| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * memory buffer pool support. Such pools are mostly used |
| 4 | * for guaranteed, deadlock-free memory allocations during |
| 5 | * extreme VM load. |
| 6 | * |
| 7 | * started by Ingo Molnar, Copyright (C) 2001 |
| 8 | * debugging by David Rientjes, Copyright (C) 2015 |
| 9 | */ |
| 10 | #include <linux/fault-inject.h> |
| 11 | #include <linux/mm.h> |
| 12 | #include <linux/slab.h> |
| 13 | #include <linux/highmem.h> |
| 14 | #include <linux/kasan.h> |
| 15 | #include <linux/kmemleak.h> |
| 16 | #include <linux/export.h> |
| 17 | #include <linux/mempool.h> |
| 18 | #include <linux/writeback.h> |
| 19 | #include "slab.h" |
| 20 | |
| 21 | static DECLARE_FAULT_ATTR(fail_mempool_alloc); |
| 22 | static DECLARE_FAULT_ATTR(fail_mempool_alloc_bulk); |
| 23 | |
| 24 | static int __init mempool_faul_inject_init(void) |
| 25 | { |
| 26 | int error; |
| 27 | |
| 28 | error = PTR_ERR_OR_ZERO(ptr: fault_create_debugfs_attr(name: "fail_mempool_alloc" , |
| 29 | NULL, attr: &fail_mempool_alloc)); |
| 30 | if (error) |
| 31 | return error; |
| 32 | |
| 33 | /* booting will fail on error return here, don't bother to cleanup */ |
| 34 | return PTR_ERR_OR_ZERO( |
| 35 | ptr: fault_create_debugfs_attr(name: "fail_mempool_alloc_bulk" , NULL, |
| 36 | attr: &fail_mempool_alloc_bulk)); |
| 37 | } |
| 38 | late_initcall(mempool_faul_inject_init); |
| 39 | |
| 40 | #ifdef CONFIG_SLUB_DEBUG_ON |
| 41 | static void poison_error(struct mempool *pool, void *element, size_t size, |
| 42 | size_t byte) |
| 43 | { |
| 44 | const int nr = pool->curr_nr; |
| 45 | const int start = max_t(int, byte - (BITS_PER_LONG / 8), 0); |
| 46 | const int end = min_t(int, byte + (BITS_PER_LONG / 8), size); |
| 47 | int i; |
| 48 | |
| 49 | pr_err("BUG: mempool element poison mismatch\n" ); |
| 50 | pr_err("Mempool %p size %zu\n" , pool, size); |
| 51 | pr_err(" nr=%d @ %p: %s0x" , nr, element, start > 0 ? "... " : "" ); |
| 52 | for (i = start; i < end; i++) |
| 53 | pr_cont("%x " , *(u8 *)(element + i)); |
| 54 | pr_cont("%s\n" , end < size ? "..." : "" ); |
| 55 | dump_stack(); |
| 56 | } |
| 57 | |
| 58 | static void __check_element(struct mempool *pool, void *element, size_t size) |
| 59 | { |
| 60 | u8 *obj = element; |
| 61 | size_t i; |
| 62 | |
| 63 | for (i = 0; i < size; i++) { |
| 64 | u8 exp = (i < size - 1) ? POISON_FREE : POISON_END; |
| 65 | |
| 66 | if (obj[i] != exp) { |
| 67 | poison_error(pool, element, size, byte: i); |
| 68 | return; |
| 69 | } |
| 70 | } |
| 71 | memset(obj, POISON_INUSE, size); |
| 72 | } |
| 73 | |
| 74 | static void check_element(struct mempool *pool, void *element) |
| 75 | { |
| 76 | /* Skip checking: KASAN might save its metadata in the element. */ |
| 77 | if (kasan_enabled()) |
| 78 | return; |
| 79 | |
| 80 | /* Mempools backed by slab allocator */ |
| 81 | if (pool->free == mempool_kfree) { |
| 82 | __check_element(pool, element, size: (size_t)pool->pool_data); |
| 83 | } else if (pool->free == mempool_free_slab) { |
| 84 | __check_element(pool, element, size: kmem_cache_size(s: pool->pool_data)); |
| 85 | } else if (pool->free == mempool_free_pages) { |
| 86 | /* Mempools backed by page allocator */ |
| 87 | int order = (int)(long)pool->pool_data; |
| 88 | |
| 89 | #ifdef CONFIG_HIGHMEM |
| 90 | for (int i = 0; i < (1 << order); i++) { |
| 91 | struct page *page = (struct page *)element; |
| 92 | void *addr = kmap_local_page(page + i); |
| 93 | |
| 94 | __check_element(pool, addr, PAGE_SIZE); |
| 95 | kunmap_local(addr); |
| 96 | } |
| 97 | #else |
| 98 | void *addr = page_address((struct page *)element); |
| 99 | |
| 100 | __check_element(pool, element: addr, PAGE_SIZE << order); |
