Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Contiguous Memory Allocator |
| 3 | * |
| 4 | * Copyright (c) 2010-2011 by Samsung Electronics. |
| 5 | * Copyright IBM Corporation, 2013 |
| 6 | * Copyright LG Electronics Inc., 2014 |
| 7 | * Written by: |
| 8 | * Marek Szyprowski <m.szyprowski@samsung.com> |
| 9 | * Michal Nazarewicz <mina86@mina86.com> |
| 10 | * Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> |
| 11 | * Joonsoo Kim <iamjoonsoo.kim@lge.com> |
| 12 | * |
| 13 | * This program is free software; you can redistribute it and/or |
| 14 | * modify it under the terms of the GNU General Public License as |
| 15 | * published by the Free Software Foundation; either version 2 of the |
| 16 | * License or (at your optional) any later version of the license. |
| 17 | */ |
| 18 | |
| 19 | #define pr_fmt(fmt) "cma: " fmt |
| 20 | |
| 21 | #ifdef CONFIG_CMA_DEBUG |
| 22 | #ifndef DEBUG |
| 23 | # define DEBUG |
| 24 | #endif |
| 25 | #endif |
| 26 | |
| 27 | #include <linux/memblock.h> |
| 28 | #include <linux/err.h> |
| 29 | #include <linux/mm.h> |
| 30 | #include <linux/mutex.h> |
| 31 | #include <linux/sizes.h> |
| 32 | #include <linux/slab.h> |
| 33 | #include <linux/log2.h> |
| 34 | #include <linux/cma.h> |
| 35 | |
| 36 | struct cma { |
| 37 | unsigned long base_pfn; |
| 38 | unsigned long count; |
| 39 | unsigned long *bitmap; |
| 40 | unsigned int order_per_bit; /* Order of pages represented by one bit */ |
| 41 | struct mutex lock; |
| 42 | }; |
| 43 | |
| 44 | static struct cma cma_areas[MAX_CMA_AREAS]; |
| 45 | static unsigned cma_area_count; |
| 46 | static DEFINE_MUTEX(cma_mutex); |
| 47 | |
| 48 | phys_addr_t cma_get_base(struct cma *cma) |
| 49 | { |
| 50 | return PFN_PHYS(cma->base_pfn); |
| 51 | } |
| 52 | |
| 53 | unsigned long cma_get_size(struct cma *cma) |
| 54 | { |
| 55 | return cma->count << PAGE_SHIFT; |
| 56 | } |
| 57 | |
| 58 | static unsigned long cma_bitmap_aligned_mask(struct cma *cma, int align_order) |
| 59 | { |
| 60 | return (1UL << (align_order >> cma->order_per_bit)) - 1; |
| 61 | } |
| 62 | |
| 63 | static unsigned long cma_bitmap_maxno(struct cma *cma) |
| 64 | { |
| 65 | return cma->count >> cma->order_per_bit; |
| 66 | } |
| 67 | |
| 68 | static unsigned long cma_bitmap_pages_to_bits(struct cma *cma, |
| 69 | unsigned long pages) |
| 70 | { |
| 71 | return ALIGN(pages, 1UL << cma->order_per_bit) >> cma->order_per_bit; |
| 72 | } |
| 73 | |
| 74 | static void cma_clear_bitmap(struct cma *cma, unsigned long pfn, int count) |
| 75 | { |
| 76 | unsigned long bitmap_no, bitmap_count; |
| 77 | |
| 78 | bitmap_no = (pfn - cma->base_pfn) >> cma->order_per_bit; |
| 79 | bitmap_count = cma_bitmap_pages_to_bits(cma, count); |
| 80 | |
| 81 | mutex_lock(&cma->lock); |
| 82 | bitmap_clear(cma->bitmap, bitmap_no, bitmap_count); |
| 83 | mutex_unlock(&cma->lock); |
| 84 | } |
| 85 | |
| 86 | static int __init cma_activate_area(struct cma *cma) |
| 87 | { |
| 88 | int bitmap_size = BITS_TO_LONGS(cma_bitmap_maxno(cma)) * sizeof(long); |
| 89 | unsigned long base_pfn = cma->base_pfn, pfn = base_pfn; |
| 90 | unsigned i = cma->count >> pageblock_order; |
| 91 | struct zone *zone; |
| 92 | |
| 93 | cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL); |
