_int32_t);

typedef struct mcache {
	/*
	 * Cache properties
	 */
	LIST_ENTRY(mcache) mc_list;	/* cache linkage */
	char		mc_name[32];	/* cache name */
	struct zone	*mc_slab_zone;	/* backend zone allocator */
	mcache_allocfn_t mc_slab_alloc;	/* slab layer allocate callback */
	mcache_freefn_t	mc_slab_free;	/* slab layer free callback */
	mcache_auditfn_t mc_slab_audit;	/* slab layer audit callback */
	mcache_logfn_t mc_slab_log;	/* slab layer log callback */
	mcache_notifyfn_t mc_slab_notify; /* slab layer notify callback */
	void		*mc_private;	/* opaque arg to callbacks */
	size_t		mc_bufsize;	/* object size */
	size_t		mc_align;	/* object alignment */
	u_int32_t	mc_flags;	/* cache creation flags */
	u_int32_t	mc_purge_cnt;	/* # of purges requested by slab */
	u_int32_t	mc_enable_cnt;	/* # of reenables due to purges */
	u_int32_t	mc_waiter_cnt;	/* # of slab layer waiters */
	u_int32_t	mc_wretry_cnt;	/* # of wait retries */
	u_int32_t	mc_nwretry_cnt;	/* # of no-wait retry attempts */
	u_int32_t	mc_nwfail_cnt;	/* # of no-wait retries that failed */
	decl_lck_mtx_data(, mc_sync_lock); /* protects purges and reenables */
	lck_attr_t	*mc_sync_lock_attr;
	lck_grp_t	*mc_sync_lock_grp;
	lck_grp_attr_t	*mc_sync_lock_grp_attr;
	/*
	 * Keep CPU and buckets layers lock statistics separate.
	 */
	lck_attr_t	*mc_cpu_lock_attr;
	lck_grp_t	*mc_cpu_lock_grp;
	lck_grp_attr_t	*mc_cpu_lock_grp_attr;

	/*
	 * Bucket layer common to all CPUs
	 */
	decl_lck_mtx_data(, mc_bkt_lock);
	lck_attr_t	*mc_bkt_lock_attr;
	lck_grp_t	*mc_bkt_lock_grp;
	lck_grp_attr_t  *mc_bkt_lock_grp_attr;
	mcache_bkttype_t *cache_bkttype;	/* bucket type */
	mcache_bktlist_t mc_full;		/* full buckets */
	mcache_bktlist_t mc_empty;		/* empty buckets */
	size_t		mc_chunksize;		/* bufsize + alignment */
	u_int32_t	mc_bkt_contention;	/* lock contention count */
	u_int32_t	mc_bkt_contention_prev;	/* previous snapshot */

	/*
	 * Per-CPU layer, aligned at cache line boundary
	 */
	mcache_cpu_t	mc_cpu[1];
} mcache_t;

#define	MCACHE_ALIGN	8	/* default guaranteed alignment */

/* Valid values for mc_flags */
#define	MCF_VERIFY	0x00000001	/* enable verification */
#define	MCF_TRACE	0x00000002	/* enable transaction auditing */
#define	MCF_NOCPUCACHE	0x00000010	/* disable CPU layer caching */
#define	MCF_NOLEAKLOG	0x00000100	/* disable leak logging */
#define	MCF_EXPLEAKLOG	0x00000200	/* expose leak info to user space */

#define	MCF_DEBUG	(MCF_VERIFY | MCF_TRACE)
#define	MCF_FLAGS_MASK	\
	(MCF_DEBUG | MCF_NOCPUCACHE | MCF_NOLEAKLOG | MCF_EXPLEAKLOG)

/* Valid values for notify callback */
#define	MCN_RETRYALLOC	0x00000001	/* Allocation should be retried */

#define	MCACHE_STACK_DEPTH 16

#define	MCA_TRN_MAX	2		/* Number of transactions to record */

typedef struct mcache_audit {
	struct mcache_audit *mca_next;	/* next audit struct */
	void		*mca_addr;	/* address of buffer */
	mcache_t	*mca_cache;	/* parent cache of the buffer */
	size_t		mca_contents_size; /* size of saved contents */
	void		*mca_contents;	/* user-specific saved contents */
	void		*mca_uptr;	/* user-specific pointer */
	uint32_t	mca_uflags;	/* user-specific flags */
	uint32_t	mca_next_trn;
	struct mca_trn {
		struct thread	*mca_thread;	/* thread doing transaction */
		uint32_t	mca_tstamp;
		uint16_t	mca_depth;
		void		*mca_stack[MCACHE_STACK_DEPTH];
	} mca_trns[MCA_TRN_MAX];
} mcache_audit_t;