| 101 | #endif |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | static void __poison_element(void *element, size_t size) |
| 106 | { |
| 107 | u8 *obj = element; |
| 108 | |
| 109 | memset(obj, POISON_FREE, size - 1); |
| 110 | obj[size - 1] = POISON_END; |
| 111 | } |
| 112 | |
| 113 | static void poison_element(struct mempool *pool, void *element) |
| 114 | { |
| 115 | /* Skip poisoning: KASAN might save its metadata in the element. */ |
| 116 | if (kasan_enabled()) |
| 117 | return; |
| 118 | |
| 119 | /* Mempools backed by slab allocator */ |
| 120 | if (pool->alloc == mempool_kmalloc) { |
| 121 | __poison_element(element, size: (size_t)pool->pool_data); |
| 122 | } else if (pool->alloc == mempool_alloc_slab) { |
| 123 | __poison_element(element, size: kmem_cache_size(s: pool->pool_data)); |
| 124 | } else if (pool->alloc == mempool_alloc_pages) { |
| 125 | /* Mempools backed by page allocator */ |
| 126 | int order = (int)(long)pool->pool_data; |
| 127 | |
| 128 | #ifdef CONFIG_HIGHMEM |
| 129 | for (int i = 0; i < (1 << order); i++) { |
| 130 | struct page *page = (struct page *)element; |
| 131 | void *addr = kmap_local_page(page + i); |
| 132 | |
| 133 | __poison_element(addr, PAGE_SIZE); |
| 134 | kunmap_local(addr); |
| 135 | } |
| 136 | #else |
| 137 | void *addr = page_address((struct page *)element); |
| 138 | |
| 139 | __poison_element(element: addr, PAGE_SIZE << order); |
| 140 | #endif |
| 141 | } |
| 142 | } |
| 143 | #else /* CONFIG_SLUB_DEBUG_ON */ |
| 144 | static inline void check_element(struct mempool *pool, void *element) |
| 145 | { |
| 146 | } |
| 147 | static inline void poison_element(struct mempool *pool, void *element) |
| 148 | { |
| 149 | } |
| 150 | #endif /* CONFIG_SLUB_DEBUG_ON */ |
| 151 | |
| 152 | static __always_inline bool kasan_poison_element(struct mempool *pool, |
| 153 | void *element) |
| 154 | { |
| 155 | if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc) |
| 156 | return kasan_mempool_poison_object(ptr: element); |
| 157 | else if (pool->alloc == mempool_alloc_pages) |
| 158 | return kasan_mempool_poison_pages(page: element, |
| 159 | order: (unsigned long)pool->pool_data); |
| 160 | return true; |
| 161 | } |
| 162 | |
| 163 | static void kasan_unpoison_element(struct mempool *pool, void *element) |
| 164 | { |
| 165 | if (pool->alloc == mempool_kmalloc) |
| 166 | kasan_mempool_unpoison_object(ptr: element, size: (size_t)pool->pool_data); |
| 167 | else if (pool->alloc == mempool_alloc_slab) |
| 168 | kasan_mempool_unpoison_object(ptr: element, |
| 169 | size: kmem_cache_size(s: pool->pool_data)); |
| 170 | else if (pool->alloc == mempool_alloc_pages) |
| 171 | kasan_mempool_unpoison_pages(page: element, |
| 172 | order: (unsigned long)pool->pool_data); |
| 173 | } |
| 174 | |
| 175 | static __always_inline void add_element(struct mempool *pool, void *element) |
| 176 | { |
| 177 | BUG_ON(pool->min_nr != 0 && pool->curr_nr >= pool->min_nr); |
| 178 | poison_element(pool, element); |
| 179 | if (kasan_poison_element(pool, element)) |
| 180 | pool->elements[pool->curr_nr++] = element; |
| 181 | } |
| 182 | |
| 183 | static void *remove_element(struct mempool *pool) |
| 184 | { |
| 185 | void *element = pool->elements[--pool->curr_nr]; |
| 186 | |
| 187 | BUG_ON(pool->curr_nr < 0); |
| 188 | kasan_unpoison_element(pool, element); |
| 189 | check_element(pool, element); |
| 190 | return element; |
| 191 | } |
| 192 | |
| 193 | /** |
| 194 | * mempool_exit - exit a mempool initialized with mempool_init() |
| 195 | * @pool: pointer to the memory pool which was initialized with |
| 196 | * mempool_init(). |
| 197 | * |
| 198 | * Free all reserved elements in @pool and @pool itself. This function |
