1/*
2 * Copyright (c) 2006-2007 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28#ifndef _SYS_MCACHE_H
29#define	_SYS_MCACHE_H
30
31#ifdef KERNEL_PRIVATE
32
33#ifdef  __cplusplus
34extern "C" {
35#endif
36
37#include <sys/types.h>
38#include <sys/queue.h>
39#include <mach/boolean.h>
40#include <kern/locks.h>
41
42#ifdef ASSERT
43#undef ASSERT
44#endif
45
46#ifdef VERIFY
47#undef VERIFY
48#endif
49
50/*
51 * Unlike VERIFY(), ASSERT() is evaluated only in DEBUG build.
52 */
53#define	VERIFY(EX)	((void)((EX) || assfail(#EX, __FILE__, __LINE__)))
54#if DEBUG
55#define	ASSERT(EX)	VERIFY(EX)
56#else
57#define	ASSERT(EX)	((void)0)
58#endif
59
60#if defined(__ppc__)
61#define	CPU_CACHE_SIZE	128
62#else
63#define	CPU_CACHE_SIZE	64
64#endif
65
66#ifndef IS_P2ALIGNED
67#define	IS_P2ALIGNED(v, a) \
68	((((uintptr_t)(v)) & ((uintptr_t)(a) - 1)) == 0)
69#endif /* IS_P2ALIGNED */
70
71#ifndef P2ROUNDUP
72#define	P2ROUNDUP(x, align) \
73	(-(-((uintptr_t)(x)) & -(align)))
74#endif /* P2ROUNDUP */
75
76#ifndef P2ROUNDDOWN
77#define	P2ROUNDDOWN(x, align) \
78	(((uintptr_t)(x)) & ~((uintptr_t)(align) - 1))
79#endif /* P2ROUNDDOWN */
80
81#define	MCACHE_FREE_PATTERN		0xdeadbeefdeadbeefULL
82#define	MCACHE_UNINITIALIZED_PATTERN	0xbaddcafebaddcafeULL
83
84/*
85 * mcache allocation request flags.
86 *
87 * MCR_NOSLEEP and MCR_FAILOK are mutually exclusive.  The latter is used
88 * by the mbuf allocator to handle the implementation of several caches that
89 * involve multiple layers of mcache.  It implies a best effort blocking
90 * allocation request; if the request cannot be satisfied, the caller will
91 * be blocked until further notice, similar to MCR_SLEEP, except that upon
92 * a wake up it will return immediately to the caller regardless of whether
93 * the request can been fulfilled.
94 *
95 * MCR_TRYHARD implies a non-blocking allocation request, regardless of
96 * whether MCR_NOSLEEP is set.  It informs the allocator that the request
97 * should not cause the calling thread to block, and that it must have
98 * exhausted all possible schemes to fulfill the request, including doing
99 * reclaims and/or purges, before returning to the caller.
100 *
101 * Regular mcache clients should only use MCR_SLEEP or MCR_NOSLEEP.
102 */
103#define	MCR_SLEEP	0x0000		/* same as M_WAITOK */
104#define	MCR_NOSLEEP	0x0001		/* same as M_NOWAIT */
105#define	MCR_FAILOK	0x0100		/* private, for internal use only */
106#define	MCR_TRYHARD	0x0200		/* private, for internal use only */
107#define	MCR_USR1	0x1000		/* private, for internal use only */
108
109#define	MCR_NONBLOCKING	(MCR_NOSLEEP | MCR_FAILOK | MCR_TRYHARD)
110
111/*
112 * Generic one-way linked list element structure.  This is used to handle
113 * mcache_alloc_ext() requests in order to chain the allocated objects
114 * together before returning them to the caller.
