zfs/lib/libumem/include/umem_impl.h

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License, Version 1.0 only
   6  * (the "License").  You may not use this file except in compliance
   7  * with the License.
   8  *
   9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
  10  * or http://www.opensolaris.org/os/licensing.
  11  * See the License for the specific language governing permissions
  12  * and limitations under the License.
  13  *
  14  * When distributing Covered Code, include this CDDL HEADER in each
  15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  16  * If applicable, add the following below this CDDL HEADER, with the
  17  * fields enclosed by brackets "[]" replaced with your own identifying
  18  * information: Portions Copyright [yyyy] [name of copyright owner]
  19  *
  20  * CDDL HEADER END
  21  */
  22 /*
  23  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26
  27 #ifndef _UMEM_IMPL_H
  28 #define _UMEM_IMPL_H
  29
  30 #pragma ident   "%Z%%M% %I%     %E% SMI"
  31
  32 #include <umem.h>
  33
  34 #include <sys/sysmacros.h>
  35 #include <sys/time.h>
  36 #include <sys/vmem.h>
  37 #include <thread.h>
  38
  39 #ifdef  __cplusplus
  40 extern "C" {
  41 #endif
  42
  43 /*
  44  * umem memory allocator: implementation-private data structures
  45  */
  46
  47 /*
  48  * Internal flags for umem_cache_create
  49  */
  50 #define UMC_QCACHE      0x00100000
  51 #define UMC_INTERNAL    0x80000000
  52
  53 /*
  54  * Cache flags
  55  */
  56 #define UMF_AUDIT       0x00000001      /* transaction auditing */
  57 #define UMF_DEADBEEF    0x00000002      /* deadbeef checking */
  58 #define UMF_REDZONE     0x00000004      /* redzone checking */
  59 #define UMF_CONTENTS    0x00000008      /* freed-buffer content logging */
  60 #define UMF_CHECKSIGNAL 0x00000010      /* abort when in signal context */
  61 #define UMF_NOMAGAZINE  0x00000020      /* disable per-cpu magazines */
  62 #define UMF_FIREWALL    0x00000040      /* put all bufs before unmapped pages */
  63 #define UMF_LITE        0x00000100      /* lightweight debugging */
  64
  65 #define UMF_HASH        0x00000200      /* cache has hash table */
  66 #define UMF_RANDOMIZE   0x00000400      /* randomize other umem_flags */
  67
  68 #define UMF_BUFTAG      (UMF_DEADBEEF | UMF_REDZONE)
  69 #define UMF_TOUCH       (UMF_BUFTAG | UMF_LITE | UMF_CONTENTS)
  70 #define UMF_RANDOM      (UMF_TOUCH | UMF_AUDIT | UMF_NOMAGAZINE)
  71 #define UMF_DEBUG       (UMF_RANDOM | UMF_FIREWALL)
  72
  73 #define UMEM_STACK_DEPTH        umem_stack_depth
  74
  75 #define UMEM_FREE_PATTERN               0xdeadbeefdeadbeefULL
  76 #define UMEM_UNINITIALIZED_PATTERN      0xbaddcafebaddcafeULL
  77 #define UMEM_REDZONE_PATTERN            0xfeedfacefeedfaceULL
  78 #define UMEM_REDZONE_BYTE               0xbb
  79
  80 #define UMEM_FATAL_FLAGS        (UMEM_NOFAIL)
  81 #define UMEM_SLEEP_FLAGS        (0)
  82
  83 /*
  84  * Redzone size encodings for umem_alloc() / umem_free().  We encode the
  85  * allocation size, rather than storing it directly, so that umem_free()
  86  * can distinguish frees of the wrong size from redzone violations.
  87  */
  88 #define UMEM_SIZE_ENCODE(x)     (251 * (x) + 1)
  89 #define UMEM_SIZE_DECODE(x)     ((x) / 251)
  90 #define UMEM_SIZE_VALID(x)      ((x) % 251 == 1)
  91
  92 /*
  93  * The bufctl (buffer control) structure keeps some minimal information
  94  * about each buffer: its address, its slab, and its current linkage,
  95  * which is either on the slab's freelist (if the buffer is free), or
  96  * on the cache's buf-to-bufctl hash table (if the buffer is allocated).
