Marek Szyprowski | 55bb033 | 2011-12-29 13:09:51 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Contiguous Memory Allocator for DMA mapping framework |
| 3 | * Copyright (c) 2010-2011 by Samsung Electronics. |
| 4 | * Written by: |
| 5 | * Marek Szyprowski <m.szyprowski@samsung.com> |
| 6 | * Michal Nazarewicz <mina86@mina86.com> |
| 7 | * |
| 8 | * This program is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU General Public License as |
| 10 | * published by the Free Software Foundation; either version 2 of the |
| 11 | * License or (at your optional) any later version of the license. |
| 12 | */ |
| 13 | |
| 14 | #define pr_fmt(fmt) "cma: " fmt |
| 15 | |
| 16 | #ifdef CONFIG_CMA_DEBUG |
| 17 | #ifndef DEBUG |
| 18 | # define DEBUG |
| 19 | #endif |
| 20 | #endif |
| 21 | |
| 22 | #include <asm/page.h> |
| 23 | #include <asm/dma-contiguous.h> |
| 24 | |
| 25 | #include <linux/memblock.h> |
| 26 | #include <linux/err.h> |
| 27 | #include <linux/mm.h> |
| 28 | #include <linux/mutex.h> |
| 29 | #include <linux/page-isolation.h> |
| 30 | #include <linux/slab.h> |
| 31 | #include <linux/swap.h> |
| 32 | #include <linux/mm_types.h> |
| 33 | #include <linux/dma-contiguous.h> |
| 34 | |
| 35 | #ifndef SZ_1M |
| 36 | #define SZ_1M (1 << 20) |
| 37 | #endif |
| 38 | |
| 39 | struct cma { |
| 40 | unsigned long base_pfn; |
| 41 | unsigned long count; |
| 42 | unsigned long *bitmap; |
| 43 | }; |
| 44 | |
| 45 | struct cma *dma_contiguous_default_area; |
| 46 | |
| 47 | #ifdef CONFIG_CMA_SIZE_MBYTES |
| 48 | #define CMA_SIZE_MBYTES CONFIG_CMA_SIZE_MBYTES |
| 49 | #else |
| 50 | #define CMA_SIZE_MBYTES 0 |
| 51 | #endif |
| 52 | |
| 53 | /* |
| 54 | * Default global CMA area size can be defined in kernel's .config. |
| 55 | * This is usefull mainly for distro maintainers to create a kernel |
| 56 | * that works correctly for most supported systems. |
| 57 | * The size can be set in bytes or as a percentage of the total memory |
| 58 | * in the system. |
| 59 | * |
| 60 | * Users, who want to set the size of global CMA area for their system |
| 61 | * should use cma= kernel parameter. |
| 62 | */ |
| 63 | static const unsigned long size_bytes = CMA_SIZE_MBYTES * SZ_1M; |
| 64 | static long size_cmdline = -1; |
| 65 | |
| 66 | static int __init early_cma(char *p) |
| 67 | { |
| 68 | pr_debug("%s(%s)\n", __func__, p); |
| 69 | size_cmdline = memparse(p, &p); |
| 70 | return 0; |
| 71 | } |
| 72 | early_param("cma", early_cma); |
| 73 | |
| 74 | #ifdef CONFIG_CMA_SIZE_PERCENTAGE |
| 75 | |
| 76 | static unsigned long __init __maybe_unused cma_early_percent_memory(void) |
| 77 | { |
| 78 | struct memblock_region *reg; |
| 79 | unsigned long total_pages = 0; |
| 80 | |
| 81 | /* |
| 82 | * We cannot use memblock_phys_mem_size() here, because |
| 83 | * memblock_analyze() has not been called yet. |
| 84 | */ |
| 85 | for_each_memblock(memory, reg) |
| 86 | total_pages += memblock_region_memory_end_pfn(reg) - |
| 87 | memblock_region_memory_base_pfn(reg); |
| 88 | |
| 89 | return (total_pages * CONFIG_CMA_SIZE_PERCENTAGE / 100) << PAGE_SHIFT; |
| 90 | } |
| 91 | |
| 92 | #else |
| 93 | |
| 94 | static inline __maybe_unused unsigned long cma_early_percent_memory(void) |
| 95 | { |
| 96 | return 0; |
| 97 | } |
| 98 | |
| 99 | #endif |
| 100 | |
| 101 | /** |
| 102 | * dma_contiguous_reserve() - reserve area for contiguous memory handling |
