bsd/sys/mcache.h

   1 /*
   2  * Copyright (c) 2006-2010 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
  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>
  41 #include <libkern/OSAtomic.h>
  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
  61 #define atomic_add_16_ov(a, n)                                          \
  62         ((u_int16_t) OSAddAtomic16(n, (volatile SInt16 *)a))
  63
  64 #define atomic_add_16(a, n)                                             \
  65         ((void) atomic_add_16_ov(a, n))
  66
  67 #define atomic_add_32_ov(a, n)                                          \
  68         ((u_int32_t) OSAddAtomic(n, (volatile SInt32 *)a))
  69
  70 #define atomic_add_32(a, n)                                             \
  71         ((void) atomic_add_32_ov(a, n))
  72
  73 #define atomic_add_64_ov(a, n)                                          \
  74         ((u_int64_t) OSAddAtomic64(n, (volatile SInt64 *)a))
  75
  76 #define atomic_add_64(a, n)                                             \
  77         ((void) atomic_add_64_ov(a, n))
  78
  79 #define atomic_set_64(a, n) do {                                        \
  80         while (!OSCompareAndSwap64(*a, n, (volatile UInt64 *)a))        \
  81                 ;                                                       \
  82 } while (0)
  83
  84 #if defined(__LP64__)
  85 #define atomic_get_64(n, a) do {                                        \
  86         (n) = *(a);                                                     \
  87 } while (0)
  88 #else
  89 #define atomic_get_64(n, a) do {                                        \
  90         (n) = atomic_add_64_ov(a, 0);                                   \
  91 } while (0)
  92 #endif /* __LP64__ */
  93
  94 #define CPU_CACHE_SIZE  64
  95
  96 #ifndef IS_P2ALIGNED
  97 #define IS_P2ALIGNED(v, a) \
  98         ((((uintptr_t)(v)) & ((uintptr_t)(a) - 1)) == 0)
  99 #endif /* IS_P2ALIGNED */
 100
 101 #ifndef P2ROUNDUP
 102 #define P2ROUNDUP(x, align) \
 103         (-(-((uintptr_t)(x)) & -(align)))
 104 #endif /* P2ROUNDUP */
 105
 106 #ifndef P2ROUNDDOWN
 107 #define P2ROUNDDOWN(x, align) \
 108         (((uintptr_t)(x)) & ~((uintptr_t)(align) - 1))
 109 #endif /* P2ROUNDDOWN */
 110
 111 #define MCACHE_FREE_PATTERN             0xdeadbeefdeadbeefULL
 112 #define MCACHE_UNINITIALIZED_PATTERN    0xbaddcafebaddcafeULL
 113
 114 /*
 115  * mcache allocation request flags.
 116  *
 117  * MCR_NOSLEEP and MCR_FAILOK are mutually exclusive.  The latter is used
 118  * by the mbuf allocator to handle the implementation of several caches that
 119  * involve multiple layers of mcache.  It implies a best effort blocking
 120  * allocation request; if the request cannot be satisfied, the caller will
 121  * be blocked until further notice, similar to MCR_SLEEP, except that upon
 122  * a wake up it will return immediately to the caller regardless of whether
 123  * the request can been fulfilled.
 124  *
 125  * MCR_TRYHARD implies a non-blocking allocation request, regardless of
 126  * whether MCR_NOSLEEP is set.  It informs the allocator that the request
 127  * should not cause the calling thread to block, and that it must have
 128  * exhausted all possible schemes to fulfill the request, including doing
 129  * reclaims and/or purges, before returning to the caller.
 130  *
 131  * Regular mcache clients should only use MCR_SLEEP or MCR_NOSLEEP.
 132  */
 133 #define MCR_SLEEP       0x0000          /* same as M_WAITOK */
 134 #define MCR_NOSLEEP     0x0001          /* same as M_NOWAIT */
 135 #define MCR_FAILOK      0x0100          /* private, for internal use only */
 136 #define MCR_TRYHARD     0x0200          /* private, for internal use only */
 137 #define MCR_USR1        0x1000          /* private, for internal use only */
 138
 139 #define MCR_NONBLOCKING (MCR_NOSLEEP | MCR_FAILOK | MCR_TRYHARD)
 140
 141 /*
 142  * Generic one-way linked list element structure.  This is used to handle
 143  * mcache_alloc_ext() requests in order to chain the allocated objects
 144  * together before returning them to the caller.