| 94 | |
| 95 | if (!cma->bitmap) |
| 96 | return -ENOMEM; |
| 97 | |
| 98 | WARN_ON_ONCE(!pfn_valid(pfn)); |
| 99 | zone = page_zone(pfn_to_page(pfn)); |
| 100 | |
| 101 | do { |
| 102 | unsigned j; |
| 103 | |
| 104 | base_pfn = pfn; |
| 105 | for (j = pageblock_nr_pages; j; --j, pfn++) { |
| 106 | WARN_ON_ONCE(!pfn_valid(pfn)); |
| 107 | /* |
| 108 | * alloc_contig_range requires the pfn range |
| 109 | * specified to be in the same zone. Make this |
| 110 | * simple by forcing the entire CMA resv range |
| 111 | * to be in the same zone. |
| 112 | */ |
| 113 | if (page_zone(pfn_to_page(pfn)) != zone) |
| 114 | goto err; |
| 115 | } |
| 116 | init_cma_reserved_pageblock(pfn_to_page(base_pfn)); |
| 117 | } while (--i); |
| 118 | |
| 119 | mutex_init(&cma->lock); |
| 120 | return 0; |
| 121 | |
| 122 | err: |
| 123 | kfree(cma->bitmap); |
| 124 | return -EINVAL; |
| 125 | } |
| 126 | |
| 127 | static int __init cma_init_reserved_areas(void) |
| 128 | { |
| 129 | int i; |
| 130 | |
| 131 | for (i = 0; i < cma_area_count; i++) { |
| 132 | int ret = cma_activate_area(&cma_areas[i]); |
| 133 | |
| 134 | if (ret) |
| 135 | return ret; |
| 136 | } |
| 137 | |
| 138 | return 0; |
| 139 | } |
| 140 | core_initcall(cma_init_reserved_areas); |
| 141 | |
| 142 | /** |
| 143 | * cma_declare_contiguous() - reserve custom contiguous area |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 144 | * @base: Base address of the reserved area optional, use 0 for any |
Joonsoo Kim | c1f733aa | 2014-08-06 16:05:32 -0700 | [diff] [blame^] | 145 | * @size: Size of the reserved area (in bytes), |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 146 | * @limit: End address of the reserved memory (optional, 0 for any). |
| 147 | * @alignment: Alignment for the CMA area, should be power of 2 or zero |
| 148 | * @order_per_bit: Order of pages represented by one bit on bitmap. |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 149 | * @fixed: hint about where to place the reserved area |
Joonsoo Kim | c1f733aa | 2014-08-06 16:05:32 -0700 | [diff] [blame^] | 150 | * @res_cma: Pointer to store the created cma region. |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 151 | * |
| 152 | * This function reserves memory from early allocator. It should be |
| 153 | * called by arch specific code once the early allocator (memblock or bootmem) |
| 154 | * has been activated and all other subsystems have already allocated/reserved |
| 155 | * memory. This function allows to create custom reserved areas. |
| 156 | * |
| 157 | * If @fixed is true, reserve contiguous area at exactly @base. If false, |
| 158 | * reserve in range from @base to @limit. |
| 159 | */ |
Joonsoo Kim | c1f733aa | 2014-08-06 16:05:32 -0700 | [diff] [blame^] | 160 | int __init cma_declare_contiguous(phys_addr_t base, |
| 161 | phys_addr_t size, phys_addr_t limit, |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 162 | phys_addr_t alignment, unsigned int order_per_bit, |
Joonsoo Kim | c1f733aa | 2014-08-06 16:05:32 -0700 | [diff] [blame^] | 163 | bool fixed, struct cma **res_cma) |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 164 | { |