| 199 | * only sleeps if the free_fn() function sleeps. |
| 200 | * |
| 201 | * May be called on a zeroed but uninitialized mempool (i.e. allocated with |
| 202 | * kzalloc()). |
| 203 | */ |
| 204 | void mempool_exit(struct mempool *pool) |
| 205 | { |
| 206 | while (pool->curr_nr) { |
| 207 | void *element = remove_element(pool); |
| 208 | pool->free(element, pool->pool_data); |
| 209 | } |
| 210 | kfree(objp: pool->elements); |
| 211 | pool->elements = NULL; |
| 212 | } |
| 213 | EXPORT_SYMBOL(mempool_exit); |
| 214 | |
| 215 | /** |
| 216 | * mempool_destroy - deallocate a memory pool |
| 217 | * @pool: pointer to the memory pool which was allocated via |
| 218 | * mempool_create(). |
| 219 | * |
| 220 | * Free all reserved elements in @pool and @pool itself. This function |
| 221 | * only sleeps if the free_fn() function sleeps. |
| 222 | */ |
| 223 | void mempool_destroy(struct mempool *pool) |
| 224 | { |
| 225 | if (unlikely(!pool)) |
| 226 | return; |
| 227 | |
| 228 | mempool_exit(pool); |
| 229 | kfree(objp: pool); |
| 230 | } |
| 231 | EXPORT_SYMBOL(mempool_destroy); |
| 232 | |
| 233 | int mempool_init_node(struct mempool *pool, int min_nr, |
| 234 | mempool_alloc_t *alloc_fn, mempool_free_t *free_fn, |
| 235 | void *pool_data, gfp_t gfp_mask, int node_id) |
| 236 | { |
| 237 | spin_lock_init(&pool->lock); |
| 238 | pool->min_nr = min_nr; |
| 239 | pool->pool_data = pool_data; |
| 240 | pool->alloc = alloc_fn; |
| 241 | pool->free = free_fn; |
| 242 | init_waitqueue_head(&pool->wait); |
| 243 | /* |
| 244 | * max() used here to ensure storage for at least 1 element to support |
| 245 | * zero minimum pool |
| 246 | */ |
| 247 | pool->elements = kmalloc_array_node(max(1, min_nr), sizeof(void *), |
| 248 | gfp_mask, node_id); |
| 249 | if (!pool->elements) |
| 250 | return -ENOMEM; |
| 251 | |
| 252 | /* |
| 253 | * First pre-allocate the guaranteed number of buffers, |
| 254 | * also pre-allocate 1 element for zero minimum pool. |
| 255 | */ |
| 256 | while (pool->curr_nr < max(1, pool->min_nr)) { |
| 257 | void *element; |
| 258 | |
| 259 | element = pool->alloc(gfp_mask, pool->pool_data); |
| 260 | if (unlikely(!element)) { |
| 261 | mempool_exit(pool); |
| 262 | return -ENOMEM; |
| 263 | } |
| 264 | add_element(pool, element); |
| 265 | } |
| 266 | |
| 267 | return 0; |
| 268 | } |
| 269 | EXPORT_SYMBOL(mempool_init_node); |
| 270 | |
| 271 | /** |
| 272 | * mempool_init - initialize a memory pool |
| 273 | * @pool: pointer to the memory pool that should be initialized |
| 274 | * @min_nr: the minimum number of elements guaranteed to be |
| 275 | * allocated for this pool. |
| 276 | * @alloc_fn: user-defined element-allocation function. |
| 277 | * @free_fn: user-defined element-freeing function. |
| 278 | * @pool_data: optional private data available to the user-defined functions. |
| 279 | * |
| 280 | * Like mempool_create(), but initializes the pool in (i.e. embedded in another |
| 281 | * structure). |
| 282 | * |
| 283 | * Return: %0 on success, negative error code otherwise. |
| 284 | */ |
| 285 | int mempool_init_noprof(struct mempool *pool, int min_nr, |
| 286 | mempool_alloc_t *alloc_fn, mempool_free_t *free_fn, |
| 287 | void *pool_data) |
| 288 | { |
| 289 | return mempool_init_node(pool, min_nr, alloc_fn, free_fn, |
| 290 | pool_data, GFP_KERNEL, NUMA_NO_NODE); |
| 291 | |
| 292 | } |
| 293 | EXPORT_SYMBOL(mempool_init_noprof); |
| 294 | |
| 295 | /** |
| 296 | * mempool_create_node - create a memory pool |
| 297 | * @min_nr: the minimum number of elements guaranteed to be |
| 298 | * allocated for this pool. |
| 299 | * @alloc_fn: user-defined element-allocation function. |
| 300 | * @free_fn: user-defined element-freeing function. |
