mm/slab_common.c

   1 /*
   2  * Slab allocator functions that are independent of the allocator strategy
   3  *
   4  * (C) 2012 Christoph Lameter <cl@linux.com>
   5  */
   6 #include <linux/slab.h>
   7
   8 #include <linux/mm.h>
   9 #include <linux/poison.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/memory.h>
  12 #include <linux/compiler.h>
  13 #include <linux/module.h>
  14 #include <linux/cpu.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/proc_fs.h>
  18 #include <asm/cacheflush.h>
  19 #include <asm/tlbflush.h>
  20 #include <asm/page.h>
  21
  22 #include "slab.h"
  23
  24 enum slab_state slab_state;
  25 LIST_HEAD(slab_caches);
  26 DEFINE_MUTEX(slab_mutex);
  27 struct kmem_cache *kmem_cache;
  28
  29 #ifdef CONFIG_DEBUG_VM
  30 static int kmem_cache_sanity_check(const char *name, size_t size)
  31 {
  32         struct kmem_cache *s = NULL;
  33
  34         if (!name || in_interrupt() || size < sizeof(void *) ||
  35                 size > KMALLOC_MAX_SIZE) {
  36                 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
  37                 return -EINVAL;
  38         }
  39
  40         list_for_each_entry(s, &slab_caches, list) {
  41                 char tmp;
  42                 int res;
  43
  44                 /*
  45                  * This happens when the module gets unloaded and doesn't
  46                  * destroy its slab cache and no-one else reuses the vmalloc
  47                  * area of the module.  Print a warning.
  48                  */
  49                 res = probe_kernel_address(s->name, tmp);
  50                 if (res) {
  51                         pr_err("Slab cache with size %d has lost its name\n",
  52                                s->object_size);
  53                         continue;
  54                 }
  55
  56                 if (!strcmp(s->name, name)) {
  57                         pr_err("%s (%s): Cache name already exists.\n",
  58                                __func__, name);
  59                         dump_stack();
  60                         s = NULL;
  61                         return -EINVAL;
  62                 }
  63         }
  64
  65         WARN_ON(strchr(name, ' '));     /* It confuses parsers */
  66         return 0;
  67 }
  68 #else
  69 static inline int kmem_cache_sanity_check(const char *name, size_t size)
  70 {
  71         return 0;
  72 }
  73 #endif
  74
  75 /*
  76  * Figure out what the alignment of the objects will be given a set of
  77  * flags, a user specified alignment and the size of the objects.
  78  */
  79 unsigned long calculate_alignment(unsigned long flags,
  80                 unsigned long align, unsigned long size)
  81 {
  82         /*
  83          * If the user wants hardware cache aligned objects then follow that
  84          * suggestion if the object is sufficiently large.
  85          *
  86          * The hardware cache alignment cannot override the specified
  87          * alignment though. If that is greater then use it.
  88          */
  89         if (flags & SLAB_HWCACHE_ALIGN) {
  90                 unsigned long ralign = cache_line_size();
  91                 while (size <= ralign / 2)
  92                         ralign /= 2;
  93                 align = max(align, ralign);
  94         }
  95
  96         if (align < ARCH_SLAB_MINALIGN)
  97                 align = ARCH_SLAB_MINALIGN;
  98
  99         return ALIGN(align, sizeof(void *));
 100 }
 101
 102
 103 /*
 104  * kmem_cache_create - Create a cache.
 105  * @name: A string which is used in /proc/slabinfo to identify this cache.
 106  * @size: The size of objects to be created in this cache.
 107  * @align: The required alignment for the objects.
 108  * @flags: SLAB flags
 109  * @ctor: A constructor for the objects.
 110  *
 111  * Returns a ptr to the cache on success, NULL on failure.
 112  * Cannot be called within a interrupt, but can be interrupted.
 113  * The @ctor is run when new pages are allocated by the cache.
 114  *
 115  * The flags are
 116  *
 117  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
 118  * to catch references to uninitialised memory.