  97  * In the case of non-hashed, or "raw", caches (the common case), only
  98  * the freelist linkage is necessary: the buffer address is at a fixed
  99  * offset from the bufctl address, and the slab is at the end of the page.
 100  *
 101  * NOTE: bc_next must be the first field; raw buffers have linkage only.
 102  */
 103 typedef struct umem_bufctl {
 104         struct umem_bufctl      *bc_next;       /* next bufctl struct */
 105         void                    *bc_addr;       /* address of buffer */
 106         struct umem_slab        *bc_slab;       /* controlling slab */
 107 } umem_bufctl_t;
 108
 109 /*
 110  * The UMF_AUDIT version of the bufctl structure.  The beginning of this
 111  * structure must be identical to the normal bufctl structure so that
 112  * pointers are interchangeable.
 113  */
 114
 115 #define UMEM_BUFCTL_AUDIT_SIZE_DEPTH(frames) \
 116         ((size_t)(&((umem_bufctl_audit_t *)0)->bc_stack[frames]))
 117
 118 /*
 119  * umem_bufctl_audits must be allocated from a UMC_NOHASH cache, so we
 120  * require that 2 of them, plus 2 buftags, plus a umem_slab_t, all fit on
 121  * a single page.
 122  *
 123  * For ILP32, this is about 1000 frames.
 124  * For LP64, this is about 490 frames.
 125  */
 126
 127 #define UMEM_BUFCTL_AUDIT_ALIGN 32
 128
 129 #define UMEM_BUFCTL_AUDIT_MAX_SIZE                                      \
 130         (P2ALIGN((PAGESIZE - sizeof (umem_slab_t))/2 -                  \
 131             sizeof (umem_buftag_t), UMEM_BUFCTL_AUDIT_ALIGN))
 132
 133 #define UMEM_MAX_STACK_DEPTH                                            \
 134         ((UMEM_BUFCTL_AUDIT_MAX_SIZE -                                  \
 135             UMEM_BUFCTL_AUDIT_SIZE_DEPTH(0)) / sizeof (uintptr_t))
 136
 137 typedef struct umem_bufctl_audit {
 138         struct umem_bufctl      *bc_next;       /* next bufctl struct */
 139         void                    *bc_addr;       /* address of buffer */
 140         struct umem_slab        *bc_slab;       /* controlling slab */
 141         umem_cache_t            *bc_cache;      /* controlling cache */
 142         hrtime_t                bc_timestamp;   /* transaction time */
 143         thread_t                bc_thread;      /* thread doing transaction */
 144         struct umem_bufctl      *bc_lastlog;    /* last log entry */
 145         void                    *bc_contents;   /* contents at last free */
 146         int                     bc_depth;       /* stack depth */
 147         uintptr_t               bc_stack[1];    /* pc stack */
 148 } umem_bufctl_audit_t;
 149
 150 #define UMEM_LOCAL_BUFCTL_AUDIT(bcpp)                                   \
 151                 *(bcpp) = (umem_bufctl_audit_t *)                       \
 152                     alloca(UMEM_BUFCTL_AUDIT_SIZE)
 153
 154 #define UMEM_BUFCTL_AUDIT_SIZE                                          \
 155         UMEM_BUFCTL_AUDIT_SIZE_DEPTH(UMEM_STACK_DEPTH)
 156
 157 /*
 158  * A umem_buftag structure is appended to each buffer whenever any of the
 159  * UMF_BUFTAG flags (UMF_DEADBEEF, UMF_REDZONE, UMF_VERIFY) are set.
