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/module.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, node_id);
  66         if (!pool)
  67                 return NULL;
  68         memset(pool, 0, sizeof(*pool));
  69         pool->elements = kmalloc_node(min_nr * sizeof(void *),
  70                                         GFP_KERNEL, node_id);
  71         if (!pool->elements) {
  72                 kfree(pool);
  73                 return NULL;
  74         }
  75         spin_lock_init(&pool->lock);
  76         pool->min_nr = min_nr;
  77         pool->pool_data = pool_data;
  78         init_waitqueue_head(&pool->wait);
  79         pool->alloc = alloc_fn;
  80         pool->free = free_fn;
  81
  82         /*
  83          * First pre-allocate the guaranteed number of buffers.
  84          */
  85         while (pool->curr_nr < pool->min_nr) {
  86                 void *element;
  87
  88                 element = pool->alloc(GFP_KERNEL, pool->pool_data);
  89                 if (unlikely(!element)) {
  90                         free_pool(pool);
  91                         return NULL;
  92                 }
  93                 add_element(pool, element);
  94         }
  95         return pool;
  96 }
  97 EXPORT_SYMBOL(mempool_create_node);
  98
  99 /**
 100  * mempool_resize - resize an existing memory pool
 101  * @pool:       pointer to the memory pool which was allocated via
 102  *              mempool_create().
 103  * @new_min_nr: the new minimum number of elements guaranteed to be
 104  *              allocated for this pool.
 105  * @gfp_mask:   the usual allocation bitmask.
 106  *
 107  * This function shrinks/grows the pool. In the case of growing,
 108  * it cannot be guaranteed that the pool will be grown to the new
 109  * size immediately, but new mempool_free() calls will refill it.
 110  *
 111  * Note, the caller must guarantee that no mempool_destroy is called
 112  * while this function is running. mempool_alloc() & mempool_free()
 113  * might be called (eg. from IRQ contexts) while this function executes.
 114  */
 115 int mempool_resize(mempool_t *pool, int new_min_nr, gfp_t gfp_mask)
 116 {
 117         void *element;
 118         void **new_elements;
 119         unsigned long flags;
 120
 121         BUG_ON(new_min_nr <= 0);
 122
 123         spin_lock_irqsave(&pool->lock, flags);
 124         if (new_min_nr <= pool->min_nr) {
 125                 while (new_min_nr < pool->curr_nr) {
 126                         element = remove_element(pool);
 127                         spin_unlock_irqrestore(&pool->lock, flags);
 128                         pool->free(element, pool->pool_data);
 129                         spin_lock_irqsave(&pool->lock, flags);
 130                 }
 131                 pool->min_nr = new_min_nr;
 132                 goto out_unlock;
 133         }
 134         spin_unlock_irqrestore(&pool->lock, flags);
 135
 136         /* Grow the pool */
 137         new_elements = kmalloc(new_min_nr * sizeof(*new_elements), gfp_mask);
 138         if (!new_elements)
 139                 return -ENOMEM;
 140
 141         spin_lock_irqsave(&pool->lock, flags);
 142         if (unlikely(new_min_nr <= pool->min_nr)) {
 143                 /* Raced, other resize will do our work */
 144                 spin_unlock_irqrestore(&pool->lock, flags);
 145                 kfree(new_elements);
 146                 goto out;
 147         }
 148         memcpy(new_elements, pool->elements,
 149                         pool->curr_nr * sizeof(*new_elements));
 150         kfree(pool->elements);
 151         pool->elements = new_elements;
 152         pool->min_nr = new_min_nr;
 153
 154         while (pool->curr_nr < pool->min_nr) {
 155                 spin_unlock_irqrestore(&pool->lock, flags);
 156                 element = pool->alloc(gfp_mask, pool->pool_data);
 157                 if (!element)
 158                         goto out;
 159                 spin_lock_irqsave(&pool->lock, flags);
 160                 if (pool->curr_nr < pool->min_nr) {
 161                         add_element(pool, element);
 162                 } else {
 163                         spin_unlock_irqrestore(&pool->lock, flags);
 164                         pool->free(element, pool->pool_data);   /* Raced */
 165                         goto out;
 166                 }
 167         }
 168 out_unlock:
 169         spin_unlock_irqrestore(&pool->lock, flags);
 170 out:
 171         return 0;
 172 }
 173 EXPORT_SYMBOL(mempool_resize);
 174
 175 /**
 176  * mempool_destroy - deallocate a memory pool
 177  * @pool:      pointer to the memory pool which was allocated via
 178  *             mempool_create().
 179  *
 180  * this function only sleeps if the free_fn() function sleeps. The caller
 181  * has to guarantee that all elements have been returned to the pool (ie:
 182  * freed) prior to calling mempool_destroy().
 183  */
 184 void mempool_destroy(mempool_t *pool)
 185 {
 186         /* Check for outstanding elements */
 187         BUG_ON(pool->curr_nr != pool->min_nr);
 188         free_pool(pool);
 189 }
 190 EXPORT_SYMBOL(mempool_destroy);
 191
 192 /**
 193  * mempool_alloc - allocate an element from a specific memory pool
 194  * @pool:      pointer to the memory pool which was allocated via
 195  *             mempool_create().
