include/sys/kmem.h

   1 /*
   2  *  This file is part of the SPL: Solaris Porting Layer.
   3  *
   4  *  Copyright (c) 2008 Lawrence Livermore National Security, LLC.
   5  *  Produced at Lawrence Livermore National Laboratory
   6  *  Written by:
   7  *          Brian Behlendorf <behlendorf1@llnl.gov>,
   8  *          Herb Wartens <wartens2@llnl.gov>,
   9  *          Jim Garlick <garlick@llnl.gov>
  10  *  UCRL-CODE-235197
  11  *
  12  *  This is free software; you can redistribute it and/or modify it
  13  *  under the terms of the GNU General Public License as published by
  14  *  the Free Software Foundation; either version 2 of the License, or
  15  *  (at your option) any later version.
  16  *
  17  *  This is distributed in the hope that it will be useful, but WITHOUT
  18  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  19  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  20  *  for more details.
  21  *
  22  *  You should have received a copy of the GNU General Public License along
  23  *  with this program; if not, write to the Free Software Foundation, Inc.,
  24  *  51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  25  */
  26
  27 #ifndef _SPL_KMEM_H
  28 #define _SPL_KMEM_H
  29
  30 #ifdef  __cplusplus
  31 extern "C" {
  32 #endif
  33
  34 #undef DEBUG_KMEM_UNIMPLEMENTED
  35
  36 #include <linux/module.h>
  37 #include <linux/slab.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/mm_compat.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/rwsem.h>
  42 #include <linux/hash.h>
  43 #include <linux/ctype.h>
  44 #include <asm/atomic_compat.h>
  45 #include <sys/types.h>
  46 #include <sys/debug.h>
  47 #include <sys/vmsystm.h>
  48
  49 /*
  50  * Memory allocation interfaces
  51  */
  52 #define KM_SLEEP                        GFP_KERNEL
  53 #define KM_NOSLEEP                      GFP_ATOMIC
  54 #undef  KM_PANIC                        /* No linux analog */
  55 #define KM_PUSHPAGE                     (KM_SLEEP | __GFP_HIGH)
  56 #define KM_VMFLAGS                      GFP_LEVEL_MASK
  57 #define KM_FLAGS                        __GFP_BITS_MASK
  58
  59 /*
  60  * Used internally, the kernel does not need to support this flag
  61  */
  62 #ifndef __GFP_ZERO
  63 # define __GFP_ZERO                     0x8000
  64 #endif
  65
  66 /*
  67  * __GFP_NOFAIL looks like it will be removed from the kernel perhaps as
  68  * early as 2.6.32.  To avoid this issue when it occurs in upstream kernels
  69  * we retry the allocation here as long as it is not __GFP_WAIT (GFP_ATOMIC).
  70  * I would prefer the caller handle the failure case cleanly but we are
  71  * trying to emulate Solaris and those are not the Solaris semantics.