__private_extern__ int assfail(const char *, const char *, int);
__private_extern__ void mcache_init(void);
__private_extern__ unsigned int mcache_getflags(void);
__private_extern__ unsigned int mcache_cache_line_size(void);
__private_extern__ mcache_t *mcache_create(const char *, size_t,
    size_t, u_int32_t, int);
__private_extern__ void *mcache_alloc(mcache_t *, int);
__private_extern__ void mcache_free(mcache_t *, void *);
__private_extern__ mcache_t *mcache_create_ext(const char *, size_t,
    mcache_allocfn_t, mcache_freefn_t, mcache_auditfn_t, mcache_logfn_t,
    mcache_notifyfn_t, void *, u_int32_t, int);
__private_extern__ void mcache_destroy(mcache_t *);
__private_extern__ unsigned int mcache_alloc_ext(mcache_t *, mcache_obj_t **,
    unsigned int, int);
__private_extern__ void mcache_free_ext(mcache_t *, mcache_obj_t *);
__private_extern__ void mcache_reap(void);
__private_extern__ boolean_t mcache_purge_cache(mcache_t *, boolean_t);
__private_extern__ void mcache_waiter_inc(mcache_t *);
__private_extern__ void mcache_waiter_dec(mcache_t *);
__private_extern__ boolean_t mcache_bkt_isempty(mcache_t *);

__private_extern__ void mcache_buffer_log(mcache_audit_t *, void *, mcache_t *,
    struct timeval *);
__private_extern__ void mcache_set_pattern(u_int64_t, void *, size_t);
__private_extern__ void *mcache_verify_pattern(u_int64_t, void *, size_t);
__private_extern__ void *mcache_verify_set_pattern(u_int64_t, u_int64_t,
    void *, size_t);
__private_extern__ void mcache_audit_free_verify(mcache_audit_t *,
    void *, size_t, size_t);
__private_extern__ void mcache_audit_free_verify_set(mcache_audit_t *,
    void *, size_t, size_t);
__private_extern__ char *mcache_dump_mca(mcache_audit_t *);
__private_extern__ void mcache_audit_panic(mcache_audit_t *, void *, size_t,
    int64_t, int64_t);

extern int32_t total_sbmb_cnt;
extern int32_t total_sbmb_cnt_peak;
extern int64_t sbmb_limreached;
extern mcache_t *mcache_audit_cache;