 160  */
 161 typedef struct umem_buftag {
 162         uint64_t                bt_redzone;     /* 64-bit redzone pattern */
 163         umem_bufctl_t           *bt_bufctl;     /* bufctl */
 164         intptr_t                bt_bxstat;      /* bufctl ^ (alloc/free) */
 165 } umem_buftag_t;
 166
 167 #define UMEM_BUFTAG(cp, buf)            \
 168         ((umem_buftag_t *)((char *)(buf) + (cp)->cache_buftag))
 169
 170 #define UMEM_BUFCTL(cp, buf)            \
 171         ((umem_bufctl_t *)((char *)(buf) + (cp)->cache_bufctl))
 172
 173 #define UMEM_BUF(cp, bcp)               \
 174         ((void *)((char *)(bcp) - (cp)->cache_bufctl))
 175
 176 #define UMEM_SLAB(cp, buf)              \
 177         ((umem_slab_t *)P2END((uintptr_t)(buf), (cp)->cache_slabsize) - 1)
 178
 179 #define UMEM_CPU_CACHE(cp, cpu)         \
 180         (umem_cpu_cache_t *)((char *)cp + cpu->cpu_cache_offset)
 181
 182 #define UMEM_MAGAZINE_VALID(cp, mp)     \
 183         (((umem_slab_t *)P2END((uintptr_t)(mp), PAGESIZE) - 1)->slab_cache == \
 184             (cp)->cache_magtype->mt_cache)
 185
 186 #define UMEM_SLAB_MEMBER(sp, buf)       \
 187         ((size_t)(buf) - (size_t)(sp)->slab_base < \
 188             (sp)->slab_cache->cache_slabsize)
 189
 190 #define UMEM_BUFTAG_ALLOC       0xa110c8edUL
 191 #define UMEM_BUFTAG_FREE        0xf4eef4eeUL
 192
 193 typedef struct umem_slab {
 194         struct umem_cache       *slab_cache;    /* controlling cache */
 195         void                    *slab_base;     /* base of allocated memory */
 196         struct umem_slab        *slab_next;     /* next slab on freelist */
 197         struct umem_slab        *slab_prev;     /* prev slab on freelist */
 198         struct umem_bufctl      *slab_head;     /* first free buffer */
 199         long                    slab_refcnt;    /* outstanding allocations */
 200         long                    slab_chunks;    /* chunks (bufs) in this slab */
 201 } umem_slab_t;
 202
 203 #define UMEM_HASH_INITIAL       64
 204
 205 #define UMEM_HASH(cp, buf)      \
 206         ((cp)->cache_hash_table +       \
 207         (((uintptr_t)(buf) >> (cp)->cache_hash_shift) & (cp)->cache_hash_mask))
 208
 209 typedef struct umem_magazine {
 210         void    *mag_next;
 211         void    *mag_round[1];          /* one or more rounds */
 212 } umem_magazine_t;
 213
 214 /*
 215  * The magazine types for fast per-cpu allocation
 216  */
 217 typedef struct umem_magtype {
 218         int             mt_magsize;     /* magazine size (number of rounds) */
 219         int             mt_align;       /* magazine alignment */
 220         size_t          mt_minbuf;      /* all smaller buffers qualify */
 221         size_t          mt_maxbuf;      /* no larger buffers qualify */
 222         umem_cache_t    *mt_cache;      /* magazine cache */
 223 } umem_magtype_t;
 224
 225 #define UMEM_CPU_CACHE_SIZE     64      /* must be power of 2 */
 226 #define UMEM_CPU_PAD            (UMEM_CPU_CACHE_SIZE - sizeof (mutex_t) - \
 227         2 * sizeof (uint_t) - 2 * sizeof (void *) - 4 * sizeof (int))
 228 #define UMEM_CACHE_SIZE(ncpus)  \
 229         ((size_t)(&((umem_cache_t *)0)->cache_cpu[ncpus]))
 230
 231 typedef struct umem_cpu_cache {
 232         mutex_t         cc_lock;        /* protects this cpu's local cache */
 233         uint_t          cc_alloc;       /* allocations from this cpu */
 234         uint_t          cc_free;        /* frees to this cpu */
 235         umem_magazine_t *cc_loaded;     /* the currently loaded magazine */
 236         umem_magazine_t *cc_ploaded;    /* the previously loaded magazine */
 237         int             cc_rounds;      /* number of objects in loaded mag */
 238         int             cc_prounds;     /* number of objects in previous mag */
 239         int             cc_magsize;     /* number of rounds in a full mag */
 240         int             cc_flags;       /* CPU-local copy of cache_flags */
 241 #ifndef _LP64
 242         char            cc_pad[UMEM_CPU_PAD]; /* for nice alignment (32-bit) */
 243 #endif
 244 } umem_cpu_cache_t;
 245
 246 /*
 247  * The magazine lists used in the depot.