| 301 | * @pool_data: optional private data available to the user-defined functions. |
| 302 | * @gfp_mask: memory allocation flags |
| 303 | * @node_id: numa node to allocate on |
| 304 | * |
| 305 | * this function creates and allocates a guaranteed size, preallocated |
| 306 | * memory pool. The pool can be used from the mempool_alloc() and mempool_free() |
| 307 | * functions. This function might sleep. Both the alloc_fn() and the free_fn() |
| 308 | * functions might sleep - as long as the mempool_alloc() function is not called |
| 309 | * from IRQ contexts. |
| 310 | * |
| 311 | * Return: pointer to the created memory pool object or %NULL on error. |
| 312 | */ |
| 313 | struct mempool *mempool_create_node_noprof(int min_nr, |
| 314 | mempool_alloc_t *alloc_fn, mempool_free_t *free_fn, |
| 315 | void *pool_data, gfp_t gfp_mask, int node_id) |
| 316 | { |
| 317 | struct mempool *pool; |
| 318 | |
| 319 | pool = kmalloc_node_noprof(size: sizeof(*pool), flags: gfp_mask | __GFP_ZERO, node: node_id); |
| 320 | if (!pool) |
| 321 | return NULL; |
| 322 | |
| 323 | if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data, |
| 324 | gfp_mask, node_id)) { |
| 325 | kfree(objp: pool); |
| 326 | return NULL; |
| 327 | } |
| 328 | |
| 329 | return pool; |
| 330 | } |
| 331 | EXPORT_SYMBOL(mempool_create_node_noprof); |
| 332 | |
| 333 | /** |
| 334 | * mempool_resize - resize an existing memory pool |
| 335 | * @pool: pointer to the memory pool which was allocated via |
| 336 | * mempool_create(). |
| 337 | * @new_min_nr: the new minimum number of elements guaranteed to be |
| 338 | * allocated for this pool. |
| 339 | * |
| 340 | * This function shrinks/grows the pool. In the case of growing, |
| 341 | * it cannot be guaranteed that the pool will be grown to the new |
| 342 | * size immediately, but new mempool_free() calls will refill it. |
| 343 | * This function may sleep. |
| 344 | * |
| 345 | * Note, the caller must guarantee that no mempool_destroy is called |
| 346 | * while this function is running. mempool_alloc() & mempool_free() |
| 347 | * might be called (eg. from IRQ contexts) while this function executes. |
| 348 | * |
| 349 | * Return: %0 on success, negative error code otherwise. |
| 350 | */ |
| 351 | int mempool_resize(struct mempool *pool, int new_min_nr) |
| 352 | { |
| 353 | void *element; |
| 354 | void **new_elements; |
| 355 | unsigned long flags; |
| 356 | |
| 357 | BUG_ON(new_min_nr <= 0); |
| 358 | might_sleep(); |
| 359 | |
| 360 | spin_lock_irqsave(&pool->lock, flags); |
| 361 | if (new_min_nr <= pool->min_nr) { |
| 362 | while (new_min_nr < pool->curr_nr) { |
| 363 | element = remove_element(pool); |
| 364 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 365 | pool->free(element, pool->pool_data); |
| 366 | spin_lock_irqsave(&pool->lock, flags); |
| 367 | } |
| 368 | pool->min_nr = new_min_nr; |
| 369 | goto out_unlock; |
| 370 | } |
| 371 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 372 | |
| 373 | /* Grow the pool */ |
| 374 | new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements), |
| 375 | GFP_KERNEL); |
| 376 | if (!new_elements) |
| 377 | return -ENOMEM; |
| 378 | |
| 379 | spin_lock_irqsave(&pool->lock, flags); |
| 380 | if (unlikely(new_min_nr <= pool->min_nr)) { |
| 381 | /* Raced, other resize will do our work */ |
| 382 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 383 | kfree(objp: new_elements); |
| 384 | goto out; |
| 385 | } |
| 386 | memcpy(new_elements, pool->elements, |
| 387 | pool->curr_nr * sizeof(*new_elements)); |
| 388 | kfree(objp: pool->elements); |
| 389 | pool->elements = new_elements; |
| 390 | pool->min_nr = new_min_nr; |
| 391 | |
| 392 | while (pool->curr_nr < pool->min_nr) { |