 145  */
 146 typedef struct mcache_obj {
 147         struct mcache_obj       *obj_next;
 148 } mcache_obj_t;
 149
 150 typedef struct mcache_bkt {
 151         void            *bkt_next;      /* next bucket in list */
 152         void            *bkt_obj[1];    /* one or more objects */
 153 } mcache_bkt_t;
 154
 155 typedef struct mcache_bktlist {
 156         mcache_bkt_t    *bl_list;       /* bucket list */
 157         u_int32_t       bl_total;       /* number of buckets */
 158         u_int32_t       bl_min;         /* min since last update */
 159         u_int32_t       bl_reaplimit;   /* max reapable buckets */
 160         u_int64_t       bl_alloc;       /* allocations from this list */
 161 } mcache_bktlist_t;
 162
 163 typedef struct mcache_bkttype {
 164         int             bt_bktsize;     /* bucket size (number of elements) */
 165         size_t          bt_minbuf;      /* all smaller buffers qualify */
 166         size_t          bt_maxbuf;      /* no larger bfufers qualify */
 167         struct mcache   *bt_cache;      /* bucket cache */
 168 } mcache_bkttype_t;
 169
 170 typedef struct mcache_cpu {
 171         decl_lck_mtx_data(, cc_lock);
 172         mcache_bkt_t    *cc_filled;     /* the currently filled bucket */
 173         mcache_bkt_t    *cc_pfilled;    /* the previously filled bucket */
 174         u_int64_t       cc_alloc;       /* allocations from this cpu */
 175         u_int64_t       cc_free;        /* frees to this cpu */
 176         int             cc_objs;        /* number of objects in filled bkt */
 177         int             cc_pobjs;       /* number of objects in previous bkt */
 178         int             cc_bktsize;     /* number of elements in a full bkt */
 179 } __attribute__((aligned(CPU_CACHE_SIZE), packed)) mcache_cpu_t;
 180
 181 typedef unsigned int (*mcache_allocfn_t)(void *, mcache_obj_t ***,
 182     unsigned int, int);
 183 typedef void (*mcache_freefn_t)(void *, mcache_obj_t *, boolean_t);
 184 typedef void (*mcache_auditfn_t)(void *, mcache_obj_t *, boolean_t);
 185 typedef void (*mcache_logfn_t)(u_int32_t, mcache_obj_t *, boolean_t);
 186 typedef void (*mcache_notifyfn_t)(void *, u_int32_t);
 187
 188 typedef struct mcache {
 189         /*
 190          * Cache properties
 191          */
 192         LIST_ENTRY(mcache) mc_list;     /* cache linkage */
 193         char            mc_name[32];    /* cache name */
 194         struct zone     *mc_slab_zone;  /* backend zone allocator */
 195         mcache_allocfn_t mc_slab_alloc; /* slab layer allocate callback */
 196         mcache_freefn_t mc_slab_free;   /* slab layer free callback */
 197         mcache_auditfn_t mc_slab_audit; /* slab layer audit callback */
 198         mcache_logfn_t mc_slab_log;     /* slab layer log callback */
 199         mcache_notifyfn_t mc_slab_notify; /* slab layer notify callback */
 200         void            *mc_private;    /* opaque arg to callbacks */
 201         size_t          mc_bufsize;     /* object size */
 202         size_t          mc_align;       /* object alignment */
 203         u_int32_t       mc_flags;       /* cache creation flags */
 204         u_int32_t       mc_purge_cnt;   /* # of purges requested by slab */
 205         u_int32_t       mc_enable_cnt;  /* # of reenables due to purges */
 206         u_int32_t       mc_waiter_cnt;  /* # of slab layer waiters */
 207         u_int32_t       mc_wretry_cnt;  /* # of wait retries */
 208         u_int32_t       mc_nwretry_cnt; /* # of no-wait retry attempts */
 209         u_int32_t       mc_nwfail_cnt;  /* # of no-wait retries that failed */
 210         decl_lck_mtx_data(, mc_sync_lock); /* protects purges and reenables */
 211         lck_attr_t      *mc_sync_lock_attr;
 212         lck_grp_t       *mc_sync_lock_grp;
 213         lck_grp_attr_t  *mc_sync_lock_grp_attr;
 214         /*
 215          * Keep CPU and buckets layers lock statistics separate.
