Merge remote-tracking branch 'upstream/master' into kvm-devel
[linux-2.6/kvm.git] / lib / genalloc.c
blobf352cc42f4f8a58b8e4c8304de32cb31b4e29b67
1 /*
2 * Basic general purpose allocator for managing special purpose
3 * memory, for example, memory that is not managed by the regular
4 * kmalloc/kfree interface. Uses for this includes on-device special
5 * memory, uncached memory etc.
7 * It is safe to use the allocator in NMI handlers and other special
8 * unblockable contexts that could otherwise deadlock on locks. This
9 * is implemented by using atomic operations and retries on any
10 * conflicts. The disadvantage is that there may be livelocks in
11 * extreme cases. For better scalability, one allocator can be used
12 * for each CPU.
14 * The lockless operation only works if there is enough memory
15 * available. If new memory is added to the pool a lock has to be
16 * still taken. So any user relying on locklessness has to ensure
17 * that sufficient memory is preallocated.
19 * The basic atomic operation of this allocator is cmpxchg on long.
20 * On architectures that don't have NMI-safe cmpxchg implementation,
21 * the allocator can NOT be used in NMI handler. So code uses the
22 * allocator in NMI handler should depend on
23 * CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG.
25 * Copyright 2005 (C) Jes Sorensen <jes@trained-monkey.org>
27 * This source code is licensed under the GNU General Public License,
28 * Version 2. See the file COPYING for more details.
31 #include <linux/slab.h>
32 #include <linux/module.h>
33 #include <linux/bitmap.h>
34 #include <linux/rculist.h>
35 #include <linux/interrupt.h>
36 #include <linux/genalloc.h>
38 static int set_bits_ll(unsigned long *addr, unsigned long mask_to_set)
40 unsigned long val, nval;
42 nval = *addr;
43 do {
44 val = nval;
45 if (val & mask_to_set)
46 return -EBUSY;
47 cpu_relax();
48 } while ((nval = cmpxchg(addr, val, val | mask_to_set)) != val);
50 return 0;
53 static int clear_bits_ll(unsigned long *addr, unsigned long mask_to_clear)
55 unsigned long val, nval;
57 nval = *addr;
58 do {
59 val = nval;
60 if ((val & mask_to_clear) != mask_to_clear)
61 return -EBUSY;
62 cpu_relax();
63 } while ((nval = cmpxchg(addr, val, val & ~mask_to_clear)) != val);
65 return 0;
69 * bitmap_set_ll - set the specified number of bits at the specified position
70 * @map: pointer to a bitmap
71 * @start: a bit position in @map
72 * @nr: number of bits to set
74 * Set @nr bits start from @start in @map lock-lessly. Several users
75 * can set/clear the same bitmap simultaneously without lock. If two
76 * users set the same bit, one user will return remain bits, otherwise
77 * return 0.
79 static int bitmap_set_ll(unsigned long *map, int start, int nr)
81 unsigned long *p = map + BIT_WORD(start);
82 const int size = start + nr;
83 int bits_to_set = BITS_PER_LONG - (start % BITS_PER_LONG);
84 unsigned long mask_to_set = BITMAP_FIRST_WORD_MASK(start);
86 while (nr - bits_to_set >= 0) {
87 if (set_bits_ll(p, mask_to_set))
88 return nr;
89 nr -= bits_to_set;
90 bits_to_set = BITS_PER_LONG;
91 mask_to_set = ~0UL;
92 p++;
94 if (nr) {
95 mask_to_set &= BITMAP_LAST_WORD_MASK(size);
96 if (set_bits_ll(p, mask_to_set))
97 return nr;
100 return 0;
104 * bitmap_clear_ll - clear the specified number of bits at the specified position
105 * @map: pointer to a bitmap
106 * @start: a bit position in @map
107 * @nr: number of bits to set
109 * Clear @nr bits start from @start in @map lock-lessly. Several users
110 * can set/clear the same bitmap simultaneously without lock. If two
111 * users clear the same bit, one user will return remain bits,
112 * otherwise return 0.
114 static int bitmap_clear_ll(unsigned long *map, int start, int nr)
116 unsigned long *p = map + BIT_WORD(start);
117 const int size = start + nr;
118 int bits_to_clear = BITS_PER_LONG - (start % BITS_PER_LONG);
119 unsigned long mask_to_clear = BITMAP_FIRST_WORD_MASK(start);
121 while (nr - bits_to_clear >= 0) {
122 if (clear_bits_ll(p, mask_to_clear))
123 return nr;
124 nr -= bits_to_clear;
125 bits_to_clear = BITS_PER_LONG;
126 mask_to_clear = ~0UL;
127 p++;
129 if (nr) {
130 mask_to_clear &= BITMAP_LAST_WORD_MASK(size);
131 if (clear_bits_ll(p, mask_to_clear))
132 return nr;
135 return 0;
139 * gen_pool_create - create a new special memory pool
140 * @min_alloc_order: log base 2 of number of bytes each bitmap bit represents
141 * @nid: node id of the node the pool structure should be allocated on, or -1
143 * Create a new special memory pool that can be used to manage special purpose
144 * memory not managed by the regular kmalloc/kfree interface.
