mm/mempool.c

   1 /*
   2  *  linux/mm/mempool.c
   3  *
   4  *  memory buffer pool support. Such pools are mostly used
   5  *  for guaranteed, deadlock-free memory allocations during
   6  *  extreme VM load.
   7  *
   8  *  started by Ingo Molnar, Copyright (C) 2001
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/slab.h>
  13 #include <linux/export.h>
  14 #include <linux/mempool.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/writeback.h>
  17
  18 static void add_element(mempool_t *pool, void *element)
  19 {
  20         BUG_ON(pool->curr_nr >= pool->min_nr);
  21         pool->elements[pool->curr_nr++] = element;
  22 }
  23
  24 static void *remove_element(mempool_t *pool)
  25 {
  26         BUG_ON(pool->curr_nr <= 0);
  27         return pool->elements[--pool->curr_nr];
  28 }
  29
  30 static void free_pool(mempool_t *pool)
  31 {
  32         while (pool->curr_nr) {
  33                 void *element = remove_element(pool);
  34                 pool->free(element, pool->pool_data);
  35         }
  36         kfree(pool->elements);
  37         kfree(pool);
  38 }
  39
  40 /**
  41  * mempool_create - create a memory pool
  42  * @min_nr:    the minimum number of elements guaranteed to be
  43  *             allocated for this pool.
  44  * @alloc_fn:  user-defined element-allocation function.
  45  * @free_fn:   user-defined element-freeing function.
  46  * @pool_data: optional private data available to the user-defined functions.
  47  *
  48  * this function creates and allocates a guaranteed size, preallocated
  49  * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
  50  * functions. This function might sleep. Both the alloc_fn() and the free_fn()
  51  * functions might sleep - as long as the mempool_alloc() function is not called
  52  * from IRQ contexts.
  53  */
  54 mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn,
  55                                 mempool_free_t *free_fn, void *pool_data)
  56 {
  57         return  mempool_create_node(min_nr,alloc_fn,free_fn, pool_data,-1);
  58 }
  59 EXPORT_SYMBOL(mempool_create);
  60
  61 mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
  62                         mempool_free_t *free_fn, void *pool_data, int node_id)
  63 {
  64         mempool_t *pool;
  65         pool = kmalloc_node(sizeof(*pool), GFP_KERNEL | __GFP_ZERO, node_id);
  66         if (!pool)
  67                 return NULL;
  68         pool->elements = kmalloc_node(min_nr * sizeof(void *),
  69                                         GFP_KERNEL, node_id);
  70         if (!pool->elements) {
  71                 kfree(pool);
  72                 return NULL;
  73         }
  74         spin_lock_init(&pool->lock);
  75         pool->min_nr = min_nr;
  76         pool->pool_data = pool_data;
  77         init_waitqueue_head(&pool->wait);
  78         pool->alloc = alloc_fn;
  79         pool->free = free_fn;
  80
  81         /*
  82          * First pre-allocate the guaranteed number of buffers.
  83          */
  84         while (pool->curr_nr < pool->min_nr) {
  85                 void *element;
  86
  87                 element = pool->alloc(GFP_KERNEL, pool->pool_data);
  88                 if (unlikely(!element)) {
  89                         free_pool(pool);
  90                         return NULL;
  91                 }
  92                 add_element(pool, element);
  93         }
  94         return pool;
  95 }
  96 EXPORT_SYMBOL(mempool_create_node);
  97
  98 /**
  99  * mempool_resize - resize an existing memory pool
 100  * @pool:       pointer to the memory pool which was allocated via
 101  *              mempool_create().
 102  * @new_min_nr: the new minimum number of elements guaranteed to be
 103  *              allocated for this pool.
 104  * @gfp_mask:   the usual allocation bitmask.
 105  *
 106  * This function shrinks/grows the pool. In the case of growing,
 107  * it cannot be guaranteed that the pool will be grown to the new
 108  * size immediately, but new mempool_free() calls will refill it.
 109  *
 110  * Note, the caller must guarantee that no mempool_destroy is called
 111  * while this function is running. mempool_alloc() & mempool_free()
 112  * might be called (eg. from IRQ contexts) while this function executes.
