drivers/base/dma-contiguous.c

   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/sizes.h>
  31 #include <linux/slab.h>
  32 #include <linux/swap.h>
  33 #include <linux/mm_types.h>
  34 #include <linux/dma-contiguous.h>
  35 #include <linux/log2.h>
  36
  37 struct cma {
  38         unsigned long   base_pfn;
  39         unsigned long   count;
  40         unsigned long   *bitmap;
  41         struct mutex    lock;
  42 };
  43
  44 struct cma *dma_contiguous_default_area;
  45
  46 #ifdef CONFIG_CMA_SIZE_MBYTES
  47 #define CMA_SIZE_MBYTES CONFIG_CMA_SIZE_MBYTES
  48 #else
  49 #define CMA_SIZE_MBYTES 0
  50 #endif
  51
  52 /*
  53  * Default global CMA area size can be defined in kernel's .config.
  54  * This is useful mainly for distro maintainers to create a kernel
  55  * that works correctly for most supported systems.
  56  * The size can be set in bytes or as a percentage of the total memory
  57  * in the system.
  58  *
  59  * Users, who want to set the size of global CMA area for their system
  60  * should use cma= kernel parameter.
  61  */
  62 static const phys_addr_t size_bytes = CMA_SIZE_MBYTES * SZ_1M;
  63 static phys_addr_t size_cmdline = -1;
  64 static phys_addr_t base_cmdline;
  65 static phys_addr_t limit_cmdline;
  66
  67 static int __init early_cma(char *p)
  68 {
  69         pr_debug("%s(%s)\n", __func__, p);
  70         size_cmdline = memparse(p, &p);
  71         if (*p != '@')
  72                 return 0;
  73         base_cmdline = memparse(p + 1, &p);
  74         if (*p != '-') {
  75                 limit_cmdline = base_cmdline + size_cmdline;
  76                 return 0;
  77         }
  78         limit_cmdline = memparse(p + 1, &p);
  79
  80         return 0;
  81 }
  82 early_param("cma", early_cma);
  83
  84 #ifdef CONFIG_CMA_SIZE_PERCENTAGE
  85
  86 static phys_addr_t __init __maybe_unused cma_early_percent_memory(void)
  87 {
  88         struct memblock_region *reg;
  89         unsigned long total_pages = 0;
  90
  91         /*
  92          * We cannot use memblock_phys_mem_size() here, because
  93          * memblock_analyze() has not been called yet.
  94          */
  95         for_each_memblock(memory, reg)
  96                 total_pages += memblock_region_memory_end_pfn(reg) -
  97                                memblock_region_memory_base_pfn(reg);
  98
  99         return (total_pages * CONFIG_CMA_SIZE_PERCENTAGE / 100) << PAGE_SHIFT;
 100 }
 101
 102 #else
 103
 104 static inline __maybe_unused phys_addr_t cma_early_percent_memory(void)
 105 {
 106         return 0;
 107 }
 108
 109 #endif
 110
 111 /**
 112  * dma_contiguous_reserve() - reserve area(s) for contiguous memory handling
 113  * @limit: End address of the reserved memory (optional, 0 for any).
 114  *
 115  * This function reserves memory from early allocator. It should be
 116  * called by arch specific code once the early allocator (memblock or bootmem)
 117  * has been activated and all other subsystems have already allocated/reserved
 118  * memory.