115 */
116typedef struct mcache_obj {
117	struct mcache_obj	*obj_next;
118} mcache_obj_t;
119
120typedef struct mcache_bkt {
121	void		*bkt_next;	/* next bucket in list */
122	void		*bkt_obj[1];	/* one or more objects */
123} mcache_bkt_t;
124
125typedef struct mcache_bktlist {
126	mcache_bkt_t	*bl_list;	/* bucket list */
127	u_int32_t	bl_total;	/* number of buckets */
128	u_int32_t	bl_min;		/* min since last update */
129	u_int32_t	bl_reaplimit;	/* max reapable buckets */
130	u_int64_t	bl_alloc;	/* allocations from this list */
131} mcache_bktlist_t;
132
133typedef struct mcache_bkttype {
134	int		bt_bktsize;	/* bucket size (number of elements) */
135	size_t		bt_minbuf;	/* all smaller buffers qualify */
136	size_t		bt_maxbuf;	/* no larger bfufers qualify */
137	struct mcache	*bt_cache;	/* bucket cache */
138} mcache_bkttype_t;
139
140typedef struct mcache_cpu {
141	decl_lck_mtx_data(, cc_lock);
142	mcache_bkt_t	*cc_filled;	/* the currently filled bucket */
143	mcache_bkt_t	*cc_pfilled;	/* the previously filled bucket */
144	u_int64_t	cc_alloc;	/* allocations from this cpu */
145	u_int64_t	cc_free;	/* frees to this cpu */
146	int		cc_objs;	/* number of objects in filled bkt */
147	int		cc_pobjs;	/* number of objects in previous bkt */
148	int		cc_bktsize;	/* number of elements in a full bkt */
149} __attribute__((aligned(CPU_CACHE_SIZE), packed)) mcache_cpu_t;
150
151typedef unsigned int (*mcache_allocfn_t)(void *, mcache_obj_t ***,
152    unsigned int, int);
153typedef void (*mcache_freefn_t)(void *, mcache_obj_t *, boolean_t);
154typedef void (*mcache_auditfn_t)(void *, mcache_obj_t *, boolean_t);
155typedef void (*mcache_notifyfn_t)(void *, u_int32_t);
156
157typedef struct mcache {
158	/*
159	 * Cache properties
160	 */
161	LIST_ENTRY(mcache) mc_list;	/* cache linkage */
162	char		mc_name[32];	/* cache name */
163	struct zone	*mc_slab_zone;	/* backend zone allocator */
164	mcache_allocfn_t mc_slab_alloc;	/* slab layer allocate callback */
165	mcache_freefn_t	mc_slab_free;	/* slab layer free callback */
166	mcache_auditfn_t mc_slab_audit;	/* slab layer audit callback */
167	mcache_notifyfn_t mc_slab_notify; /* slab layer notify callback */
168	void		*mc_private;	/* opaque arg to callbacks */
169	size_t		mc_bufsize;	/* object size */
170	size_t		mc_align;	/* object alignment */
171	u_int32_t	mc_flags;	/* cache creation flags */
172	u_int32_t	mc_purge_cnt;	/* # of purges requested by slab */
173	u_int32_t	mc_enable_cnt;	/* # of reenables due to purges */
174	u_int32_t	mc_waiter_cnt;	/* # of slab layer waiters */
175	u_int32_t	mc_wretry_cnt;	/* # of wait retries */
176	u_int32_t	mc_nwretry_cnt;	/* # of no-wait retry attempts */
177	u_int32_t	mc_nwfail_cnt;	/* # of no-wait retries that failed */
178	decl_lck_mtx_data(, mc_sync_lock); /* protects purges and reenables */
179	lck_attr_t	*mc_sync_lock_attr;
180	lck_grp_t	*mc_sync_lock_grp;
181	lck_grp_attr_t	*mc_sync_lock_grp_attr;
182	/*
183	 * Keep CPU and buckets layers lock statistics separate.