Joonsoo Kim | c1f733aa | 2014-08-06 16:05:32 -0700 | [diff] [blame^] | 165 | struct cma *cma; |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 166 | int ret = 0; |
| 167 | |
| 168 | pr_debug("%s(size %lx, base %08lx, limit %08lx alignment %08lx)\n", |
| 169 | __func__, (unsigned long)size, (unsigned long)base, |
| 170 | (unsigned long)limit, (unsigned long)alignment); |
| 171 | |
| 172 | if (cma_area_count == ARRAY_SIZE(cma_areas)) { |
| 173 | pr_err("Not enough slots for CMA reserved regions!\n"); |
| 174 | return -ENOSPC; |
| 175 | } |
| 176 | |
| 177 | if (!size) |
| 178 | return -EINVAL; |
| 179 | |
| 180 | if (alignment && !is_power_of_2(alignment)) |
| 181 | return -EINVAL; |
| 182 | |
| 183 | /* |
| 184 | * Sanitise input arguments. |
| 185 | * Pages both ends in CMA area could be merged into adjacent unmovable |
| 186 | * migratetype page by page allocator's buddy algorithm. In the case, |
| 187 | * you couldn't get a contiguous memory, which is not what we want. |
| 188 | */ |
| 189 | alignment = max(alignment, |
| 190 | (phys_addr_t)PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order)); |
| 191 | base = ALIGN(base, alignment); |
| 192 | size = ALIGN(size, alignment); |
| 193 | limit &= ~(alignment - 1); |
| 194 | |
| 195 | /* size should be aligned with order_per_bit */ |
| 196 | if (!IS_ALIGNED(size >> PAGE_SHIFT, 1 << order_per_bit)) |
| 197 | return -EINVAL; |
| 198 | |
| 199 | /* Reserve memory */ |
| 200 | if (base && fixed) { |
| 201 | if (memblock_is_region_reserved(base, size) || |
| 202 | memblock_reserve(base, size) < 0) { |
| 203 | ret = -EBUSY; |
| 204 | goto err; |
| 205 | } |
| 206 | } else { |
| 207 | phys_addr_t addr = memblock_alloc_range(size, alignment, base, |
| 208 | limit); |
| 209 | if (!addr) { |
| 210 | ret = -ENOMEM; |
| 211 | goto err; |
| 212 | } else { |
| 213 | base = addr; |
| 214 | } |
| 215 | } |
| 216 | |
| 217 | /* |
| 218 | * Each reserved area must be initialised later, when more kernel |
| 219 | * subsystems (like slab allocator) are available. |
| 220 | */ |
Joonsoo Kim | c1f733aa | 2014-08-06 16:05:32 -0700 | [diff] [blame^] | 221 | cma = &cma_areas[cma_area_count]; |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 222 | cma->base_pfn = PFN_DOWN(base); |
| 223 | cma->count = size >> PAGE_SHIFT; |
| 224 | cma->order_per_bit = order_per_bit; |
| 225 | *res_cma = cma; |
| 226 | cma_area_count++; |
| 227 | |
| 228 | pr_info("CMA: reserved %ld MiB at %08lx\n", (unsigned long)size / SZ_1M, |
| 229 | (unsigned long)base); |
| 230 | return 0; |
| 231 | |
| 232 | err: |
| 233 | pr_err("CMA: failed to reserve %ld MiB\n", (unsigned long)size / SZ_1M); |
| 234 | return ret; |
| 235 | } |
| 236 | |
| 237 | /** |
| 238 | * cma_alloc() - allocate pages from contiguous area |
| 239 | * @cma: Contiguous memory region for which the allocation is performed. |
| 240 | * @count: Requested number of pages. |
| 241 | * @align: Requested alignment of pages (in PAGE_SIZE order). |
| 242 | * |
| 243 | * This function allocates part of contiguous memory on specific |
| 244 | * contiguous memory area. |
| 245 | */ |
| 246 | struct page *cma_alloc(struct cma *cma, int count, unsigned int align) |
| 247 | { |
| 248 | unsigned long mask, pfn, start = 0; |