#ifdef  __cplusplus
}
#endif

#endif /* KERNEL_PRIVATE */

#endif /* _SYS_MCACHE_H */
Commit	Line	Data
2d21ac55	1	/*
fe8ab488	2	* Copyright (c) 2006-2014 Apple Inc. All rights reserved.
2d21ac55 A	3	*
2d21ac55 A	4	* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
39236c6e	5	*
2d21ac55 A	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.
39236c6e	14	*
2d21ac55 A	15	* Please obtain a copy of the License at
2d21ac55 A	16	* http://www.opensource.apple.com/apsl/ and read it before using this file.
39236c6e	17	*
2d21ac55 A	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.
39236c6e	25	*
2d21ac55 A	26	* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
	27	*/
	28	#ifndef _SYS_MCACHE_H
	29	#define _SYS_MCACHE_H
	30
	31	#ifdef KERNEL_PRIVATE
	32
	33	#ifdef __cplusplus
	34	extern "C" {
	35	#endif
	36
	37	#include <sys/types.h>
	38	#include <sys/queue.h>
	39	#include <mach/boolean.h>
	40	#include <kern/locks.h>
6d2010ae	41	#include <libkern/OSAtomic.h>
2d21ac55 A	42
	43	#ifdef ASSERT
	44	#undef ASSERT
	45	#endif
	46
	47	#ifdef VERIFY
	48	#undef VERIFY
	49	#endif
	50
	51	/*
	52	* Unlike VERIFY(), ASSERT() is evaluated only in DEBUG build.
	53	*/
	54	#define VERIFY(EX) ((void)((EX) \|\| assfail(#EX, __FILE__, __LINE__)))
	55	#if DEBUG
	56	#define ASSERT(EX) VERIFY(EX)
	57	#else
	58	#define ASSERT(EX) ((void)0)
	59	#endif
	60
316670eb A	61	/*
	62	* Compile time assert; this should be on its own someday.
	63	*/
	64	#define _CASSERT(x) \
	65	switch (0) { case 0: case (x): ; }
	66
	67	/*
	68	* Atomic macros; these should be on their own someday.
	69	*/
6d2010ae A	70	#define atomic_add_16_ov(a, n) \
	71	((u_int16_t) OSAddAtomic16(n, (volatile SInt16 *)a))
	72
	73	#define atomic_add_16(a, n) \
	74	((void) atomic_add_16_ov(a, n))
	75
	76	#define atomic_add_32_ov(a, n) \
	77	((u_int32_t) OSAddAtomic(n, (volatile SInt32 *)a))
	78
	79	#define atomic_add_32(a, n) \
	80	((void) atomic_add_32_ov(a, n))
	81
	82	#define atomic_add_64_ov(a, n) \
	83	((u_int64_t) OSAddAtomic64(n, (volatile SInt64 *)a))
	84
	85	#define atomic_add_64(a, n) \
	86	((void) atomic_add_64_ov(a, n))
	87
	88	#define atomic_set_64(a, n) do { \
	89	while (!OSCompareAndSwap64(a, n, (volatile UInt64 )a)) \
	90	; \
	91	} while (0)
	92
	93	#if defined(__LP64__)
	94	#define atomic_get_64(n, a) do { \
	95	(n) = *(a); \
	96	} while (0)
2d21ac55	97	#else
6d2010ae A	98	#define atomic_get_64(n, a) do { \
	99	(n) = atomic_add_64_ov(a, 0); \
	100	} while (0)
	101	#endif /* __LP64__ */
	102
39236c6e A	103	#define atomic_or_8_ov(a, n) \
	104	((u_int8_t) OSBitOrAtomic8(n, (volatile UInt8 *)a))
	105
	106	#define atomic_or_8(a, n) \
	107	((void) atomic_or_8_ov(a, n))
	108
	109	#define atomic_bitset_8(a, n) \
	110	atomic_or_8(a, n)
	111
	112	#define atomic_or_16_ov(a, n) \
	113	((u_int16_t) OSBitOrAtomic16(n, (volatile UInt16 *)a))
	114
	115	#define atomic_or_16(a, n) \
	116	((void) atomic_or_16_ov(a, n))
	117
	118	#define atomic_bitset_16(a, n) \
	119	atomic_or_16(a, n)
	120
	121	#define atomic_or_32_ov(a, n) \
	122	((u_int32_t) OSBitOrAtomic(n, (volatile UInt32 *)a))
	123
	124	#define atomic_or_32(a, n) \
	125	((void) atomic_or_32_ov(a, n))
	126
	127	#define atomic_bitset_32(a, n) \
	128	atomic_or_32(a, n)
	129
	130	#define atomic_and_8_ov(a, n) \
	131	((u_int8_t) OSBitAndAtomic8(n, (volatile UInt8 *)a))
	132
	133	#define atomic_and_8(a, n) \
	134	((void) atomic_and_8_ov(a, n))
	135
	136	#define atomic_bitclear_8(a, n) \
	137	atomic_and_8(a, ~(n))
	138
	139	#define atomic_and_16_ov(a, n) \
	140	((u_int16_t) OSBitAndAtomic16(n, (volatile UInt16 *)a))
	141
	142	#define atomic_and_16(a, n) \