 119  *
 120  * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
 121  * for buffer overruns.
 122  *
 123  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
 124  * cacheline.  This can be beneficial if you're counting cycles as closely
 125  * as davem.
 126  */
 127
 128 struct kmem_cache *kmem_cache_create(const char *name, size_t size, size_t align,
 129                 unsigned long flags, void (*ctor)(void *))
 130 {
 131         struct kmem_cache *s = NULL;
 132         int err = 0;
 133
 134         get_online_cpus();
 135         mutex_lock(&slab_mutex);
 136
 137         if (!kmem_cache_sanity_check(name, size) == 0)
 138                 goto out_locked;
 139
 140         /*
 141          * Some allocators will constraint the set of valid flags to a subset
 142          * of all flags. We expect them to define CACHE_CREATE_MASK in this
 143          * case, and we'll just provide them with a sanitized version of the
 144          * passed flags.
 145          */
 146         flags &= CACHE_CREATE_MASK;
 147
 148         s = __kmem_cache_alias(name, size, align, flags, ctor);
 149         if (s)
 150                 goto out_locked;
 151
 152         s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
 153         if (s) {
 154                 s->object_size = s->size = size;
 155                 s->align = calculate_alignment(flags, align, size);
 156                 s->ctor = ctor;
 157                 s->name = kstrdup(name, GFP_KERNEL);
 158                 if (!s->name) {
 159                         kmem_cache_free(kmem_cache, s);
 160                         err = -ENOMEM;
 161                         goto out_locked;
 162                 }
 163
 164                 err = __kmem_cache_create(s, flags);
 165                 if (!err) {
 166
 167                         s->refcount = 1;
 168                         list_add(&s->list, &slab_caches);
 169
 170                 } else {
 171                         kfree(s->name);
 172                         kmem_cache_free(kmem_cache, s);
 173                 }
 174         } else
 175                 err = -ENOMEM;
 176
 177 out_locked:
 178         mutex_unlock(&slab_mutex);
 179         put_online_cpus();
 180
 181         if (err) {
 182
 183                 if (flags & SLAB_PANIC)
 184                         panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
 185                                 name, err);
 186                 else {
 187                         printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
 188                                 name, err);
 189                         dump_stack();
 190                 }
 191
 192                 return NULL;
 193         }
 194
 195         return s;
 196 }
 197 EXPORT_SYMBOL(kmem_cache_create);
 198
 199 void kmem_cache_destroy(struct kmem_cache *s)
 200 {
 201         get_online_cpus();
 202         mutex_lock(&slab_mutex);
 203         s->refcount--;
 204         if (!s->refcount) {
 205                 list_del(&s->list);
 206
 207                 if (!__kmem_cache_shutdown(s)) {
 208                         mutex_unlock(&slab_mutex);
 209                         if (s->flags & SLAB_DESTROY_BY_RCU)
 210                                 rcu_barrier();
 211
 212                         kfree(s->name);
 213                         kmem_cache_free(kmem_cache, s);
 214                 } else {
 215                         list_add(&s->list, &slab_caches);
 216                         mutex_unlock(&slab_mutex);
 217                         printk(KERN_ERR "kmem_cache_destroy %s: Slab cache still has objects\n",
 218                                 s->name);
 219                         dump_stack();
 220                 }
 221         } else {
 222                 mutex_unlock(&slab_mutex);
 223         }
 224         put_online_cpus();
 225 }
 226 EXPORT_SYMBOL(kmem_cache_destroy);
 227
 228 int slab_is_available(void)
 229 {
 230         return slab_state >= UP;
 231 }
 232
 233 #ifndef CONFIG_SLOB
 234 /* Create a cache during boot when no slab services are available yet */
 235 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
 236                 unsigned long flags)
 237 {
 238         int err;
 239
 240         s->name = name;
 241         s->size = s->object_size = size;
 242         s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
 243         err = __kmem_cache_create(s, flags);
 244
 245         if (err)
 246                 panic("Creation of kmalloc slab %s size=%zd failed. Reason %d\n",
 247                                         name, size, err);
 248
 249         s->refcount = -1;       /* Exempt from merging for now */
 250 }
 251
 252 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
 253                                 unsigned long flags)
 254 {
 255         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
 256
 257         if (!s)
 258                 panic("Out of memory when creating slab %s\n", name);
 259
 260         create_boot_cache(s, name, size, flags);
 261         list_add(&s->list, &slab_caches);
 262         s->refcount = 1;
 263         return s;
 264 }
 265
 266 #endif /* !CONFIG_SLOB */
 267
 268
 269 #ifdef CONFIG_SLABINFO
 270 static void print_slabinfo_header(struct seq_file *m)
 271 {
 272         /*
 273          * Output format version, so at least we can change it
 274          * without _too_ many complaints.