 248  */
 249 typedef struct umem_maglist {
 250         umem_magazine_t *ml_list;       /* magazine list */
 251         long            ml_total;       /* number of magazines */
 252         long            ml_min;         /* min since last update */
 253         long            ml_reaplimit;   /* max reapable magazines */
 254         uint64_t        ml_alloc;       /* allocations from this list */
 255 } umem_maglist_t;
 256
 257 #define UMEM_CACHE_NAMELEN      31
 258
 259 struct umem_cache {
 260         /*
 261          * Statistics
 262          */
 263         uint64_t        cache_slab_create;      /* slab creates */
 264         uint64_t        cache_slab_destroy;     /* slab destroys */
 265         uint64_t        cache_slab_alloc;       /* slab layer allocations */
 266         uint64_t        cache_slab_free;        /* slab layer frees */
 267         uint64_t        cache_alloc_fail;       /* total failed allocations */
 268         uint64_t        cache_buftotal;         /* total buffers */
 269         uint64_t        cache_bufmax;           /* max buffers ever */
 270         uint64_t        cache_rescale;          /* # of hash table rescales */
 271         uint64_t        cache_lookup_depth;     /* hash lookup depth */
 272         uint64_t        cache_depot_contention; /* mutex contention count */
 273         uint64_t        cache_depot_contention_prev; /* previous snapshot */
 274
 275         /*
 276          * Cache properties
 277          */
 278         char            cache_name[UMEM_CACHE_NAMELEN + 1];
 279         size_t          cache_bufsize;          /* object size */
 280         size_t          cache_align;            /* object alignment */
 281         umem_constructor_t *cache_constructor;
 282         umem_destructor_t *cache_destructor;
 283         umem_reclaim_t  *cache_reclaim;
 284         void            *cache_private;         /* opaque arg to callbacks */
 285         vmem_t          *cache_arena;           /* vmem source for slabs */
 286         int             cache_cflags;           /* cache creation flags */
 287         int             cache_flags;            /* various cache state info */
 288         int             cache_uflags;           /* UMU_* flags */
 289         uint32_t        cache_mtbf;             /* induced alloc failure rate */
 290         umem_cache_t    *cache_next;            /* forward cache linkage */
 291         umem_cache_t    *cache_prev;            /* backward cache linkage */
 292         umem_cache_t    *cache_unext;           /* next in update list */
 293         umem_cache_t    *cache_uprev;           /* prev in update list */
 294         uint32_t        cache_cpu_mask;         /* mask for cpu offset */
 295
 296         /*
 297          * Slab layer
 298          */
 299         mutex_t         cache_lock;             /* protects slab layer */
 300         size_t          cache_chunksize;        /* buf + alignment [+ debug] */
 301         size_t          cache_slabsize;         /* size of a slab */
 302         size_t          cache_bufctl;           /* buf-to-bufctl distance */
 303         size_t          cache_buftag;           /* buf-to-buftag distance */
 304         size_t          cache_verify;           /* bytes to verify */
 305         size_t          cache_contents;         /* bytes of saved content */
 306         size_t          cache_color;            /* next slab color */
 307         size_t          cache_mincolor;         /* maximum slab color */
 308         size_t          cache_maxcolor;         /* maximum slab color */
 309         size_t          cache_hash_shift;       /* get to interesting bits */
