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