 119  */
 120 void __init dma_contiguous_reserve(phys_addr_t limit)
 121 {
 122         phys_addr_t selected_size = 0;
 123         phys_addr_t selected_base = 0;
 124         phys_addr_t selected_limit = limit;
 125         bool fixed = false;
 126
 127         pr_debug("%s(limit %08lx)\n", __func__, (unsigned long)limit);
 128
 129         if (size_cmdline != -1) {
 130                 selected_size = size_cmdline;
 131                 selected_base = base_cmdline;
 132                 selected_limit = min_not_zero(limit_cmdline, limit);
 133                 if (base_cmdline + size_cmdline == limit_cmdline)
 134                         fixed = true;
 135         } else {
 136 #ifdef CONFIG_CMA_SIZE_SEL_MBYTES
 137                 selected_size = size_bytes;
 138 #elif defined(CONFIG_CMA_SIZE_SEL_PERCENTAGE)
 139                 selected_size = cma_early_percent_memory();
 140 #elif defined(CONFIG_CMA_SIZE_SEL_MIN)
 141                 selected_size = min(size_bytes, cma_early_percent_memory());
 142 #elif defined(CONFIG_CMA_SIZE_SEL_MAX)
 143                 selected_size = max(size_bytes, cma_early_percent_memory());
 144 #endif
 145         }
 146
 147         if (selected_size && !dma_contiguous_default_area) {
 148                 pr_debug("%s: reserving %ld MiB for global area\n", __func__,
 149                          (unsigned long)selected_size / SZ_1M);
 150
 151                 dma_contiguous_reserve_area(selected_size, selected_base,
 152                                             selected_limit,
 153                                             &dma_contiguous_default_area,
 154                                             fixed);
 155         }
 156 }
 157
 158 static DEFINE_MUTEX(cma_mutex);
 159
 160 static int __init cma_activate_area(struct cma *cma)
 161 {
 162         int bitmap_size = BITS_TO_LONGS(cma->count) * sizeof(long);
 163         unsigned long base_pfn = cma->base_pfn, pfn = base_pfn;
 164         unsigned i = cma->count >> pageblock_order;
 165         struct zone *zone;
 166
 167         cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL);
 168
 169         if (!cma->bitmap)
 170                 return -ENOMEM;
 171
 172         WARN_ON_ONCE(!pfn_valid(pfn));
 173         zone = page_zone(pfn_to_page(pfn));
 174
 175         do {
 176                 unsigned j;
 177                 base_pfn = pfn;
 178                 for (j = pageblock_nr_pages; j; --j, pfn++) {
 179                         WARN_ON_ONCE(!pfn_valid(pfn));
 180                         /*
 181                          * alloc_contig_range requires the pfn range
 182                          * specified to be in the same zone. Make this
 183                          * simple by forcing the entire CMA resv range
 184                          * to be in the same zone.
 185                          */
 186                         if (page_zone(pfn_to_page(pfn)) != zone)
 187                                 goto err;
 188                 }
 189                 init_cma_reserved_pageblock(pfn_to_page(base_pfn));
 190         } while (--i);
 191
 192         mutex_init(&cma->lock);
 193         return 0;
 194
 195 err:
 196         kfree(cma->bitmap);
 197         return -EINVAL;
 198 }
 199
 200 static struct cma cma_areas[MAX_CMA_AREAS];
 201 static unsigned cma_area_count;
 202
 203 static int __init cma_init_reserved_areas(void)
 204 {
 205         int i;
 206
 207         for (i = 0; i < cma_area_count; i++) {
 208                 int ret = cma_activate_area(&cma_areas[i]);
 209                 if (ret)
 210                         return ret;
 211         }
 212
 213         return 0;
 214 }
 215 core_initcall(cma_init_reserved_areas);
 216
 217 static int __init __dma_contiguous_reserve_area(phys_addr_t size,
 218                                 phys_addr_t base, phys_addr_t limit,
 219                                 phys_addr_t alignment,
 220                                 struct cma **res_cma, bool fixed)
 221 {
 222         struct cma *cma = &cma_areas[cma_area_count];
 223         int ret = 0;
 224
 225         pr_debug("%s(size %lx, base %08lx, limit %08lx alignment %08lx)\n",
 226                 __func__, (unsigned long)size, (unsigned long)base,
 227                 (unsigned long)limit, (unsigned long)alignment);
 228
 229         if (cma_area_count == ARRAY_SIZE(cma_areas)) {
 230                 pr_err("Not enough slots for CMA reserved regions!\n");
 231                 return -ENOSPC;
 232         }
 233
 234         if (!size)
 235                 return -EINVAL;
 236
 237         if (alignment && !is_power_of_2(alignment))
 238                 return -EINVAL;
 239
 240         /*
 241          * Sanitise input arguments.