	143	((void) atomic_and_16_ov(a, n))
	144
	145	#define atomic_bitclear_16(a, n) \
	146	atomic_and_16(a, ~(n))
	147
	148	#define atomic_and_32_ov(a, n) \
	149	((u_int32_t) OSBitAndAtomic(n, (volatile UInt32 *)a))
	150
	151	#define atomic_and_32(a, n) \
	152	((void) atomic_and_32_ov(a, n))
	153
	154	#define atomic_bitclear_32(a, n) \
	155	atomic_and_32(a, ~(n))
	156
	157	/*
	158	* Use CPU_CACHE_LINE_SIZE instead of MAX_CPU_CACHE_LINE_SIZE, unless
	159	* wasting space is of no concern.
	160	*/
	161	#define MAX_CPU_CACHE_LINE_SIZE 64
	162	#define CPU_CACHE_LINE_SIZE mcache_cache_line_size()
2d21ac55 A	163
	164	#ifndef IS_P2ALIGNED
	165	#define IS_P2ALIGNED(v, a) \
	166	((((uintptr_t)(v)) & ((uintptr_t)(a) - 1)) == 0)
	167	#endif /* IS_P2ALIGNED */
	168
	169	#ifndef P2ROUNDUP
	170	#define P2ROUNDUP(x, align) \
39236c6e	171	(-(-((uintptr_t)(x)) & -((uintptr_t)align)))
2d21ac55 A	172	#endif /* P2ROUNDUP */
	173
	174	#ifndef P2ROUNDDOWN
	175	#define P2ROUNDDOWN(x, align) \
	176	(((uintptr_t)(x)) & ~((uintptr_t)(align) - 1))
	177	#endif /* P2ROUNDDOWN */
	178
	179	#define MCACHE_FREE_PATTERN 0xdeadbeefdeadbeefULL
	180	#define MCACHE_UNINITIALIZED_PATTERN 0xbaddcafebaddcafeULL
	181
	182	/*
	183	* mcache allocation request flags.
	184	*
	185	* MCR_NOSLEEP and MCR_FAILOK are mutually exclusive. The latter is used
	186	* by the mbuf allocator to handle the implementation of several caches that
	187	* involve multiple layers of mcache. It implies a best effort blocking
	188	* allocation request; if the request cannot be satisfied, the caller will
	189	* be blocked until further notice, similar to MCR_SLEEP, except that upon
	190	* a wake up it will return immediately to the caller regardless of whether
	191	* the request can been fulfilled.
	192	*
	193	* MCR_TRYHARD implies a non-blocking allocation request, regardless of
	194	* whether MCR_NOSLEEP is set. It informs the allocator that the request
	195	* should not cause the calling thread to block, and that it must have
	196	* exhausted all possible schemes to fulfill the request, including doing
	197	* reclaims and/or purges, before returning to the caller.
	198	*
	199	* Regular mcache clients should only use MCR_SLEEP or MCR_NOSLEEP.
	200	*/
	201	#define MCR_SLEEP 0x0000 /* same as M_WAITOK */
	202	#define MCR_NOSLEEP 0x0001 /* same as M_NOWAIT */
	203	#define MCR_FAILOK 0x0100 /* private, for internal use only */
	204	#define MCR_TRYHARD 0x0200 /* private, for internal use only */
	205	#define MCR_USR1 0x1000 /* private, for internal use only */
	206
	207	#define MCR_NONBLOCKING (MCR_NOSLEEP \| MCR_FAILOK \| MCR_TRYHARD)
	208
	209	/*
	210	* Generic one-way linked list element structure. This is used to handle
	211	* mcache_alloc_ext() requests in order to chain the allocated objects
	212	* together before returning them to the caller.
	213	*/
	214	typedef struct mcache_obj {
	215	struct mcache_obj *obj_next;
	216	} mcache_obj_t;
	217
	218	typedef struct mcache_bkt {
	219	void bkt_next; / next bucket in list */
	220	void bkt_obj[1]; / one or more objects */
	221	} mcache_bkt_t;
	222
	223	typedef struct mcache_bktlist {
	224	mcache_bkt_t bl_list; / bucket list */
	225	u_int32_t bl_total; /* number of buckets */
	226	u_int32_t bl_min; /* min since last update */
	227	u_int32_t bl_reaplimit; /* max reapable buckets */
	228	u_int64_t bl_alloc; /* allocations from this list */
	229	} mcache_bktlist_t;
	230
	231	typedef struct mcache_bkttype {
	232	int bt_bktsize; /* bucket size (number of elements) */
	233	size_t bt_minbuf; /* all smaller buffers qualify */
	234	size_t bt_maxbuf; /* no larger bfufers qualify */