  72  */
  73 static inline void *
  74 kmalloc_nofail(size_t size, gfp_t flags)
  75 {
  76         void *ptr;
  77
  78         do {
  79                 ptr = kmalloc(size, flags);
  80         } while (ptr == NULL && (flags & __GFP_WAIT));
  81
  82         return ptr;
  83 }
  84
  85 static inline void *
  86 kzalloc_nofail(size_t size, gfp_t flags)
  87 {
  88         void *ptr;
  89
  90         do {
  91                 ptr = kzalloc(size, flags);
  92         } while (ptr == NULL && (flags & __GFP_WAIT));
  93
  94         return ptr;
  95 }
  96
  97 #ifdef HAVE_KMALLOC_NODE
  98 static inline void *
  99 kmalloc_node_nofail(size_t size, gfp_t flags, int node)
 100 {
 101         void *ptr;
 102
 103         do {
 104                 ptr = kmalloc_node(size, flags, node);
 105         } while (ptr == NULL && (flags & __GFP_WAIT));
 106
 107         return ptr;
 108 }
 109 #endif /* HAVE_KMALLOC_NODE */
 110
 111 #ifdef DEBUG_KMEM
 112
 113 extern atomic64_t kmem_alloc_used;
 114 extern unsigned long long kmem_alloc_max;
 115 extern atomic64_t vmem_alloc_used;
 116 extern unsigned long long vmem_alloc_max;
 117
 118 # define kmem_alloc(size, flags)             __kmem_alloc((size), (flags), 0, 0)
 119 # define kmem_zalloc(size, flags)            __kmem_alloc((size), ((flags) |  \
 120                                                  __GFP_ZERO), 0, 0)
 121
 122 /* The node alloc functions are only used by the SPL code itself */
 123 # ifdef HAVE_KMALLOC_NODE
 124 #  define kmem_alloc_node(size, flags, node) __kmem_alloc((size), (flags), 1, \
 125                                                  node)
 126 # else
 127 #  define kmem_alloc_node(size, flags, node) __kmem_alloc((size), (flags), 0, 0)
 128 # endif
 129
 130 # define vmem_zalloc(size, flags)            vmem_alloc((size), ((flags) |    \
 131                                                  __GFP_ZERO))
 132
 133 # ifdef DEBUG_KMEM_TRACKING
 134
 135 extern void *kmem_alloc_track(size_t size, int flags, const char *func,
 136     int line, int node_alloc, int node);
 137 extern void kmem_free_track(void *ptr, size_t size);
 138 extern void *vmem_alloc_track(size_t size, int flags, const char *func,
 139     int line);
 140 extern void vmem_free_track(void *ptr, size_t size);
 141
 142 #  define __kmem_alloc(size, flags, na, node) kmem_alloc_track((size),        \
 143                                                   (flags), __FUNCTION__,      \
 144                                                   __LINE__, (na), (node))
 145 #  define kmem_free(ptr, size)                kmem_free_track((ptr), (size))
 146 #  define vmem_alloc(size, flags)             vmem_alloc_track((size),        \
 147                                                   (flags),__FUNCTION__,       \
 148                                                   __LINE__)
 149 #  define vmem_free(ptr, size)                vmem_free_track((ptr), (size))
 150
 151 # else /* DEBUG_KMEM_TRACKING */
 152
 153 extern void *kmem_alloc_debug(size_t size, int flags, const char *func,
 154     int line, int node_alloc, int node);
 155 extern void kmem_free_debug(void *ptr, size_t size);
 156 extern void *vmem_alloc_debug(size_t size, int flags, const char *func,
 157     int line);
 158 extern void vmem_free_debug(void *ptr, size_t size);
 159
 160 #  define __kmem_alloc(size, flags, na, node) kmem_alloc_debug((size),        \
 161                                                   (flags), __FUNCTION__,      \
 162                                                   __LINE__, (na), (node))
 163 #  define kmem_free(ptr, size)                kmem_free_debug((ptr), (size))
 164 #  define vmem_alloc(size, flags)             vmem_alloc_debug((size),        \
 165                                                   (flags), __FUNCTION__,      \
 166                                                   __LINE__)
 167 #  define vmem_free(ptr, size)                vmem_free_debug((ptr), (size))
 168
 169 # endif /* DEBUG_KMEM_TRACKING */
 170
 171 #else /* DEBUG_KMEM */
 172
 173 # define kmem_alloc(size, flags)              kmalloc_nofail((size), (flags))