 242          * Pages both ends in CMA area could be merged into adjacent unmovable
 243          * migratetype page by page allocator's buddy algorithm. In the case,
 244          * you couldn't get a contiguous memory, which is not what we want.
 245          */
 246         alignment = max(alignment,
 247                 (phys_addr_t)PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order));
 248         base = ALIGN(base, alignment);
 249         size = ALIGN(size, alignment);
 250         limit &= ~(alignment - 1);
 251
 252         /* Reserve memory */
 253         if (base && fixed) {
 254                 if (memblock_is_region_reserved(base, size) ||
 255                     memblock_reserve(base, size) < 0) {
 256                         ret = -EBUSY;
 257                         goto err;
 258                 }
 259         } else {
 260                 phys_addr_t addr = memblock_alloc_range(size, alignment, base,
 261                                                         limit);
 262                 if (!addr) {
 263                         ret = -ENOMEM;
 264                         goto err;
 265                 } else {
 266                         base = addr;
 267                 }
 268         }
 269
 270         /*
 271          * Each reserved area must be initialised later, when more kernel
 272          * subsystems (like slab allocator) are available.
 273          */
 274         cma->base_pfn = PFN_DOWN(base);
 275         cma->count = size >> PAGE_SHIFT;
 276         *res_cma = cma;
 277         cma_area_count++;
 278
 279         pr_info("CMA: reserved %ld MiB at %08lx\n", (unsigned long)size / SZ_1M,
 280                 (unsigned long)base);
 281         return 0;
 282
 283 err:
 284         pr_err("CMA: failed to reserve %ld MiB\n", (unsigned long)size / SZ_1M);
 285         return ret;
 286 }
 287
 288 /**
 289  * dma_contiguous_reserve_area() - reserve custom contiguous area
 290  * @size: Size of the reserved area (in bytes),
 291  * @base: Base address of the reserved area optional, use 0 for any
 292  * @limit: End address of the reserved memory (optional, 0 for any).
 293  * @res_cma: Pointer to store the created cma region.
 294  * @fixed: hint about where to place the reserved area
 295  *
 296  * This function reserves memory from early allocator. It should be
 297  * called by arch specific code once the early allocator (memblock or bootmem)
 298  * has been activated and all other subsystems have already allocated/reserved
 299  * memory. This function allows to create custom reserved areas for specific
 300  * devices.
 301  *
 302  * If @fixed is true, reserve contiguous area at exactly @base.  If false,
 303  * reserve in range from @base to @limit.
 304  */
 305 int __init dma_contiguous_reserve_area(phys_addr_t size, phys_addr_t base,
 306                                        phys_addr_t limit, struct cma **res_cma,
 307                                        bool fixed)
 308 {
 309         int ret;
 310
 311         ret = __dma_contiguous_reserve_area(size, base, limit, 0,
 312                                                 res_cma, fixed);
 313         if (ret)
 314                 return ret;
 315
 316         /* Architecture specific contiguous memory fixup. */
 317         dma_contiguous_early_fixup(PFN_PHYS((*res_cma)->base_pfn),
 318                                 (*res_cma)->count << PAGE_SHIFT);
 319
 320         return 0;
 321 }
 322
 323 static void clear_cma_bitmap(struct cma *cma, unsigned long pfn, int count)
 324 {
 325         mutex_lock(&cma->lock);
 326         bitmap_clear(cma->bitmap, pfn - cma->base_pfn, count);
 327         mutex_unlock(&cma->lock);
 328 }
 329
 330 static struct page *__dma_alloc_from_contiguous(struct cma *cma, int count,
 331                                        unsigned int align)
 332 {
 333         unsigned long mask, pfn, pageno, start = 0;
 334         struct page *page = NULL;
 335         int ret;
 336
 337         if (!cma || !cma->count)
 338                 return NULL;
 339
 340         pr_debug("%s(cma %p, count %d, align %d)\n", __func__, (void *)cma,
 341                  count, align);
 342
 343         if (!count)
 344                 return NULL;
 345
 346         mask = (1 << align) - 1;
 347
 348
 349         for (;;) {
 350                 mutex_lock(&cma->lock);
 351                 pageno = bitmap_find_next_zero_area(cma->bitmap, cma->count,
 352                                                     start, count, mask);
 353                 if (pageno >= cma->count) {
 354                         mutex_unlock(&cma->lock);
 355                         break;
 356                 }
 357                 bitmap_set(cma->bitmap, pageno, count);
 358                 /*
 359                  * It's safe to drop the lock here. We've marked this region for
 360                  * our exclusive use. If the migration fails we will take the
 361                  * lock again and unmark it.