| 393 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 394 | element = pool->alloc(GFP_KERNEL, pool->pool_data); |
| 395 | if (!element) |
| 396 | goto out; |
| 397 | spin_lock_irqsave(&pool->lock, flags); |
| 398 | if (pool->curr_nr < pool->min_nr) { |
| 399 | add_element(pool, element); |
| 400 | } else { |
| 401 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 402 | pool->free(element, pool->pool_data); /* Raced */ |
| 403 | goto out; |
| 404 | } |
| 405 | } |
| 406 | out_unlock: |
| 407 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 408 | out: |
| 409 | return 0; |
| 410 | } |
| 411 | EXPORT_SYMBOL(mempool_resize); |
| 412 | |
| 413 | static unsigned int mempool_alloc_from_pool(struct mempool *pool, void **elems, |
| 414 | unsigned int count, unsigned int allocated, |
| 415 | gfp_t gfp_mask) |
| 416 | { |
| 417 | unsigned long flags; |
| 418 | unsigned int i; |
| 419 | |
| 420 | spin_lock_irqsave(&pool->lock, flags); |
| 421 | if (unlikely(pool->curr_nr < count - allocated)) |
| 422 | goto fail; |
| 423 | for (i = 0; i < count; i++) { |
| 424 | if (!elems[i]) { |
| 425 | elems[i] = remove_element(pool); |
| 426 | allocated++; |
| 427 | } |
| 428 | } |
| 429 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 430 | |
| 431 | /* Paired with rmb in mempool_free(), read comment there. */ |
| 432 | smp_wmb(); |
| 433 | |
| 434 | /* |
| 435 | * Update the allocation stack trace as this is more useful for |
| 436 | * debugging. |
| 437 | */ |
| 438 | for (i = 0; i < count; i++) |
| 439 | kmemleak_update_trace(ptr: elems[i]); |
| 440 | return allocated; |
| 441 | |
| 442 | fail: |
| 443 | if (gfp_mask & __GFP_DIRECT_RECLAIM) { |
| 444 | DEFINE_WAIT(wait); |
| 445 | |
| 446 | prepare_to_wait(wq_head: &pool->wait, wq_entry: &wait, TASK_UNINTERRUPTIBLE); |
| 447 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 448 | |
| 449 | /* |
| 450 | * Wait for someone else to return an element to @pool, but wake |
| 451 | * up occasionally as memory pressure might have reduced even |
| 452 | * and the normal allocation in alloc_fn could succeed even if |
| 453 | * no element was returned. |
| 454 | */ |
| 455 | io_schedule_timeout(timeout: 5 * HZ); |
| 456 | finish_wait(wq_head: &pool->wait, wq_entry: &wait); |
| 457 | } else { |
| 458 | /* We must not sleep if __GFP_DIRECT_RECLAIM is not set. */ |
| 459 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 460 | } |
| 461 | |
| 462 | return allocated; |
| 463 | } |
| 464 | |
| 465 | /* |
| 466 | * Adjust the gfp flags for mempool allocations, as we never want to dip into |
| 467 | * the global emergency reserves or retry in the page allocator. |
| 468 | * |
| 469 | * The first pass also doesn't want to go reclaim, but the next passes do, so |
| 470 | * return a separate subset for that first iteration. |
| 471 | */ |
| 472 | static inline gfp_t mempool_adjust_gfp(gfp_t *gfp_mask) |
| 473 | { |
| 474 | *gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
| 475 | return *gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); |
| 476 | } |
| 477 | |
| 478 | /** |
| 479 | * mempool_alloc_bulk - allocate multiple elements from a memory pool |
| 480 | * @pool: pointer to the memory pool |
| 481 | * @elems: partially or fully populated elements array |
| 482 | * @count: number of entries in @elem that need to be allocated |
| 483 | * @allocated: number of entries in @elem already allocated |
| 484 | * |
| 485 | * Allocate elements for each slot in @elem that is non-%NULL. This is done by |
| 486 | * first calling into the alloc_fn supplied at pool initialization time, and |
| 487 | * dipping into the reserved pool when alloc_fn fails to allocate an element. |
| 488 | * |