| 103 | * @limit: End address of the reserved memory (optional, 0 for any). |
| 104 | * |
| 105 | * This function reserves memory from early allocator. It should be |
| 106 | * called by arch specific code once the early allocator (memblock or bootmem) |
| 107 | * has been activated and all other subsystems have already allocated/reserved |
| 108 | * memory. |
| 109 | */ |
| 110 | void __init dma_contiguous_reserve(phys_addr_t limit) |
| 111 | { |
| 112 | unsigned long selected_size = 0; |
| 113 | |
| 114 | pr_debug("%s(limit %08lx)\n", __func__, (unsigned long)limit); |
| 115 | |
| 116 | if (size_cmdline != -1) { |
| 117 | selected_size = size_cmdline; |
| 118 | } else { |
| 119 | #ifdef CONFIG_CMA_SIZE_SEL_MBYTES |
| 120 | selected_size = size_bytes; |
| 121 | #elif defined(CONFIG_CMA_SIZE_SEL_PERCENTAGE) |
| 122 | selected_size = cma_early_percent_memory(); |
| 123 | #elif defined(CONFIG_CMA_SIZE_SEL_MIN) |
| 124 | selected_size = min(size_bytes, cma_early_percent_memory()); |
| 125 | #elif defined(CONFIG_CMA_SIZE_SEL_MAX) |
| 126 | selected_size = max(size_bytes, cma_early_percent_memory()); |
| 127 | #endif |
| 128 | } |
| 129 | |
| 130 | if (selected_size) { |
| 131 | pr_debug("%s: reserving %ld MiB for global area\n", __func__, |
| 132 | selected_size / SZ_1M); |
| 133 | |
| 134 | dma_declare_contiguous(NULL, selected_size, 0, limit); |
| 135 | } |
| 136 | }; |
| 137 | |
| 138 | static DEFINE_MUTEX(cma_mutex); |
| 139 | |
| 140 | static __init int cma_activate_area(unsigned long base_pfn, unsigned long count) |
| 141 | { |
| 142 | unsigned long pfn = base_pfn; |
| 143 | unsigned i = count >> pageblock_order; |
| 144 | struct zone *zone; |
| 145 | |
| 146 | WARN_ON_ONCE(!pfn_valid(pfn)); |
| 147 | zone = page_zone(pfn_to_page(pfn)); |
| 148 | |
| 149 | do { |
| 150 | unsigned j; |
| 151 | base_pfn = pfn; |
| 152 | for (j = pageblock_nr_pages; j; --j, pfn++) { |
| 153 | WARN_ON_ONCE(!pfn_valid(pfn)); |
| 154 | if (page_zone(pfn_to_page(pfn)) != zone) |
| 155 | return -EINVAL; |
| 156 | } |
| 157 | init_cma_reserved_pageblock(pfn_to_page(base_pfn)); |
| 158 | } while (--i); |
| 159 | return 0; |
| 160 | } |
| 161 | |
| 162 | static __init struct cma *cma_create_area(unsigned long base_pfn, |
| 163 | unsigned long count) |
| 164 | { |
| 165 | int bitmap_size = BITS_TO_LONGS(count) * sizeof(long); |
| 166 | struct cma *cma; |
| 167 | int ret = -ENOMEM; |
| 168 | |
| 169 | pr_debug("%s(base %08lx, count %lx)\n", __func__, base_pfn, count); |
| 170 | |
| 171 | cma = kmalloc(sizeof *cma, GFP_KERNEL); |
| 172 | if (!cma) |
| 173 | return ERR_PTR(-ENOMEM); |
| 174 | |
| 175 | cma->base_pfn = base_pfn; |
| 176 | cma->count = count; |
| 177 | cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL); |
| 178 | |
| 179 | if (!cma->bitmap) |
| 180 | goto no_mem; |
| 181 | |
| 182 | ret = cma_activate_area(base_pfn, count); |
| 183 | if (ret) |
| 184 | goto error; |
| 185 | |
| 186 | pr_debug("%s: returned %p\n", __func__, (void *)cma); |
| 187 | return cma; |
| 188 | |
| 189 | error: |
| 190 | kfree(cma->bitmap); |
| 191 | no_mem: |
| 192 | kfree(cma); |
| 193 | return ERR_PTR(ret); |
| 194 | } |
| 195 | |
| 196 | static struct cma_reserved { |
| 197 | phys_addr_t start; |
| 198 | unsigned long size; |
| 199 | struct device *dev; |
| 200 | } cma_reserved[MAX_CMA_AREAS] __initdata; |
| 201 | static unsigned cma_reserved_count __initdata; |
| 202 | |
| 203 | static int __init cma_init_reserved_areas(void) |