 275          */
 276 #ifdef CONFIG_DEBUG_SLAB
 277         seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
 278 #else
 279         seq_puts(m, "slabinfo - version: 2.1\n");
 280 #endif
 281         seq_puts(m, "# name            <active_objs> <num_objs> <objsize> "
 282                  "<objperslab> <pagesperslab>");
 283         seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
 284         seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
 285 #ifdef CONFIG_DEBUG_SLAB
 286         seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
 287                  "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
 288         seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
 289 #endif
 290         seq_putc(m, '\n');
 291 }
 292
 293 static void *s_start(struct seq_file *m, loff_t *pos)
 294 {
 295         loff_t n = *pos;
 296
 297         mutex_lock(&slab_mutex);
 298         if (!n)
 299                 print_slabinfo_header(m);
 300
 301         return seq_list_start(&slab_caches, *pos);
 302 }
 303
 304 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
 305 {
 306         return seq_list_next(p, &slab_caches, pos);
 307 }
 308
 309 static void s_stop(struct seq_file *m, void *p)
 310 {
 311         mutex_unlock(&slab_mutex);
 312 }
 313
 314 static int s_show(struct seq_file *m, void *p)
 315 {
 316         struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
 317         struct slabinfo sinfo;
 318
 319         memset(&sinfo, 0, sizeof(sinfo));
 320         get_slabinfo(s, &sinfo);
 321
 322         seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
 323                    s->name, sinfo.active_objs, sinfo.num_objs, s->size,
 324                    sinfo.objects_per_slab, (1 << sinfo.cache_order));
 325
 326         seq_printf(m, " : tunables %4u %4u %4u",
 327                    sinfo.limit, sinfo.batchcount, sinfo.shared);
 328         seq_printf(m, " : slabdata %6lu %6lu %6lu",
 329                    sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
 330         slabinfo_show_stats(m, s);
 331         seq_putc(m, '\n');
 332         return 0;
 333 }
 334
 335 /*
 336  * slabinfo_op - iterator that generates /proc/slabinfo
 337  *
 338  * Output layout:
 339  * cache-name
 340  * num-active-objs
 341  * total-objs
 342  * object size
 343  * num-active-slabs
 344  * total-slabs
 345  * num-pages-per-slab
 346  * + further values on SMP and with statistics enabled
 347  */
 348 static const struct seq_operations slabinfo_op = {
 349         .start = s_start,
 350         .next = s_next,
 351         .stop = s_stop,
 352         .show = s_show,
 353 };
 354
 355 static int slabinfo_open(struct inode *inode, struct file *file)
 356 {
 357         return seq_open(file, &slabinfo_op);
 358 }
 359
 360 static const struct file_operations proc_slabinfo_operations = {
 361         .open           = slabinfo_open,
 362         .read           = seq_read,
 363         .write          = slabinfo_write,
 364         .llseek         = seq_lseek,
 365         .release        = seq_release,
 366 };
 367
 368 static int __init slab_proc_init(void)
 369 {
 370         proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations);
 371         return 0;
 372 }
 373 module_init(slab_proc_init);
 374 #endif /* CONFIG_SLABINFO */