| 489 | * On return all @count elements in @elems will be populated. |
| 490 | * |
| 491 | * Return: Always 0. If it wasn't for %$#^$ alloc tags, it would return void. |
| 492 | */ |
| 493 | int mempool_alloc_bulk_noprof(struct mempool *pool, void **elems, |
| 494 | unsigned int count, unsigned int allocated) |
| 495 | { |
| 496 | gfp_t gfp_mask = GFP_KERNEL; |
| 497 | gfp_t gfp_temp = mempool_adjust_gfp(gfp_mask: &gfp_mask); |
| 498 | unsigned int i = 0; |
| 499 | |
| 500 | VM_WARN_ON_ONCE(count > pool->min_nr); |
| 501 | might_alloc(gfp_mask); |
| 502 | |
| 503 | /* |
| 504 | * If an error is injected, fail all elements in a bulk allocation so |
| 505 | * that we stress the multiple elements missing path. |
| 506 | */ |
| 507 | if (should_fail_ex(attr: &fail_mempool_alloc_bulk, size: 1, flags: FAULT_NOWARN)) { |
| 508 | pr_info("forcing mempool usage for %pS\n" , |
| 509 | (void *)_RET_IP_); |
| 510 | goto use_pool; |
| 511 | } |
| 512 | |
| 513 | repeat_alloc: |
| 514 | /* |
| 515 | * Try to allocate the elements using the allocation callback first as |
| 516 | * that might succeed even when the caller's bulk allocation did not. |
| 517 | */ |
| 518 | for (i = 0; i < count; i++) { |
| 519 | if (elems[i]) |
| 520 | continue; |
| 521 | elems[i] = pool->alloc(gfp_temp, pool->pool_data); |
| 522 | if (unlikely(!elems[i])) |
| 523 | goto use_pool; |
| 524 | allocated++; |
| 525 | } |
| 526 | |
| 527 | return 0; |
| 528 | |
| 529 | use_pool: |
| 530 | allocated = mempool_alloc_from_pool(pool, elems, count, allocated, |
| 531 | gfp_mask: gfp_temp); |
| 532 | gfp_temp = gfp_mask; |
| 533 | goto repeat_alloc; |
| 534 | } |
| 535 | EXPORT_SYMBOL_GPL(mempool_alloc_bulk_noprof); |
| 536 | |
| 537 | /** |
| 538 | * mempool_alloc - allocate an element from a memory pool |
| 539 | * @pool: pointer to the memory pool |
| 540 | * @gfp_mask: GFP_* flags. %__GFP_ZERO is not supported. |
| 541 | * |
| 542 | * Allocate an element from @pool. This is done by first calling into the |
| 543 | * alloc_fn supplied at pool initialization time, and dipping into the reserved |
| 544 | * pool when alloc_fn fails to allocate an element. |
| 545 | * |
| 546 | * This function only sleeps if the alloc_fn callback sleeps, or when waiting |
| 547 | * for elements to become available in the pool. |
| 548 | * |
| 549 | * Return: pointer to the allocated element or %NULL when failing to allocate |
| 550 | * an element. Allocation failure can only happen when @gfp_mask does not |
| 551 | * include %__GFP_DIRECT_RECLAIM. |
| 552 | */ |
| 553 | void *mempool_alloc_noprof(struct mempool *pool, gfp_t gfp_mask) |
| 554 | { |
| 555 | gfp_t gfp_temp = mempool_adjust_gfp(gfp_mask: &gfp_mask); |
| 556 | void *element; |
| 557 | |
| 558 | VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO); |
| 559 | might_alloc(gfp_mask); |
| 560 | |
| 561 | repeat_alloc: |
| 562 | if (should_fail_ex(attr: &fail_mempool_alloc, size: 1, flags: FAULT_NOWARN)) { |
| 563 | pr_info("forcing mempool usage for %pS\n" , |
| 564 | (void *)_RET_IP_); |
| 565 | element = NULL; |
| 566 | } else { |
| 567 | element = pool->alloc(gfp_temp, pool->pool_data); |
| 568 | } |
| 569 | |
| 570 | if (unlikely(!element)) { |
| 571 | /* |
| 572 | * Try to allocate an element from the pool. |
| 573 | * |
| 574 | * The first pass won't have __GFP_DIRECT_RECLAIM and won't |
| 575 | * sleep in mempool_alloc_from_pool. Retry the allocation |
| 576 | * with all flags set in that case. |
| 577 | */ |
| 578 | if (!mempool_alloc_from_pool(pool, elems: &element, count: 1, allocated: 0, gfp_mask: gfp_temp)) { |
| 579 | if (gfp_temp != gfp_mask) { |
| 580 | gfp_temp = gfp_mask; |
| 581 | goto repeat_alloc; |
| 582 | } |
| 583 | if (gfp_mask & __GFP_DIRECT_RECLAIM) { |