 196  * @gfp_mask:  the usual allocation bitmask.
 197  *
 198  * this function only sleeps if the alloc_fn function sleeps or
 199  * returns NULL. Note that due to preallocation, this function
 200  * *never* fails when called from process contexts. (it might
 201  * fail if called from an IRQ context.)
 202  */
 203 void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
 204 {
 205         void *element;
 206         unsigned long flags;
 207         wait_queue_t wait;
 208         gfp_t gfp_temp;
 209
 210         might_sleep_if(gfp_mask & __GFP_WAIT);
 211
 212         gfp_mask |= __GFP_NOMEMALLOC;   /* don't allocate emergency reserves */
 213         gfp_mask |= __GFP_NORETRY;      /* don't loop in __alloc_pages */
 214         gfp_mask |= __GFP_NOWARN;       /* failures are OK */
 215
 216         gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO);
 217
 218 repeat_alloc:
 219
 220         element = pool->alloc(gfp_temp, pool->pool_data);
 221         if (likely(element != NULL))
 222                 return element;
 223
 224         spin_lock_irqsave(&pool->lock, flags);
 225         if (likely(pool->curr_nr)) {
 226                 element = remove_element(pool);
 227                 spin_unlock_irqrestore(&pool->lock, flags);
 228                 return element;
 229         }
 230         spin_unlock_irqrestore(&pool->lock, flags);
 231
 232         /* We must not sleep in the GFP_ATOMIC case */
 233         if (!(gfp_mask & __GFP_WAIT))
 234                 return NULL;
 235
 236         /* Now start performing page reclaim */
 237         gfp_temp = gfp_mask;
 238         init_wait(&wait);
 239         prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
 240         smp_mb();
 241         if (!pool->curr_nr) {
 242                 /*
 243                  * FIXME: this should be io_schedule().  The timeout is there
 244                  * as a workaround for some DM problems in 2.6.18.
 245                  */
 246                 io_schedule_timeout(5*HZ);
 247         }
 248         finish_wait(&pool->wait, &wait);
 249
 250         goto repeat_alloc;
 251 }
 252 EXPORT_SYMBOL(mempool_alloc);
 253
 254 /**
 255  * mempool_free - return an element to the pool.
 256  * @element:   pool element pointer.
 257  * @pool:      pointer to the memory pool which was allocated via
 258  *             mempool_create().
 259  *
 260  * this function only sleeps if the free_fn() function sleeps.
 261  */
 262 void mempool_free(void *element, mempool_t *pool)
 263 {
 264         unsigned long flags;
 265
 266         smp_mb();
 267         if (pool->curr_nr < pool->min_nr) {
 268                 spin_lock_irqsave(&pool->lock, flags);
 269                 if (pool->curr_nr < pool->min_nr) {
 270                         add_element(pool, element);
 271                         spin_unlock_irqrestore(&pool->lock, flags);
 272                         wake_up(&pool->wait);
 273                         return;
 274                 }
 275                 spin_unlock_irqrestore(&pool->lock, flags);
 276         }
 277         pool->free(element, pool->pool_data);
 278 }
 279 EXPORT_SYMBOL(mempool_free);
 280
 281 /*
 282  * A commonly used alloc and free fn.
 283  */
 284 void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
 285 {
 286         struct kmem_cache *mem = pool_data;
 287         return kmem_cache_alloc(mem, gfp_mask);
 288 }
 289 EXPORT_SYMBOL(mempool_alloc_slab);
 290
 291 void mempool_free_slab(void *element, void *pool_data)
 292 {
 293         struct kmem_cache *mem = pool_data;
 294         kmem_cache_free(mem, element);
 295 }
 296 EXPORT_SYMBOL(mempool_free_slab);
 297
 298 /*
 299  * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
 300  * specfied by pool_data
 301  */
 302 void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
 303 {
 304         size_t size = (size_t)(long)pool_data;
 305         return kmalloc(size, gfp_mask);
 306 }
 307 EXPORT_SYMBOL(mempool_kmalloc);
 308
 309 void *mempool_kzalloc(gfp_t gfp_mask, void *pool_data)
 310 {
 311         size_t size = (size_t) pool_data;
 312         return kzalloc(size, gfp_mask);
 313 }
 314 EXPORT_SYMBOL(mempool_kzalloc);
 315
 316 void mempool_kfree(void *element, void *pool_data)
 317 {
 318         kfree(element);
 319 }
 320 EXPORT_SYMBOL(mempool_kfree);
 321
 322 /*
 323  * A simple mempool-backed page allocator that allocates pages
 324  * of the order specified by pool_data.
 325  */
 326 void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
 327 {
 328         int order = (int)(long)pool_data;
 329         return alloc_pages(gfp_mask, order);
 330 }
 331 EXPORT_SYMBOL(mempool_alloc_pages);
 332
 333 void mempool_free_pages(void *element, void *pool_data)
 334 {
 335         int order = (int)(long)pool_data;
 336         __free_pages(element, order);
 337 }
 338 EXPORT_SYMBOL(mempool_free_pages);