 174 # define kmem_zalloc(size, flags)             kzalloc_nofail((size), (flags))
 175 # define kmem_free(ptr, size)                 ((void)(size), kfree(ptr))
 176
 177 # ifdef HAVE_KMALLOC_NODE
 178 #  define kmem_alloc_node(size, flags, node)                                  \
 179           kmalloc_node_nofail((size), (flags), (node))
 180 # else
 181 #  define kmem_alloc_node(size, flags, node)                                  \
 182           kmalloc_nofail((size), (flags))
 183 # endif
 184
 185 # define vmem_alloc(size, flags)              __vmalloc((size), ((flags) |    \
 186                                                   __GFP_HIGHMEM), PAGE_KERNEL)
 187 # define vmem_zalloc(size, flags)                                             \
 188 ({                                                                            \
 189         void *_ptr_ = __vmalloc((size),((flags)|__GFP_HIGHMEM),PAGE_KERNEL);  \
 190         if (_ptr_)                                                            \
 191                 memset(_ptr_, 0, (size));                                     \
 192         _ptr_;                                                                \
 193 })
 194 # define vmem_free(ptr, size)           ((void)(size), vfree(ptr))
 195
 196 #endif /* DEBUG_KMEM */
 197
 198 #ifdef DEBUG_KMEM_UNIMPLEMENTED
 199 static __inline__ void *
 200 kmem_alloc_tryhard(size_t size, size_t *alloc_size, int kmflags)
 201 {
 202 #error "kmem_alloc_tryhard() not implemented"
 203 }
 204 #endif /* DEBUG_KMEM_UNIMPLEMENTED */
 205
 206 /*
 207  * Slab allocation interfaces
 208  */
 209 enum {
 210         KMC_BIT_NOTOUCH         = 0,    /* Don't update ages */
 211         KMC_BIT_NODEBUG         = 1,    /* Default behavior */
 212         KMC_BIT_NOMAGAZINE      = 2,    /* XXX: Unsupported */
 213         KMC_BIT_NOHASH          = 3,    /* XXX: Unsupported */
 214         KMC_BIT_QCACHE          = 4,    /* XXX: Unsupported */
 215         KMC_BIT_KMEM            = 5,    /* Use kmem cache */
 216         KMC_BIT_VMEM            = 6,    /* Use vmem cache */
 217         KMC_BIT_OFFSLAB         = 7,    /* Objects not on slab */
 218         KMC_BIT_REAPING         = 16,   /* Reaping in progress */
 219         KMC_BIT_DESTROY         = 17,   /* Destroy in progress */
 220 };
 221
 222 #define KMC_NOTOUCH             (1 << KMC_BIT_NOTOUCH)
 223 #define KMC_NODEBUG             (1 << KMC_BIT_NODEBUG)
 224 #define KMC_NOMAGAZINE          (1 << KMC_BIT_NOMAGAZINE)
 225 #define KMC_NOHASH              (1 << KMC_BIT_NOHASH)
 226 #define KMC_QCACHE              (1 << KMC_BIT_QCACHE)
 227 #define KMC_KMEM                (1 << KMC_BIT_KMEM)
 228 #define KMC_VMEM                (1 << KMC_BIT_VMEM)
 229 #define KMC_OFFSLAB             (1 << KMC_BIT_OFFSLAB)
 230 #define KMC_REAPING             (1 << KMC_BIT_REAPING)
 231 #define KMC_DESTROY             (1 << KMC_BIT_DESTROY)
 232
 233 #define KMC_REAP_CHUNK                  INT_MAX
 234 #define KMC_DEFAULT_SEEKS               1
 235
 236 #ifdef DEBUG_KMEM_UNIMPLEMENTED
 237 static __inline__ void kmem_init(void) {
 238 #error "kmem_init() not implemented"
 239 }
 240
 241 static __inline__ void kmem_thread_init(void) {
 242 #error "kmem_thread_init() not implemented"
 243 }
 244
 245 static __inline__ void kmem_mp_init(void) {
 246 #error "kmem_mp_init() not implemented"
 247 }
 248
 249 static __inline__ void kmem_reap_idspace(void) {
 250 #error "kmem_reap_idspace() not implemented"
 251 }
 252
 253 static __inline__ size_t kmem_avail(void) {
 254 #error "kmem_avail() not implemented"
 255 }
 256
 257 static __inline__ size_t kmem_maxavail(void) {
 258 #error "kmem_maxavail() not implemented"
 259 }
 260
 261 static __inline__ uint64_t kmem_cache_stat(spl_kmem_cache_t *cache) {
 262 #error "kmem_cache_stat() not implemented"
 263 }
 264 #endif /* DEBUG_KMEM_UNIMPLEMENTED */
 265
 266 /* XXX - Used by arc.c to adjust its memory footprint. We may want
 267  *       to use this hook in the future to adjust behavior based on
 268  *       debug levels.  For now it's safe to always return 0.