 216          */
 217         lck_attr_t      *mc_cpu_lock_attr;
 218         lck_grp_t       *mc_cpu_lock_grp;
 219         lck_grp_attr_t  *mc_cpu_lock_grp_attr;
 220
 221         /*
 222          * Bucket layer common to all CPUs
 223          */
 224         decl_lck_mtx_data(, mc_bkt_lock);
 225         lck_attr_t      *mc_bkt_lock_attr;
 226         lck_grp_t       *mc_bkt_lock_grp;
 227         lck_grp_attr_t  *mc_bkt_lock_grp_attr;
 228         mcache_bkttype_t *cache_bkttype;        /* bucket type */
 229         mcache_bktlist_t mc_full;               /* full buckets */
 230         mcache_bktlist_t mc_empty;              /* empty buckets */
 231         size_t          mc_chunksize;           /* bufsize + alignment */
 232         u_int32_t       mc_bkt_contention;      /* lock contention count */
 233         u_int32_t       mc_bkt_contention_prev; /* previous snapshot */
 234
 235         /*
 236          * Per-CPU layer, aligned at cache line boundary
 237          */
 238         mcache_cpu_t    mc_cpu[1];
 239 } mcache_t;
 240
 241 #define MCACHE_ALIGN    8       /* default guaranteed alignment */
 242
 243 /* Valid values for mc_flags */
 244 #define MCF_VERIFY      0x00000001      /* enable verification */
 245 #define MCF_TRACE       0x00000002      /* enable transaction auditing */
 246 #define MCF_NOCPUCACHE  0x00000010      /* disable CPU layer caching */
 247 #define MCF_NOLEAKLOG   0x00000100      /* disable leak logging */
 248
 249 #define MCF_DEBUG       (MCF_VERIFY | MCF_TRACE)
 250 #define MCF_FLAGS_MASK  (MCF_DEBUG | MCF_NOCPUCACHE | MCF_NOLEAKLOG)
 251
 252 /* Valid values for notify callback */
 253 #define MCN_RETRYALLOC  0x00000001      /* Allocation should be retried */
 254
 255 #define MCACHE_STACK_DEPTH 16
 256
 257 typedef struct mcache_audit {
 258         struct mcache_audit *mca_next;  /* next audit struct */
 259         void            *mca_addr;      /* address of buffer */
 260         mcache_t        *mca_cache;     /* parent cache of the buffer */
 261         struct thread   *mca_thread;    /* thread doing transaction */
 262         struct thread   *mca_pthread;   /* previous transaction thread */
 263         size_t          mca_contents_size; /* size of contents */
 264         void            *mca_contents;  /* contents at last free */
 265         uint16_t        mca_depth;      /* pc stack depth */
 266         uint16_t        mca_pdepth;     /* previous transaction pc stack */
 267         void            *mca_stack[MCACHE_STACK_DEPTH];
 268         void            *mca_pstack[MCACHE_STACK_DEPTH];
 269         void            *mca_uptr;      /* user-specific pointer */
 270         uint32_t        mca_uflags;     /* user-specific flags */
 271 } mcache_audit_t;
 272
 273 __private_extern__ int assfail(const char *, const char *, int);
 274 __private_extern__ void mcache_init(void);
 275 __private_extern__ unsigned int mcache_getflags(void);
 276 __private_extern__ mcache_t *mcache_create(const char *, size_t,
 277     size_t, u_int32_t, int);
 278 __private_extern__ void *mcache_alloc(mcache_t *, int);
 279 __private_extern__ void mcache_free(mcache_t *, void *);
 280 __private_extern__ mcache_t *mcache_create_ext(const char *, size_t,
 281     mcache_allocfn_t, mcache_freefn_t, mcache_auditfn_t, mcache_logfn_t,
 282     mcache_notifyfn_t, void *, u_int32_t, int);
 283 __private_extern__ void mcache_destroy(mcache_t *);
 284 __private_extern__ unsigned int mcache_alloc_ext(mcache_t *, mcache_obj_t **,
 285     unsigned int, int);
 286 __private_extern__ void mcache_free_ext(mcache_t *, mcache_obj_t *);
 287 __private_extern__ void mcache_reap(void);
 288 __private_extern__ boolean_t mcache_purge_cache(mcache_t *);
 289 __private_extern__ void mcache_waiter_inc(mcache_t *);
 290 __private_extern__ void mcache_waiter_dec(mcache_t *);
 291 __private_extern__ boolean_t mcache_bkt_isempty(mcache_t *);
 292
 293 __private_extern__ void mcache_buffer_log(mcache_audit_t *, void *, mcache_t *);
 294 __private_extern__ void mcache_set_pattern(u_int64_t, void *, size_t);
 295 __private_extern__ void *mcache_verify_pattern(u_int64_t, void *, size_t);
 296 __private_extern__ void *mcache_verify_set_pattern(u_int64_t, u_int64_t,
 297     void *, size_t);
 298 __private_extern__ void mcache_audit_free_verify(mcache_audit_t *,
 299     void *, size_t, size_t);
 300 __private_extern__ void mcache_audit_free_verify_set(mcache_audit_t *,
 301     void *, size_t, size_t);
 302 __private_extern__ char *mcache_dump_mca(mcache_audit_t *);
 303 __private_extern__ void mcache_audit_panic(mcache_audit_t *, void *, size_t,
 304     int64_t, int64_t);
 305
 306 __private_extern__ mcache_t *mcache_audit_cache;
 307
 308 #ifdef  __cplusplus
 309 }
 310 #endif
 311
 312 #endif /* KERNEL_PRIVATE */
 313
 314 #endif /* _SYS_MCACHE_H */