 310         size_t          cache_hash_mask;        /* hash table mask */
 311         umem_slab_t     *cache_freelist;        /* slab free list */
 312         umem_slab_t     cache_nullslab;         /* end of freelist marker */
 313         umem_cache_t    *cache_bufctl_cache;    /* source of bufctls */
 314         umem_bufctl_t   **cache_hash_table;     /* hash table base */
 315         /*
 316          * Depot layer
 317          */
 318         mutex_t         cache_depot_lock;       /* protects depot */
 319         umem_magtype_t  *cache_magtype;         /* magazine type */
 320         umem_maglist_t  cache_full;             /* full magazines */
 321         umem_maglist_t  cache_empty;            /* empty magazines */
 322
 323         /*
 324          * Per-CPU layer
 325          */
 326         umem_cpu_cache_t cache_cpu[1];          /* cache_cpu_mask + 1 entries */
 327 };
 328
 329 typedef struct umem_cpu_log_header {
 330         mutex_t         clh_lock;
 331         char            *clh_current;
 332         size_t          clh_avail;
 333         int             clh_chunk;
 334         int             clh_hits;
 335         char            clh_pad[64 - sizeof (mutex_t) - sizeof (char *) -
 336                                 sizeof (size_t) - 2 * sizeof (int)];
 337 } umem_cpu_log_header_t;
 338
 339 typedef struct umem_log_header {
 340         mutex_t         lh_lock;
 341         char            *lh_base;
 342         int             *lh_free;
 343         size_t          lh_chunksize;
 344         int             lh_nchunks;
 345         int             lh_head;
 346         int             lh_tail;
 347         int             lh_hits;
 348         umem_cpu_log_header_t lh_cpu[1];        /* actually umem_max_ncpus */
 349 } umem_log_header_t;
 350
 351 typedef struct umem_cpu {
 352         uint32_t cpu_cache_offset;
 353         uint32_t cpu_number;
 354 } umem_cpu_t;
 355
 356 #define UMEM_MAXBUF     16384
 357
 358 #define UMEM_ALIGN              8       /* min guaranteed alignment */
 359 #define UMEM_ALIGN_SHIFT        3       /* log2(UMEM_ALIGN) */
 360 #define UMEM_VOID_FRACTION      8       /* never waste more than 1/8 of slab */
 361
 362 /*
 363  * For 64 bits, buffers >= 16 bytes must be 16-byte aligned
 364  */
 365 #ifdef _LP64
 366 #define UMEM_SECOND_ALIGN 16
 367 #else
 368 #define UMEM_SECOND_ALIGN UMEM_ALIGN
 369 #endif
 370
 371 #define MALLOC_MAGIC                    0x3a10c000 /* 8-byte tag */
 372 #define MEMALIGN_MAGIC                  0x3e3a1000
 373
 374 #ifdef _LP64
 375 #define MALLOC_SECOND_MAGIC             0x16ba7000 /* 8-byte tag, 16-aligned */
 376 #define MALLOC_OVERSIZE_MAGIC           0x06e47000 /* 16-byte tag, _LP64 */
 377 #endif
 378
 379 #define UMEM_MALLOC_ENCODE(type, sz)    (uint32_t)((type) - (sz))
 380 #define UMEM_MALLOC_DECODE(stat, sz)    (uint32_t)((stat) + (sz))
 381 #define UMEM_FREE_PATTERN_32            (uint32_t)(UMEM_FREE_PATTERN)
 382
 383 #define UMU_MAGAZINE_RESIZE     0x00000001
 384 #define UMU_HASH_RESCALE        0x00000002
 385 #define UMU_REAP                0x00000004
 386 #define UMU_NOTIFY              0x08000000
 387 #define UMU_ACTIVE              0x80000000
 388
 389 #define UMEM_READY_INIT_FAILED          -1
 390 #define UMEM_READY_STARTUP              1
 391 #define UMEM_READY_INITING              2
 392 #define UMEM_READY                      3
 393
 394 #ifdef UMEM_STANDALONE
 395 extern void umem_startup(caddr_t, size_t, size_t, caddr_t, caddr_t);
 396 extern int umem_add(caddr_t, size_t);
 397 #endif
 398
 399 #ifdef  __cplusplus
 400 }
 401 #endif
 402
 403 #endif  /* _UMEM_IMPL_H */