 269  */
 270 static __inline__ int
 271 kmem_debugging(void)
 272 {
 273         return 0;
 274 }
 275
 276 extern int kmem_set_warning(int flag);
 277
 278 extern struct list_head spl_kmem_cache_list;
 279 extern struct rw_semaphore spl_kmem_cache_sem;
 280
 281 #define SKM_MAGIC                       0x2e2e2e2e
 282 #define SKO_MAGIC                       0x20202020
 283 #define SKS_MAGIC                       0x22222222
 284 #define SKC_MAGIC                       0x2c2c2c2c
 285
 286 #define SPL_KMEM_CACHE_DELAY            15      /* Minimum slab release age */
 287 #define SPL_KMEM_CACHE_REAP             0       /* Default reap everything */
 288 #define SPL_KMEM_CACHE_OBJ_PER_SLAB     32      /* Target objects per slab */
 289 #define SPL_KMEM_CACHE_OBJ_PER_SLAB_MIN 8       /* Minimum objects per slab */
 290 #define SPL_KMEM_CACHE_ALIGN            8       /* Default object alignment */
 291
 292 typedef int (*spl_kmem_ctor_t)(void *, void *, int);
 293 typedef void (*spl_kmem_dtor_t)(void *, void *);
 294 typedef void (*spl_kmem_reclaim_t)(void *);
 295
 296 typedef struct spl_kmem_magazine {
 297         uint32_t                skm_magic;      /* Sanity magic */
 298         uint32_t                skm_avail;      /* Available objects */
 299         uint32_t                skm_size;       /* Magazine size */
 300         uint32_t                skm_refill;     /* Batch refill size */
 301         struct spl_kmem_cache   *skm_cache;     /* Owned by cache */
 302         struct delayed_work     skm_work;       /* Magazine reclaim work */
 303         unsigned long           skm_age;        /* Last cache access */
 304         void                    *skm_objs[0];   /* Object pointers */
 305 } spl_kmem_magazine_t;
 306
 307 typedef struct spl_kmem_obj {
 308         uint32_t                sko_magic;      /* Sanity magic */
 309         void                    *sko_addr;      /* Buffer address */
 310         struct spl_kmem_slab    *sko_slab;      /* Owned by slab */
 311         struct list_head        sko_list;       /* Free object list linkage */
 312 } spl_kmem_obj_t;
 313
 314 typedef struct spl_kmem_slab {
 315         uint32_t                sks_magic;      /* Sanity magic */
 316         uint32_t                sks_objs;       /* Objects per slab */
 317         struct spl_kmem_cache   *sks_cache;     /* Owned by cache */
 318         struct list_head        sks_list;       /* Slab list linkage */
 319         struct list_head        sks_free_list;  /* Free object list */
 320         unsigned long           sks_age;        /* Last modify jiffie */
 321         uint32_t                sks_ref;        /* Ref count used objects */
 322 } spl_kmem_slab_t;
 323
 324 typedef struct spl_kmem_cache {
 325         uint32_t                skc_magic;      /* Sanity magic */
 326         uint32_t                skc_name_size;  /* Name length */
 327         char                    *skc_name;      /* Name string */
 328         spl_kmem_magazine_t     *skc_mag[NR_CPUS]; /* Per-CPU warm cache */
 329         uint32_t                skc_mag_size;   /* Magazine size */
 330         uint32_t                skc_mag_refill; /* Magazine refill count */
 331         spl_kmem_ctor_t         skc_ctor;       /* Constructor */
 332         spl_kmem_dtor_t         skc_dtor;       /* Destructor */