	235	struct mcache bt_cache; / bucket cache */
236	} mcache_bkttype_t;
237
238	typedef struct mcache_cpu {
239	decl_lck_mtx_data(, cc_lock);
240	mcache_bkt_t cc_filled; / the currently filled bucket */
241	mcache_bkt_t cc_pfilled; / the previously filled bucket */
242	u_int64_t cc_alloc; /* allocations from this cpu */
243	u_int64_t cc_free; /* frees to this cpu */
244	int cc_objs; /* number of objects in filled bkt */
245	int cc_pobjs; /* number of objects in previous bkt */
246	int cc_bktsize; /* number of elements in a full bkt */
39236c6e	247	} __attribute__((aligned(MAX_CPU_CACHE_LINE_SIZE), packed)) mcache_cpu_t;
2d21ac55 A	248
	249	typedef unsigned int (mcache_allocfn_t)(void , mcache_obj_t ***,
	250	unsigned int, int);
	251	typedef void (mcache_freefn_t)(void , mcache_obj_t *, boolean_t);
	252	typedef void (mcache_auditfn_t)(void , mcache_obj_t *, boolean_t);
6d2010ae	253	typedef void (mcache_logfn_t)(u_int32_t, mcache_obj_t , boolean_t);
2d21ac55 A	254	typedef void (mcache_notifyfn_t)(void , u_int32_t);
	255
	256	typedef struct mcache {
	257	/*
	258	* Cache properties
	259	*/
	260	LIST_ENTRY(mcache) mc_list; /* cache linkage */
	261	char mc_name[32]; /* cache name */
	262	struct zone mc_slab_zone; / backend zone allocator */
	263	mcache_allocfn_t mc_slab_alloc; /* slab layer allocate callback */
	264	mcache_freefn_t mc_slab_free; /* slab layer free callback */
	265	mcache_auditfn_t mc_slab_audit; /* slab layer audit callback */
6d2010ae	266	mcache_logfn_t mc_slab_log; /* slab layer log callback */
2d21ac55 A	267	mcache_notifyfn_t mc_slab_notify; /* slab layer notify callback */
	268	void mc_private; / opaque arg to callbacks */
	269	size_t mc_bufsize; /* object size */
	270	size_t mc_align; /* object alignment */
	271	u_int32_t mc_flags; /* cache creation flags */
	272	u_int32_t mc_purge_cnt; /* # of purges requested by slab */
	273	u_int32_t mc_enable_cnt; /* # of reenables due to purges */
	274	u_int32_t mc_waiter_cnt; /* # of slab layer waiters */
	275	u_int32_t mc_wretry_cnt; /* # of wait retries */
	276	u_int32_t mc_nwretry_cnt; /* # of no-wait retry attempts */
	277	u_int32_t mc_nwfail_cnt; /* # of no-wait retries that failed */
	278	decl_lck_mtx_data(, mc_sync_lock); /* protects purges and reenables */
	279	lck_attr_t *mc_sync_lock_attr;
	280	lck_grp_t *mc_sync_lock_grp;
	281	lck_grp_attr_t *mc_sync_lock_grp_attr;
	282	/*
	283	* Keep CPU and buckets layers lock statistics separate.
	284	*/
	285	lck_attr_t *mc_cpu_lock_attr;
	286	lck_grp_t *mc_cpu_lock_grp;
	287	lck_grp_attr_t *mc_cpu_lock_grp_attr;
	288
	289	/*
	290	* Bucket layer common to all CPUs
	291	*/
	292	decl_lck_mtx_data(, mc_bkt_lock);
	293	lck_attr_t *mc_bkt_lock_attr;
	294	lck_grp_t *mc_bkt_lock_grp;
	295	lck_grp_attr_t *mc_bkt_lock_grp_attr;
	296	mcache_bkttype_t cache_bkttype; / bucket type */
	297	mcache_bktlist_t mc_full; /* full buckets */
	298	mcache_bktlist_t mc_empty; /* empty buckets */
	299	size_t mc_chunksize; /* bufsize + alignment */
	300	u_int32_t mc_bkt_contention; /* lock contention count */
	301	u_int32_t mc_bkt_contention_prev; /* previous snapshot */
	302
	303	/*
	304	* Per-CPU layer, aligned at cache line boundary
	305	*/
	306	mcache_cpu_t mc_cpu[1];
	307	} mcache_t;
	308
	309	#define MCACHE_ALIGN 8 /* default guaranteed alignment */
	310
	311	/* Valid values for mc_flags */
	312	#define MCF_VERIFY 0x00000001 /* enable verification */
6d2010ae	313	#define MCF_TRACE 0x00000002 /* enable transaction auditing */
2d21ac55	314	#define MCF_NOCPUCACHE 0x00000010 /* disable CPU layer caching */
6d2010ae	315	#define MCF_NOLEAKLOG 0x00000100 /* disable leak logging */