| 584 | goto repeat_alloc; |
| 585 | } |
| 586 | } |
| 587 | } |
| 588 | |
| 589 | return element; |
| 590 | } |
| 591 | EXPORT_SYMBOL(mempool_alloc_noprof); |
| 592 | |
| 593 | /** |
| 594 | * mempool_alloc_preallocated - allocate an element from preallocated elements |
| 595 | * belonging to a memory pool |
| 596 | * @pool: pointer to the memory pool |
| 597 | * |
| 598 | * This function is similar to mempool_alloc(), but it only attempts allocating |
| 599 | * an element from the preallocated elements. It only takes a single spinlock_t |
| 600 | * and immediately returns if no preallocated elements are available. |
| 601 | * |
| 602 | * Return: pointer to the allocated element or %NULL if no elements are |
| 603 | * available. |
| 604 | */ |
| 605 | void *mempool_alloc_preallocated(struct mempool *pool) |
| 606 | { |
| 607 | void *element = NULL; |
| 608 | |
| 609 | mempool_alloc_from_pool(pool, elems: &element, count: 1, allocated: 0, GFP_NOWAIT); |
| 610 | return element; |
| 611 | } |
| 612 | EXPORT_SYMBOL(mempool_alloc_preallocated); |
| 613 | |
| 614 | /** |
| 615 | * mempool_free_bulk - return elements to a mempool |
| 616 | * @pool: pointer to the memory pool |
| 617 | * @elems: elements to return |
| 618 | * @count: number of elements to return |
| 619 | * |
| 620 | * Returns a number of elements from the start of @elem to @pool if @pool needs |
| 621 | * replenishing and sets their slots in @elem to NULL. Other elements are left |
| 622 | * in @elem. |
| 623 | * |
| 624 | * Return: number of elements transferred to @pool. Elements are always |
| 625 | * transferred from the beginning of @elem, so the return value can be used as |
| 626 | * an offset into @elem for the freeing the remaining elements in the caller. |
| 627 | */ |
| 628 | unsigned int mempool_free_bulk(struct mempool *pool, void **elems, |
| 629 | unsigned int count) |
| 630 | { |
| 631 | unsigned long flags; |
| 632 | unsigned int freed = 0; |
| 633 | bool added = false; |
| 634 | |
| 635 | /* |
| 636 | * Paired with the wmb in mempool_alloc(). The preceding read is |
| 637 | * for @element and the following @pool->curr_nr. This ensures |
| 638 | * that the visible value of @pool->curr_nr is from after the |
| 639 | * allocation of @element. This is necessary for fringe cases |
| 640 | * where @element was passed to this task without going through |
| 641 | * barriers. |
| 642 | * |
| 643 | * For example, assume @p is %NULL at the beginning and one task |
| 644 | * performs "p = mempool_alloc(...);" while another task is doing |
| 645 | * "while (!p) cpu_relax(); mempool_free(p, ...);". This function |
| 646 | * may end up using curr_nr value which is from before allocation |
| 647 | * of @p without the following rmb. |
| 648 | */ |
| 649 | smp_rmb(); |
| 650 | |
| 651 | /* |
| 652 | * For correctness, we need a test which is guaranteed to trigger |
| 653 | * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr |
| 654 | * without locking achieves that and refilling as soon as possible |
| 655 | * is desirable. |
| 656 | * |
| 657 | * Because curr_nr visible here is always a value after the |
| 658 | * allocation of @element, any task which decremented curr_nr below |
| 659 | * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets |
| 660 | * incremented to min_nr afterwards. If curr_nr gets incremented |
| 661 | * to min_nr after the allocation of @element, the elements |
| 662 | * allocated after that are subject to the same guarantee. |
| 663 | * |
| 664 | * Waiters happen iff curr_nr is 0 and the above guarantee also |
| 665 | * ensures that there will be frees which return elements to the |
| 666 | * pool waking up the waiters. |
| 667 | * |