184	 */
185	lck_attr_t	*mc_cpu_lock_attr;
186	lck_grp_t	*mc_cpu_lock_grp;
187	lck_grp_attr_t	*mc_cpu_lock_grp_attr;
188
189	/*
190	 * Bucket layer common to all CPUs
191	 */
192	decl_lck_mtx_data(, mc_bkt_lock);
193	lck_attr_t	*mc_bkt_lock_attr;
194	lck_grp_t	*mc_bkt_lock_grp;
195	lck_grp_attr_t  *mc_bkt_lock_grp_attr;
196	mcache_bkttype_t *cache_bkttype;	/* bucket type */
197	mcache_bktlist_t mc_full;		/* full buckets */
198	mcache_bktlist_t mc_empty;		/* empty buckets */
199	size_t		mc_chunksize;		/* bufsize + alignment */
200	u_int32_t	mc_bkt_contention;	/* lock contention count */
201	u_int32_t	mc_bkt_contention_prev;	/* previous snapshot */
202
203	/*
204	 * Per-CPU layer, aligned at cache line boundary
205	 */
206	mcache_cpu_t	mc_cpu[1];
207} mcache_t;
208
209#define	MCACHE_ALIGN	8	/* default guaranteed alignment */
210
211/* Valid values for mc_flags */
212#define	MCF_VERIFY	0x00000001	/* enable verification */
213#define	MCF_AUDIT	0x00000002	/* enable transaction auditing */
214#define	MCF_NOCPUCACHE	0x00000010	/* disable CPU layer caching */
215
216#define	MCF_DEBUG	(MCF_VERIFY | MCF_AUDIT)
217#define	MCF_FLAGS_MASK	(MCF_DEBUG | MCF_NOCPUCACHE)
218
219/* Valid values for notify callback */
220#define	MCN_RETRYALLOC	0x00000001	/* Allocation should be retried */
221
222#define	MCACHE_STACK_DEPTH 16
223
224typedef struct mcache_audit {
225	struct mcache_audit *mca_next;	/* next audit struct */
226	void		*mca_addr;	/* address of buffer */
227	mcache_t	*mca_cache;	/* parent cache of the buffer */
228	struct thread	*mca_thread;	/* thread doing transaction */
229	struct thread	*mca_pthread;	/* previous transaction thread */
230	size_t		mca_contents_size; /* size of contents */
231	void		*mca_contents;	/* contents at last free */
232	uint16_t	mca_depth;	/* pc stack depth */
233	uint16_t	mca_pdepth;	/* previous transaction pc stack */
234	void		*mca_stack[MCACHE_STACK_DEPTH];
235	void		*mca_pstack[MCACHE_STACK_DEPTH];
236	void		*mca_uptr;	/* user-specific pointer */
237	uint32_t	mca_uflags;	/* user-specific flags */
238} mcache_audit_t;
239
240__private_extern__ int assfail(const char *, const char *, int);
241__private_extern__ void mcache_init(void);
242__private_extern__ unsigned int mcache_getflags(void);
243__private_extern__ mcache_t *mcache_create(const char *, size_t,
244    size_t, u_int32_t, int);
245__private_extern__ void *mcache_alloc(mcache_t *, int);
246__private_extern__ void mcache_free(mcache_t *, void *);
247__private_extern__ mcache_t *mcache_create_ext(const char *, size_t,
248    mcache_allocfn_t, mcache_freefn_t, mcache_auditfn_t, mcache_notifyfn_t,
249    void *, u_int32_t, int);
250__private_extern__ void mcache_destroy(mcache_t *);
251__private_extern__ unsigned int mcache_alloc_ext(mcache_t *, mcache_obj_t **,
252    unsigned int, int);
253__private_extern__ void mcache_free_ext(mcache_t *, mcache_obj_t *);
254__private_extern__ void mcache_reap(void);
255__private_extern__ boolean_t mcache_purge_cache(mcache_t *);
256__private_extern__ void mcache_waiter_inc(mcache_t *);
257__private_extern__ void mcache_waiter_dec(mcache_t *);
258__private_extern__ boolean_t mcache_bkt_isempty(mcache_t *);
259
260__private_extern__ void mcache_buffer_log(mcache_audit_t *, void *, mcache_t *);
261__private_extern__ void mcache_set_pattern(u_int64_t, void *, size_t);
262__private_extern__ void *mcache_verify_pattern(u_int64_t, void *, size_t);
263__private_extern__ void *mcache_verify_set_pattern(u_int64_t, u_int64_t,
264    void *, size_t);
265__private_extern__ void mcache_audit_free_verify(mcache_audit_t *,
266    void *, size_t, size_t);
267__private_extern__ void mcache_audit_free_verify_set(mcache_audit_t *,
268    void *, size_t, size_t);
269__private_extern__ char *mcache_dump_mca(mcache_audit_t *);
270__private_extern__ void mcache_audit_panic(mcache_audit_t *, void *, size_t,
271    int64_t, int64_t);
272
273__private_extern__ mcache_t *mcache_audit_cache;
274
275#ifdef  __cplusplus
276}
277#endif
278
279#endif /* KERNEL_PRIVATE */
280
281#endif /* _SYS_MCACHE_H */
282