146 struct gen_pool *gen_pool_create(int min_alloc_order, int nid)
148 struct gen_pool *pool;
150 pool = kmalloc_node(sizeof(struct gen_pool), GFP_KERNEL, nid);
151 if (pool != NULL) {
152 spin_lock_init(&pool->lock);
153 INIT_LIST_HEAD(&pool->chunks);
154 pool->min_alloc_order = min_alloc_order;
156 return pool;
158 EXPORT_SYMBOL(gen_pool_create);
161 * gen_pool_add_virt - add a new chunk of special memory to the pool
162 * @pool: pool to add new memory chunk to
163 * @virt: virtual starting address of memory chunk to add to pool
164 * @phys: physical starting address of memory chunk to add to pool
165 * @size: size in bytes of the memory chunk to add to pool
166 * @nid: node id of the node the chunk structure and bitmap should be
167 * allocated on, or -1
169 * Add a new chunk of special memory to the specified pool.
171 * Returns 0 on success or a -ve errno on failure.
173 int gen_pool_add_virt(struct gen_pool *pool, unsigned long virt, phys_addr_t phys,
174 size_t size, int nid)
176 struct gen_pool_chunk *chunk;
177 int nbits = size >> pool->min_alloc_order;
178 int nbytes = sizeof(struct gen_pool_chunk) +
179 (nbits + BITS_PER_BYTE - 1) / BITS_PER_BYTE;
181 chunk = kmalloc_node(nbytes, GFP_KERNEL | __GFP_ZERO, nid);
182 if (unlikely(chunk == NULL))
183 return -ENOMEM;
185 chunk->phys_addr = phys;
186 chunk->start_addr = virt;
187 chunk->end_addr = virt + size;
188 atomic_set(&chunk->avail, size);
190 spin_lock(&pool->lock);
191 list_add_rcu(&chunk->next_chunk, &pool->chunks);
192 spin_unlock(&pool->lock);
194 return 0;
196 EXPORT_SYMBOL(gen_pool_add_virt);
199 * gen_pool_virt_to_phys - return the physical address of memory
200 * @pool: pool to allocate from
201 * @addr: starting address of memory
203 * Returns the physical address on success, or -1 on error.
205 phys_addr_t gen_pool_virt_to_phys(struct gen_pool *pool, unsigned long addr)
207 struct gen_pool_chunk *chunk;
208 phys_addr_t paddr = -1;
210 rcu_read_lock();
211 list_for_each_entry_rcu(chunk, &pool->chunks, next_chunk) {
212 if (addr >= chunk->start_addr && addr < chunk->end_addr) {
213 paddr = chunk->phys_addr + (addr - chunk->start_addr);
214 break;
217 rcu_read_unlock();
219 return paddr;
221 EXPORT_SYMBOL(gen_pool_virt_to_phys);
224 * gen_pool_destroy - destroy a special memory pool
225 * @pool: pool to destroy
227 * Destroy the specified special memory pool. Verifies that there are no
228 * outstanding allocations.
230 void gen_pool_destroy(struct gen_pool *pool)
232 struct list_head *_chunk, *_next_chunk;
233 struct gen_pool_chunk *chunk;
234 int order = pool->min_alloc_order;
235 int bit, end_bit;
237 list_for_each_safe(_chunk, _next_chunk, &pool->chunks) {
238 chunk = list_entry(_chunk, struct gen_pool_chunk, next_chunk);
239 list_del(&chunk->next_chunk);
241 end_bit = (chunk->end_addr - chunk->start_addr) >> order;
242 bit = find_next_bit(chunk->bits, end_bit, 0);
243 BUG_ON(bit < end_bit);
245 kfree(chunk);
247 kfree(pool);
248 return;
250 EXPORT_SYMBOL(gen_pool_destroy);
253 * gen_pool_alloc - allocate special memory from the pool
254 * @pool: pool to allocate from
255 * @size: number of bytes to allocate from the pool
257 * Allocate the requested number of bytes from the specified pool.
258 * Uses a first-fit algorithm. Can not be used in NMI handler on
259 * architectures without NMI-safe cmpxchg implementation.