| 249 | unsigned long bitmap_maxno, bitmap_no, bitmap_count; |
| 250 | struct page *page = NULL; |
| 251 | int ret; |
| 252 | |
| 253 | if (!cma || !cma->count) |
| 254 | return NULL; |
| 255 | |
| 256 | pr_debug("%s(cma %p, count %d, align %d)\n", __func__, (void *)cma, |
| 257 | count, align); |
| 258 | |
| 259 | if (!count) |
| 260 | return NULL; |
| 261 | |
| 262 | mask = cma_bitmap_aligned_mask(cma, align); |
| 263 | bitmap_maxno = cma_bitmap_maxno(cma); |
| 264 | bitmap_count = cma_bitmap_pages_to_bits(cma, count); |
| 265 | |
| 266 | for (;;) { |
| 267 | mutex_lock(&cma->lock); |
| 268 | bitmap_no = bitmap_find_next_zero_area(cma->bitmap, |
| 269 | bitmap_maxno, start, bitmap_count, mask); |
| 270 | if (bitmap_no >= bitmap_maxno) { |
| 271 | mutex_unlock(&cma->lock); |
| 272 | break; |
| 273 | } |
| 274 | bitmap_set(cma->bitmap, bitmap_no, bitmap_count); |
| 275 | /* |
| 276 | * It's safe to drop the lock here. We've marked this region for |
| 277 | * our exclusive use. If the migration fails we will take the |
| 278 | * lock again and unmark it. |
| 279 | */ |
| 280 | mutex_unlock(&cma->lock); |
| 281 | |
| 282 | pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit); |
| 283 | mutex_lock(&cma_mutex); |
| 284 | ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA); |
| 285 | mutex_unlock(&cma_mutex); |
| 286 | if (ret == 0) { |
| 287 | page = pfn_to_page(pfn); |
| 288 | break; |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 289 | } |
Joonsoo Kim | b7155e7 | 2014-08-06 16:05:30 -0700 | [diff] [blame] | 290 | |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 291 | cma_clear_bitmap(cma, pfn, count); |
Joonsoo Kim | b7155e7 | 2014-08-06 16:05:30 -0700 | [diff] [blame] | 292 | if (ret != -EBUSY) |
| 293 | break; |
| 294 | |
Joonsoo Kim | a254129 | 2014-08-06 16:05:25 -0700 | [diff] [blame] | 295 | pr_debug("%s(): memory range at %p is busy, retrying\n", |
| 296 | __func__, pfn_to_page(pfn)); |
| 297 | /* try again with a bit different memory target */ |
| 298 | start = bitmap_no + mask + 1; |
| 299 | } |
| 300 | |
| 301 | pr_debug("%s(): returned %p\n", __func__, page); |
| 302 | return page; |
| 303 | } |
| 304 | |
| 305 | /** |
| 306 | * cma_release() - release allocated pages |
| 307 | * @cma: Contiguous memory region for which the allocation is performed. |
| 308 | * @pages: Allocated pages. |
| 309 | * @count: Number of allocated pages. |
| 310 | * |
| 311 | * This function releases memory allocated by alloc_cma(). |
| 312 | * It returns false when provided pages do not belong to contiguous area and |
| 313 | * true otherwise. |
| 314 | */ |
| 315 | bool cma_release(struct cma *cma, struct page *pages, int count) |
| 316 | { |
| 317 | unsigned long pfn; |
| 318 | |
| 319 | if (!cma || !pages) |
| 320 | return false; |
| 321 | |
| 322 | pr_debug("%s(page %p)\n", __func__, (void *)pages); |
| 323 | |
| 324 | pfn = page_to_pfn(pages); |
| 325 | |
| 326 | if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count) |
| 327 | return false; |
| 328 | |
| 329 | VM_BUG_ON(pfn + count > cma->base_pfn + cma->count); |
| 330 | |
| 331 | free_contig_range(pfn, count); |
| 332 | cma_clear_bitmap(cma, pfn, count); |
| 333 | |
| 334 | return true; |
| 335 | } |