 362                  */
 363                 mutex_unlock(&cma->lock);
 364
 365                 pfn = cma->base_pfn + pageno;
 366                 mutex_lock(&cma_mutex);
 367                 ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA);
 368                 mutex_unlock(&cma_mutex);
 369                 if (ret == 0) {
 370                         page = pfn_to_page(pfn);
 371                         break;
 372                 } else if (ret != -EBUSY) {
 373                         clear_cma_bitmap(cma, pfn, count);
 374                         break;
 375                 }
 376                 clear_cma_bitmap(cma, pfn, count);
 377                 pr_debug("%s(): memory range at %p is busy, retrying\n",
 378                          __func__, pfn_to_page(pfn));
 379                 /* try again with a bit different memory target */
 380                 start = pageno + mask + 1;
 381         }
 382
 383         pr_debug("%s(): returned %p\n", __func__, page);
 384         return page;
 385 }
 386
 387 /**
 388  * dma_alloc_from_contiguous() - allocate pages from contiguous area
 389  * @dev:   Pointer to device for which the allocation is performed.
 390  * @count: Requested number of pages.
 391  * @align: Requested alignment of pages (in PAGE_SIZE order).
 392  *
 393  * This function allocates memory buffer for specified device. It uses
 394  * device specific contiguous memory area if available or the default
 395  * global one. Requires architecture specific dev_get_cma_area() helper
 396  * function.
 397  */
 398 struct page *dma_alloc_from_contiguous(struct device *dev, int count,
 399                                        unsigned int align)
 400 {
 401         struct cma *cma = dev_get_cma_area(dev);
 402
 403         if (align > CONFIG_CMA_ALIGNMENT)
 404                 align = CONFIG_CMA_ALIGNMENT;
 405
 406         return __dma_alloc_from_contiguous(cma, count, align);
 407 }
 408
 409 static bool __dma_release_from_contiguous(struct cma *cma, struct page *pages,
 410                                  int count)
 411 {
 412         unsigned long pfn;
 413
 414         if (!cma || !pages)
 415                 return false;
 416
 417         pr_debug("%s(page %p)\n", __func__, (void *)pages);
 418
 419         pfn = page_to_pfn(pages);
 420
 421         if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count)
 422                 return false;
 423
 424         VM_BUG_ON(pfn + count > cma->base_pfn + cma->count);
 425
 426         free_contig_range(pfn, count);
 427         clear_cma_bitmap(cma, pfn, count);
 428
 429         return true;
 430 }
 431
 432 /**
 433  * dma_release_from_contiguous() - release allocated pages
 434  * @dev:   Pointer to device for which the pages were allocated.
 435  * @pages: Allocated pages.
 436  * @count: Number of allocated pages.
 437  *
 438  * This function releases memory allocated by dma_alloc_from_contiguous().
 439  * It returns false when provided pages do not belong to contiguous area and
 440  * true otherwise.
 441  */
 442 bool dma_release_from_contiguous(struct device *dev, struct page *pages,
 443                                  int count)
 444 {
 445         struct cma *cma = dev_get_cma_area(dev);
 446
 447         return __dma_release_from_contiguous(cma, pages, count);
 448 }