261 unsigned long gen_pool_alloc(struct gen_pool *pool, size_t size)
263 struct gen_pool_chunk *chunk;
264 unsigned long addr = 0;
265 int order = pool->min_alloc_order;
266 int nbits, start_bit = 0, end_bit, remain;
268 #ifndef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
269 BUG_ON(in_nmi());
270 #endif
272 if (size == 0)
273 return 0;
275 nbits = (size + (1UL << order) - 1) >> order;
276 rcu_read_lock();
277 list_for_each_entry_rcu(chunk, &pool->chunks, next_chunk) {
278 if (size > atomic_read(&chunk->avail))
279 continue;
281 end_bit = (chunk->end_addr - chunk->start_addr) >> order;
282 retry:
283 start_bit = bitmap_find_next_zero_area(chunk->bits, end_bit,
284 start_bit, nbits, 0);
285 if (start_bit >= end_bit)
286 continue;
287 remain = bitmap_set_ll(chunk->bits, start_bit, nbits);
288 if (remain) {
289 remain = bitmap_clear_ll(chunk->bits, start_bit,
290 nbits - remain);
291 BUG_ON(remain);
292 goto retry;
295 addr = chunk->start_addr + ((unsigned long)start_bit << order);
296 size = nbits << order;
297 atomic_sub(size, &chunk->avail);
298 break;
300 rcu_read_unlock();
301 return addr;
303 EXPORT_SYMBOL(gen_pool_alloc);
306 * gen_pool_free - free allocated special memory back to the pool
307 * @pool: pool to free to
308 * @addr: starting address of memory to free back to pool
309 * @size: size in bytes of memory to free
311 * Free previously allocated special memory back to the specified
312 * pool. Can not be used in NMI handler on architectures without
313 * NMI-safe cmpxchg implementation.
315 void gen_pool_free(struct gen_pool *pool, unsigned long addr, size_t size)
317 struct gen_pool_chunk *chunk;
318 int order = pool->min_alloc_order;
319 int start_bit, nbits, remain;
321 #ifndef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
322 BUG_ON(in_nmi());
323 #endif
325 nbits = (size + (1UL << order) - 1) >> order;
326 rcu_read_lock();
327 list_for_each_entry_rcu(chunk, &pool->chunks, next_chunk) {
328 if (addr >= chunk->start_addr && addr < chunk->end_addr) {
329 BUG_ON(addr + size > chunk->end_addr);
330 start_bit = (addr - chunk->start_addr) >> order;
331 remain = bitmap_clear_ll(chunk->bits, start_bit, nbits);
332 BUG_ON(remain);
333 size = nbits << order;
334 atomic_add(size, &chunk->avail);
335 rcu_read_unlock();
336 return;
339 rcu_read_unlock();
340 BUG();
342 EXPORT_SYMBOL(gen_pool_free);
345 * gen_pool_for_each_chunk - call func for every chunk of generic memory pool
346 * @pool: the generic memory pool
347 * @func: func to call
348 * @data: additional data used by @func
350 * Call @func for every chunk of generic memory pool. The @func is
351 * called with rcu_read_lock held.
353 void gen_pool_for_each_chunk(struct gen_pool *pool,
354 void (*func)(struct gen_pool *pool, struct gen_pool_chunk *chunk, void *data),
355 void *data)
357 struct gen_pool_chunk *chunk;
359 rcu_read_lock();
360 list_for_each_entry_rcu(chunk, &(pool)->chunks, next_chunk)
361 func(pool, chunk, data);
362 rcu_read_unlock();
364 EXPORT_SYMBOL(gen_pool_for_each_chunk);
367 * gen_pool_avail - get available free space of the pool
368 * @pool: pool to get available free space
370 * Return available free space of the specified pool.
372 size_t gen_pool_avail(struct gen_pool *pool)
374 struct gen_pool_chunk *chunk;
375 size_t avail = 0;
377 rcu_read_lock();
378 list_for_each_entry_rcu(chunk, &pool->chunks, next_chunk)
379 avail += atomic_read(&chunk->avail);
380 rcu_read_unlock();
381 return avail;
383 EXPORT_SYMBOL_GPL(gen_pool_avail);
386 * gen_pool_size - get size in bytes of memory managed by the pool
387 * @pool: pool to get size
389 * Return size in bytes of memory managed by the pool.
391 size_t gen_pool_size(struct gen_pool *pool)
393 struct gen_pool_chunk *chunk;
394 size_t size = 0;
396 rcu_read_lock();
397 list_for_each_entry_rcu(chunk, &pool->chunks, next_chunk)
398 size += chunk->end_addr - chunk->start_addr;
399 rcu_read_unlock();
400 return size;
402 EXPORT_SYMBOL_GPL(gen_pool_size);