316670eb	316	#define MCF_EXPLEAKLOG 0x00000200 /* expose leak info to user space */
2d21ac55	317
6d2010ae	318	#define MCF_DEBUG (MCF_VERIFY \| MCF_TRACE)
316670eb A	319	#define MCF_FLAGS_MASK \
316670eb A	320	(MCF_DEBUG \| MCF_NOCPUCACHE \| MCF_NOLEAKLOG \| MCF_EXPLEAKLOG)
2d21ac55 A	321
	322	/* Valid values for notify callback */
	323	#define MCN_RETRYALLOC 0x00000001 /* Allocation should be retried */
	324
	325	#define MCACHE_STACK_DEPTH 16
	326
fe8ab488 A	327	#define MCA_TRN_MAX 2 /* Number of transactions to record */
fe8ab488 A	328
2d21ac55 A	329	typedef struct mcache_audit {
	330	struct mcache_audit mca_next; / next audit struct */
	331	void mca_addr; / address of buffer */
	332	mcache_t mca_cache; / parent cache of the buffer */
39236c6e A	333	size_t mca_contents_size; /* size of saved contents */
39236c6e A	334	void mca_contents; / user-specific saved contents */
2d21ac55 A	335	void mca_uptr; / user-specific pointer */
2d21ac55 A	336	uint32_t mca_uflags; /* user-specific flags */
fe8ab488 A	337	uint32_t mca_next_trn;
	338	struct mca_trn {
	339	struct thread mca_thread; / thread doing transaction */
	340	uint32_t mca_tstamp;
	341	uint16_t mca_depth;
	342	void *mca_stack[MCACHE_STACK_DEPTH];
	343	} mca_trns[MCA_TRN_MAX];
2d21ac55 A	344	} mcache_audit_t;
	345
	346	__private_extern__ int assfail(const char , const char , int);
	347	__private_extern__ void mcache_init(void);
	348	__private_extern__ unsigned int mcache_getflags(void);
39236c6e	349	__private_extern__ unsigned int mcache_cache_line_size(void);
2d21ac55 A	350	__private_extern__ mcache_t mcache_create(const char , size_t,
	351	size_t, u_int32_t, int);
	352	__private_extern__ void mcache_alloc(mcache_t , int);
	353	__private_extern__ void mcache_free(mcache_t , void );
	354	__private_extern__ mcache_t mcache_create_ext(const char , size_t,
6d2010ae A	355	mcache_allocfn_t, mcache_freefn_t, mcache_auditfn_t, mcache_logfn_t,
6d2010ae A	356	mcache_notifyfn_t, void *, u_int32_t, int);
2d21ac55 A	357	__private_extern__ void mcache_destroy(mcache_t *);
	358	__private_extern__ unsigned int mcache_alloc_ext(mcache_t , mcache_obj_t *,
	359	unsigned int, int);
	360	__private_extern__ void mcache_free_ext(mcache_t , mcache_obj_t );
	361	__private_extern__ void mcache_reap(void);
fe8ab488	362	__private_extern__ boolean_t mcache_purge_cache(mcache_t *, boolean_t);
2d21ac55 A	363	__private_extern__ void mcache_waiter_inc(mcache_t *);
	364	__private_extern__ void mcache_waiter_dec(mcache_t *);
	365	__private_extern__ boolean_t mcache_bkt_isempty(mcache_t *);
	366
39236c6e A	367	__private_extern__ void mcache_buffer_log(mcache_audit_t , void , mcache_t *,
39236c6e A	368	struct timeval *);
2d21ac55 A	369	__private_extern__ void mcache_set_pattern(u_int64_t, void *, size_t);
	370	__private_extern__ void mcache_verify_pattern(u_int64_t, void , size_t);
	371	__private_extern__ void *mcache_verify_set_pattern(u_int64_t, u_int64_t,
	372	void *, size_t);
	373	__private_extern__ void mcache_audit_free_verify(mcache_audit_t *,
	374	void *, size_t, size_t);
	375	__private_extern__ void mcache_audit_free_verify_set(mcache_audit_t *,
	376	void *, size_t, size_t);
	377	__private_extern__ char mcache_dump_mca(mcache_audit_t );
	378	__private_extern__ void mcache_audit_panic(mcache_audit_t , void , size_t,
	379	int64_t, int64_t);
	380
fe8ab488 A	381	extern int32_t total_sbmb_cnt;
	382	extern int32_t total_sbmb_cnt_peak;
	383	extern int64_t sbmb_limreached;
39236c6e	384	extern mcache_t *mcache_audit_cache;
2d21ac55 A	385
	386	#ifdef __cplusplus
	387	}
	388	#endif
	389
	390	#endif /* KERNEL_PRIVATE */
	391
	392	#endif /* _SYS_MCACHE_H */