 113  */
 114 int mempool_resize(mempool_t *pool, int new_min_nr, gfp_t gfp_mask)
 115 {
 116         void *element;
 117         void **new_elements;
 118         unsigned long flags;
 119
 120         BUG_ON(new_min_nr <= 0);
 121
 122         spin_lock_irqsave(&pool->lock, flags);
 123         if (new_min_nr <= pool->min_nr) {
 124                 while (new_min_nr < pool->curr_nr) {
 125                         element = remove_element(pool);
 126                         spin_unlock_irqrestore(&pool->lock, flags);
 127                         pool->free(element, pool->pool_data);
 128                         spin_lock_irqsave(&pool->lock, flags);
 129                 }
 130                 pool->min_nr = new_min_nr;
 131                 goto out_unlock;
 132         }
 133         spin_unlock_irqrestore(&pool->lock, flags);
 134
 135         /* Grow the pool */
 136         new_elements = kmalloc(new_min_nr * sizeof(*new_elements), gfp_mask);
 137         if (!new_elements)
 138                 return -ENOMEM;
 139
 140         spin_lock_irqsave(&pool->lock, flags);
 141         if (unlikely(new_min_nr <= pool->min_nr)) {
 142                 /* Raced, other resize will do our work */
 143                 spin_unlock_irqrestore(&pool->lock, flags);
 144                 kfree(new_elements);
 145                 goto out;
 146         }
 147         memcpy(new_elements, pool->elements,
 148                         pool->curr_nr * sizeof(*new_elements));
 149         kfree(pool->elements);
 150         pool->elements = new_elements;
 151         pool->min_nr = new_min_nr;
 152
 153         while (pool->curr_nr < pool->min_nr) {
 154                 spin_unlock_irqrestore(&pool->lock, flags);
 155                 element = pool->alloc(gfp_mask, pool->pool_data);
 156                 if (!element)
 157                         goto out;
 158                 spin_lock_irqsave(&pool->lock, flags);
 159                 if (pool->curr_nr < pool->min_nr) {
 160                         add_element(pool, element);
 161                 } else {
 162                         spin_unlock_irqrestore(&pool->lock, flags);
 163                         pool->free(element, pool->pool_data);   /* Raced */
 164                         goto out;
 165                 }
 166         }
 167 out_unlock:
 168         spin_unlock_irqrestore(&pool->lock, flags);
 169 out:
 170         return 0;
 171 }
 172 EXPORT_SYMBOL(mempool_resize);
 173
 174 /**
 175  * mempool_destroy - deallocate a memory pool
 176  * @pool:      pointer to the memory pool which was allocated via
 177  *             mempool_create().
 178  *
 179  * this function only sleeps if the free_fn() function sleeps. The caller
 180  * has to guarantee that all elements have been returned to the pool (ie:
 181  * freed) prior to calling mempool_destroy().
 182  */
 183 void mempool_destroy(mempool_t *pool)
 184 {
 185         /* Check for outstanding elements */
 186         BUG_ON(pool->curr_nr != pool->min_nr);
 187         free_pool(pool);
 188 }
 189 EXPORT_SYMBOL(mempool_destroy);
 190
 191 /**
 192  * mempool_alloc - allocate an element from a specific memory pool
 193  * @pool:      pointer to the memory pool which was allocated via
 194  *             mempool_create().
 195  * @gfp_mask:  the usual allocation bitmask.
 196  *
 197  * this function only sleeps if the alloc_fn() function sleeps or
 198  * returns NULL. Note that due to preallocation, this function
 199  * *never* fails when called from process contexts. (it might
 200  * fail if called from an IRQ context.)