| 668 | * For zero-minimum pools, curr_nr < min_nr (0 < 0) never succeeds, |
| 669 | * so waiters sleeping on pool->wait would never be woken by the |
| 670 | * wake-up path of previous test. This explicit check ensures the |
| 671 | * allocation of element when both min_nr and curr_nr are 0, and |
| 672 | * any active waiters are properly awakened. |
| 673 | */ |
| 674 | if (unlikely(READ_ONCE(pool->curr_nr) < pool->min_nr)) { |
| 675 | spin_lock_irqsave(&pool->lock, flags); |
| 676 | while (pool->curr_nr < pool->min_nr && freed < count) { |
| 677 | add_element(pool, element: elems[freed++]); |
| 678 | added = true; |
| 679 | } |
| 680 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 681 | } else if (unlikely(pool->min_nr == 0 && |
| 682 | READ_ONCE(pool->curr_nr) == 0)) { |
| 683 | /* Handle the min_nr = 0 edge case: */ |
| 684 | spin_lock_irqsave(&pool->lock, flags); |
| 685 | if (likely(pool->curr_nr == 0)) { |
| 686 | add_element(pool, element: elems[freed++]); |
| 687 | added = true; |
| 688 | } |
| 689 | spin_unlock_irqrestore(lock: &pool->lock, flags); |
| 690 | } |
| 691 | |
| 692 | if (unlikely(added) && wq_has_sleeper(wq_head: &pool->wait)) |
| 693 | wake_up(&pool->wait); |
| 694 | |
| 695 | return freed; |
| 696 | } |
| 697 | EXPORT_SYMBOL_GPL(mempool_free_bulk); |
| 698 | |
| 699 | /** |
| 700 | * mempool_free - return an element to the pool. |
| 701 | * @element: element to return |
| 702 | * @pool: pointer to the memory pool |
| 703 | * |
| 704 | * Returns @element to @pool if it needs replenishing, else frees it using |
| 705 | * the free_fn callback in @pool. |
| 706 | * |
| 707 | * This function only sleeps if the free_fn callback sleeps. |
| 708 | */ |
| 709 | void mempool_free(void *element, struct mempool *pool) |
| 710 | { |
| 711 | if (likely(element) && !mempool_free_bulk(pool, &element, 1)) |
| 712 | pool->free(element, pool->pool_data); |
| 713 | } |
| 714 | EXPORT_SYMBOL(mempool_free); |
| 715 | |
| 716 | /* |
| 717 | * A commonly used alloc and free fn. |
| 718 | */ |
| 719 | void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data) |
| 720 | { |
| 721 | struct kmem_cache *mem = pool_data; |
| 722 | VM_BUG_ON(mem->ctor); |
| 723 | return kmem_cache_alloc_noprof(cachep: mem, flags: gfp_mask); |
| 724 | } |
| 725 | EXPORT_SYMBOL(mempool_alloc_slab); |
| 726 | |
| 727 | void mempool_free_slab(void *element, void *pool_data) |
| 728 | { |
| 729 | struct kmem_cache *mem = pool_data; |
| 730 | kmem_cache_free(s: mem, objp: element); |
| 731 | } |
| 732 | EXPORT_SYMBOL(mempool_free_slab); |
| 733 | |
| 734 | /* |
| 735 | * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory |
| 736 | * specified by pool_data |
| 737 | */ |
| 738 | void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data) |
| 739 | { |
| 740 | size_t size = (size_t)pool_data; |
| 741 | return kmalloc_noprof(size, flags: gfp_mask); |
| 742 | } |
| 743 | EXPORT_SYMBOL(mempool_kmalloc); |
| 744 | |
| 745 | void mempool_kfree(void *element, void *pool_data) |
| 746 | { |
| 747 | kfree(objp: element); |
| 748 | } |
| 749 | EXPORT_SYMBOL(mempool_kfree); |
| 750 | |
| 751 | /* |
| 752 | * A simple mempool-backed page allocator that allocates pages |
| 753 | * of the order specified by pool_data. |
| 754 | */ |
| 755 | void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data) |
| 756 | { |
| 757 | int order = (int)(long)pool_data; |
| 758 | return alloc_pages_noprof(gfp: gfp_mask, order); |
| 759 | } |
| 760 | EXPORT_SYMBOL(mempool_alloc_pages); |
| 761 | |
| 762 | void mempool_free_pages(void *element, void *pool_data) |
| 763 | { |
| 764 | int order = (int)(long)pool_data; |
| 765 | __free_pages(page: element, order); |
| 766 | } |
| 767 | EXPORT_SYMBOL(mempool_free_pages); |
| 768 | |