| 204 | { |
| 205 | struct cma_reserved *r = cma_reserved; |
| 206 | unsigned i = cma_reserved_count; |
| 207 | |
| 208 | pr_debug("%s()\n", __func__); |
| 209 | |
| 210 | for (; i; --i, ++r) { |
| 211 | struct cma *cma; |
| 212 | cma = cma_create_area(PFN_DOWN(r->start), |
| 213 | r->size >> PAGE_SHIFT); |
| 214 | if (!IS_ERR(cma)) |
| 215 | dev_set_cma_area(r->dev, cma); |
| 216 | } |
| 217 | return 0; |
| 218 | } |
| 219 | core_initcall(cma_init_reserved_areas); |
| 220 | |
| 221 | /** |
| 222 | * dma_declare_contiguous() - reserve area for contiguous memory handling |
| 223 | * for particular device |
| 224 | * @dev: Pointer to device structure. |
| 225 | * @size: Size of the reserved memory. |
| 226 | * @base: Start address of the reserved memory (optional, 0 for any). |
| 227 | * @limit: End address of the reserved memory (optional, 0 for any). |
| 228 | * |
| 229 | * This function reserves memory for specified device. It should be |
| 230 | * called by board specific code when early allocator (memblock or bootmem) |
| 231 | * is still activate. |
| 232 | */ |
| 233 | int __init dma_declare_contiguous(struct device *dev, unsigned long size, |
| 234 | phys_addr_t base, phys_addr_t limit) |
| 235 | { |
| 236 | struct cma_reserved *r = &cma_reserved[cma_reserved_count]; |
| 237 | unsigned long alignment; |
| 238 | |
| 239 | pr_debug("%s(size %lx, base %08lx, limit %08lx)\n", __func__, |
| 240 | (unsigned long)size, (unsigned long)base, |
| 241 | (unsigned long)limit); |
| 242 | |
| 243 | /* Sanity checks */ |
| 244 | if (cma_reserved_count == ARRAY_SIZE(cma_reserved)) { |
| 245 | pr_err("Not enough slots for CMA reserved regions!\n"); |
| 246 | return -ENOSPC; |
| 247 | } |
| 248 | |
| 249 | if (!size) |
| 250 | return -EINVAL; |
| 251 | |
| 252 | /* Sanitise input arguments */ |
| 253 | alignment = PAGE_SIZE << max(MAX_ORDER, pageblock_order); |
| 254 | base = ALIGN(base, alignment); |
| 255 | size = ALIGN(size, alignment); |
| 256 | limit &= ~(alignment - 1); |
| 257 | |
| 258 | /* Reserve memory */ |
| 259 | if (base) { |
| 260 | if (memblock_is_region_reserved(base, size) || |
| 261 | memblock_reserve(base, size) < 0) { |
| 262 | base = -EBUSY; |
| 263 | goto err; |
| 264 | } |
| 265 | } else { |
| 266 | /* |
| 267 | * Use __memblock_alloc_base() since |
| 268 | * memblock_alloc_base() panic()s. |
| 269 | */ |
| 270 | phys_addr_t addr = __memblock_alloc_base(size, alignment, limit); |
| 271 | if (!addr) { |
| 272 | base = -ENOMEM; |
| 273 | goto err; |
| 274 | } else if (addr + size > ~(unsigned long)0) { |
| 275 | memblock_free(addr, size); |
| 276 | base = -EINVAL; |
| 277 | goto err; |
| 278 | } else { |
| 279 | base = addr; |
| 280 | } |
| 281 | } |
| 282 | |
| 283 | /* |
| 284 | * Each reserved area must be initialised later, when more kernel |
| 285 | * subsystems (like slab allocator) are available. |
| 286 | */ |
| 287 | r->start = base; |
| 288 | r->size = size; |
| 289 | r->dev = dev; |
| 290 | cma_reserved_count++; |
| 291 | pr_info("CMA: reserved %ld MiB at %08lx\n", size / SZ_1M, |
| 292 | (unsigned long)base); |
| 293 | |
| 294 | /* Architecture specific contiguous memory fixup. */ |
| 295 | dma_contiguous_early_fixup(base, size); |
| 296 | return 0; |
| 297 | err: |
| 298 | pr_err("CMA: failed to reserve %ld MiB\n", size / SZ_1M); |
| 299 | return base; |
| 300 | } |
| 301 | |
| 302 | /** |