 201  */
 202 void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
 203 {
 204         void *element;
 205         unsigned long flags;
 206         wait_queue_t wait;
 207         gfp_t gfp_temp;
 208
 209         might_sleep_if(gfp_mask & __GFP_WAIT);
 210
 211         gfp_mask |= __GFP_NOMEMALLOC;   /* don't allocate emergency reserves */
 212         gfp_mask |= __GFP_NORETRY;      /* don't loop in __alloc_pages */
 213         gfp_mask |= __GFP_NOWARN;       /* failures are OK */
 214
 215         gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO);
 216
 217 repeat_alloc:
 218
 219         element = pool->alloc(gfp_temp, pool->pool_data);
 220         if (likely(element != NULL))
 221                 return element;
 222
 223         spin_lock_irqsave(&pool->lock, flags);
 224         if (likely(pool->curr_nr)) {
 225                 element = remove_element(pool);
 226                 spin_unlock_irqrestore(&pool->lock, flags);
 227                 return element;
 228         }
 229         spin_unlock_irqrestore(&pool->lock, flags);
 230
 231         /* We must not sleep in the GFP_ATOMIC case */
 232         if (!(gfp_mask & __GFP_WAIT))
 233                 return NULL;
 234
 235         /* Now start performing page reclaim */
 236         gfp_temp = gfp_mask;
 237         init_wait(&wait);
 238         prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
 239         smp_mb();
 240         if (!pool->curr_nr) {
 241                 /*
 242                  * FIXME: this should be io_schedule().  The timeout is there
 243                  * as a workaround for some DM problems in 2.6.18.
 244                  */
 245                 io_schedule_timeout(5*HZ);
 246         }
 247         finish_wait(&pool->wait, &wait);
 248
 249         goto repeat_alloc;
 250 }
 251 EXPORT_SYMBOL(mempool_alloc);
 252
 253 /**
 254  * mempool_free - return an element to the pool.
 255  * @element:   pool element pointer.
 256  * @pool:      pointer to the memory pool which was allocated via
 257  *             mempool_create().
 258  *
 259  * this function only sleeps if the free_fn() function sleeps.
 260  */
 261 void mempool_free(void *element, mempool_t *pool)
 262 {
 263         unsigned long flags;
 264
 265         if (unlikely(element == NULL))
 266                 return;
 267
 268         smp_mb();
 269         if (pool->curr_nr < pool->min_nr) {
 270                 spin_lock_irqsave(&pool->lock, flags);
 271                 if (pool->curr_nr < pool->min_nr) {
 272                         add_element(pool, element);
 273                         spin_unlock_irqrestore(&pool->lock, flags);
 274                         wake_up(&pool->wait);
 275                         return;
 276                 }
 277                 spin_unlock_irqrestore(&pool->lock, flags);
 278         }
 279         pool->free(element, pool->pool_data);
 280 }
 281 EXPORT_SYMBOL(mempool_free);
 282
 283 /*
 284  * A commonly used alloc and free fn.
 285  */
 286 void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
 287 {
 288         struct kmem_cache *mem = pool_data;
 289         return kmem_cache_alloc(mem, gfp_mask);
 290 }
 291 EXPORT_SYMBOL(mempool_alloc_slab);
 292
 293 void mempool_free_slab(void *element, void *pool_data)
 294 {
 295         struct kmem_cache *mem = pool_data;
 296         kmem_cache_free(mem, element);
 297 }
 298 EXPORT_SYMBOL(mempool_free_slab);
 299
 300 /*
 301  * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
 302  * specified by pool_data
 303  */
 304 void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
 305 {
 306         size_t size = (size_t)pool_data;
 307         return kmalloc(size, gfp_mask);
 308 }
 309 EXPORT_SYMBOL(mempool_kmalloc);
 310
 311 void mempool_kfree(void *element, void *pool_data)
 312 {
 313         kfree(element);
 314 }
 315 EXPORT_SYMBOL(mempool_kfree);
 316
 317 /*
 318  * A simple mempool-backed page allocator that allocates pages
 319  * of the order specified by pool_data.
 320  */
 321 void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
 322 {
 323         int order = (int)(long)pool_data;
 324         return alloc_pages(gfp_mask, order);
 325 }
 326 EXPORT_SYMBOL(mempool_alloc_pages);
 327
 328 void mempool_free_pages(void *element, void *pool_data)
 329 {
 330         int order = (int)(long)pool_data;
 331         __free_pages(element, order);
 332 }
 333 EXPORT_SYMBOL(mempool_free_pages);