 333         spl_kmem_reclaim_t      skc_reclaim;    /* Reclaimator */
 334         void                    *skc_private;   /* Private data */
 335         void                    *skc_vmp;       /* Unused */
 336         unsigned long           skc_flags;      /* Flags */
 337         uint32_t                skc_obj_size;   /* Object size */
 338         uint32_t                skc_obj_align;  /* Object alignment */
 339         uint32_t                skc_slab_objs;  /* Objects per slab */
 340         uint32_t                skc_slab_size;  /* Slab size */
 341         uint32_t                skc_delay;      /* Slab reclaim interval */
 342         uint32_t                skc_reap;       /* Slab reclaim count */
 343         atomic_t                skc_ref;        /* Ref count callers */
 344         struct delayed_work     skc_work;       /* Slab reclaim work */
 345         struct list_head        skc_list;       /* List of caches linkage */
 346         struct list_head        skc_complete_list;/* Completely alloc'ed */
 347         struct list_head        skc_partial_list; /* Partially alloc'ed */
 348         spinlock_t              skc_lock;       /* Cache lock */
 349         uint64_t                skc_slab_fail;  /* Slab alloc failures */
 350         uint64_t                skc_slab_create;/* Slab creates */
 351         uint64_t                skc_slab_destroy;/* Slab destroys */
 352         uint64_t                skc_slab_total; /* Slab total current */
 353         uint64_t                skc_slab_alloc; /* Slab alloc current */
 354         uint64_t                skc_slab_max;   /* Slab max historic  */
 355         uint64_t                skc_obj_total;  /* Obj total current */
 356         uint64_t                skc_obj_alloc;  /* Obj alloc current */
 357         uint64_t                skc_obj_max;    /* Obj max historic */
 358 } spl_kmem_cache_t;
 359 #define kmem_cache_t            spl_kmem_cache_t
 360
 361 extern spl_kmem_cache_t *
 362 spl_kmem_cache_create(char *name, size_t size, size_t align,
 363         spl_kmem_ctor_t ctor, spl_kmem_dtor_t dtor, spl_kmem_reclaim_t reclaim,
 364         void *priv, void *vmp, int flags);
 365
 366 extern void spl_kmem_cache_destroy(spl_kmem_cache_t *skc);
 367 extern void *spl_kmem_cache_alloc(spl_kmem_cache_t *skc, int flags);
 368 extern void spl_kmem_cache_free(spl_kmem_cache_t *skc, void *obj);
 369 extern void spl_kmem_cache_reap_now(spl_kmem_cache_t *skc);
 370 extern void spl_kmem_reap(void);
 371
 372 int spl_kmem_init_kallsyms_lookup(void);
 373 int spl_kmem_init(void);
 374 void spl_kmem_fini(void);
 375
 376 #define kmem_cache_create(name,size,align,ctor,dtor,rclm,priv,vmp,flags) \
 377         spl_kmem_cache_create(name,size,align,ctor,dtor,rclm,priv,vmp,flags)
 378 #define kmem_cache_destroy(skc)         spl_kmem_cache_destroy(skc)
 379 #define kmem_cache_alloc(skc, flags)    spl_kmem_cache_alloc(skc, flags)
 380 #define kmem_cache_free(skc, obj)       spl_kmem_cache_free(skc, obj)
 381 #define kmem_cache_reap_now(skc)        spl_kmem_cache_reap_now(skc)
 382 #define kmem_reap()                     spl_kmem_reap()
 383 #define kmem_virt(ptr)                  (((ptr) >= (void *)VMALLOC_START) && \
 384                                          ((ptr) <  (void *)VMALLOC_END))
 385
 386 #ifdef  __cplusplus
 387 }
 388 #endif
 389
 390 #endif  /* _SPL_KMEM_H */