| 303 | * dma_alloc_from_contiguous() - allocate pages from contiguous area |
| 304 | * @dev: Pointer to device for which the allocation is performed. |
| 305 | * @count: Requested number of pages. |
| 306 | * @align: Requested alignment of pages (in PAGE_SIZE order). |
| 307 | * |
| 308 | * This function allocates memory buffer for specified device. It uses |
| 309 | * device specific contiguous memory area if available or the default |
| 310 | * global one. Requires architecture specific get_dev_cma_area() helper |
| 311 | * function. |
| 312 | */ |
| 313 | struct page *dma_alloc_from_contiguous(struct device *dev, int count, |
| 314 | unsigned int align) |
| 315 | { |
| 316 | unsigned long mask, pfn, pageno, start = 0; |
| 317 | struct cma *cma = dev_get_cma_area(dev); |
| 318 | int ret; |
| 319 | |
| 320 | if (!cma || !cma->count) |
| 321 | return NULL; |
| 322 | |
| 323 | if (align > CONFIG_CMA_ALIGNMENT) |
| 324 | align = CONFIG_CMA_ALIGNMENT; |
| 325 | |
| 326 | pr_debug("%s(cma %p, count %d, align %d)\n", __func__, (void *)cma, |
| 327 | count, align); |
| 328 | |
| 329 | if (!count) |
| 330 | return NULL; |
| 331 | |
| 332 | mask = (1 << align) - 1; |
| 333 | |
| 334 | mutex_lock(&cma_mutex); |
| 335 | |
| 336 | for (;;) { |
| 337 | pageno = bitmap_find_next_zero_area(cma->bitmap, cma->count, |
| 338 | start, count, mask); |
| 339 | if (pageno >= cma->count) { |
| 340 | ret = -ENOMEM; |
| 341 | goto error; |
| 342 | } |
| 343 | |
| 344 | pfn = cma->base_pfn + pageno; |
| 345 | ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA); |
| 346 | if (ret == 0) { |
| 347 | bitmap_set(cma->bitmap, pageno, count); |
| 348 | break; |
| 349 | } else if (ret != -EBUSY) { |
| 350 | goto error; |
| 351 | } |
| 352 | pr_debug("%s(): memory range at %p is busy, retrying\n", |
| 353 | __func__, pfn_to_page(pfn)); |
| 354 | /* try again with a bit different memory target */ |
| 355 | start = pageno + mask + 1; |
| 356 | } |
| 357 | |
| 358 | mutex_unlock(&cma_mutex); |
| 359 | |
| 360 | pr_debug("%s(): returned %p\n", __func__, pfn_to_page(pfn)); |
| 361 | return pfn_to_page(pfn); |
| 362 | error: |
| 363 | mutex_unlock(&cma_mutex); |
| 364 | return NULL; |
| 365 | } |
| 366 | |
| 367 | /** |
| 368 | * dma_release_from_contiguous() - release allocated pages |
| 369 | * @dev: Pointer to device for which the pages were allocated. |
| 370 | * @pages: Allocated pages. |
| 371 | * @count: Number of allocated pages. |
| 372 | * |
| 373 | * This function releases memory allocated by dma_alloc_from_contiguous(). |
| 374 | * It returns false when provided pages do not belong to contiguous area and |
| 375 | * true otherwise. |
| 376 | */ |
| 377 | bool dma_release_from_contiguous(struct device *dev, struct page *pages, |
| 378 | int count) |
| 379 | { |
| 380 | struct cma *cma = dev_get_cma_area(dev); |
| 381 | unsigned long pfn; |
| 382 | |
| 383 | if (!cma || !pages) |
| 384 | return false; |
| 385 | |
| 386 | pr_debug("%s(page %p)\n", __func__, (void *)pages); |
| 387 | |
| 388 | pfn = page_to_pfn(pages); |
| 389 | |
| 390 | if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count) |
| 391 | return false; |
| 392 | |
| 393 | VM_BUG_ON(pfn + count > cma->base_pfn + cma->count); |
| 394 | |
| 395 | mutex_lock(&cma_mutex); |
| 396 | bitmap_clear(cma->bitmap, pfn - cma->base_pfn, count); |
| 397 | free_contig_range(pfn, count); |
| 398 | mutex_unlock(&cma_mutex); |
| 399 | |
| 400 | return true; |
| 401 | } |