mm/cma.c

   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 #include <linux/highmem.h>
  36
  37 struct cma {
  38         unsigned long   base_pfn;
  39         unsigned long   count;
  40         unsigned long   *bitmap;
  41         unsigned int order_per_bit; /* Order of pages represented by one bit */
  42         struct mutex    lock;
  43 };
  44
  45 static struct cma cma_areas[MAX_CMA_AREAS];
  46 static unsigned cma_area_count;
  47 static DEFINE_MUTEX(cma_mutex);
  48
  49 phys_addr_t cma_get_base(struct cma *cma)
  50 {
  51         return PFN_PHYS(cma->base_pfn);
  52 }
  53
  54 unsigned long cma_get_size(struct cma *cma)
  55 {
  56         return cma->count << PAGE_SHIFT;
  57 }
  58
  59 static unsigned long cma_bitmap_aligned_mask(struct cma *cma, int align_order)
  60 {
  61         return (1UL << (align_order >> cma->order_per_bit)) - 1;
  62 }
  63
  64 static unsigned long cma_bitmap_maxno(struct cma *cma)
  65 {
  66         return cma->count >> cma->order_per_bit;
  67 }
  68
  69 static unsigned long cma_bitmap_pages_to_bits(struct cma *cma,
  70                                                 unsigned long pages)
  71 {
  72         return ALIGN(pages, 1UL << cma->order_per_bit) >> cma->order_per_bit;
  73 }
  74
  75 static void cma_clear_bitmap(struct cma *cma, unsigned long pfn, int count)
  76 {
  77         unsigned long bitmap_no, bitmap_count;
  78
  79         bitmap_no = (pfn - cma->base_pfn) >> cma->order_per_bit;
  80         bitmap_count = cma_bitmap_pages_to_bits(cma, count);
  81
  82         mutex_lock(&cma->lock);
  83         bitmap_clear(cma->bitmap, bitmap_no, bitmap_count);
  84         mutex_unlock(&cma->lock);
  85 }
  86
  87 static int __init cma_activate_area(struct cma *cma)
  88 {
  89         int bitmap_size = BITS_TO_LONGS(cma_bitmap_maxno(cma)) * sizeof(long);
  90         unsigned long base_pfn = cma->base_pfn, pfn = base_pfn;
  91         unsigned i = cma->count >> pageblock_order;
  92         struct zone *zone;
  93
  94         cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL);
  95
  96         if (!cma->bitmap)
  97                 return -ENOMEM;
  98
  99         WARN_ON_ONCE(!pfn_valid(pfn));
 100         zone = page_zone(pfn_to_page(pfn));
 101
 102         do {
 103                 unsigned j;
 104
 105                 base_pfn = pfn;
 106                 for (j = pageblock_nr_pages; j; --j, pfn++) {
 107                         WARN_ON_ONCE(!pfn_valid(pfn));
 108                         /*
 109                          * alloc_contig_range requires the pfn range
 110                          * specified to be in the same zone. Make this
 111                          * simple by forcing the entire CMA resv range
 112                          * to be in the same zone.
 113                          */
 114                         if (page_zone(pfn_to_page(pfn)) != zone)
 115                                 goto err;
 116                 }
 117                 init_cma_reserved_pageblock(pfn_to_page(base_pfn));
 118         } while (--i);
 119
 120         mutex_init(&cma->lock);
 121         return 0;
 122
 123 err:
 124         kfree(cma->bitmap);
 125         return -EINVAL;
 126 }
 127
 128 static int __init cma_init_reserved_areas(void)
 129 {
 130         int i;
 131
 132         for (i = 0; i < cma_area_count; i++) {
 133                 int ret = cma_activate_area(&cma_areas[i]);
 134
 135                 if (ret)
 136                         return ret;
 137         }
 138
 139         return 0;
 140 }
 141 core_initcall(cma_init_reserved_areas);
 142
 143 /**
 144  * cma_declare_contiguous() - reserve custom contiguous area
 145  * @base: Base address of the reserved area optional, use 0 for any
 146  * @size: Size of the reserved area (in bytes),
 147  * @limit: End address of the reserved memory (optional, 0 for any).
 148  * @alignment: Alignment for the CMA area, should be power of 2 or zero
 149  * @order_per_bit: Order of pages represented by one bit on bitmap.
 150  * @fixed: hint about where to place the reserved area
 151  * @res_cma: Pointer to store the created cma region.
 152  *
 153  * This function reserves memory from early allocator. It should be
 154  * called by arch specific code once the early allocator (memblock or bootmem)
 155  * has been activated and all other subsystems have already allocated/reserved
 156  * memory. This function allows to create custom reserved areas.
 157  *
 158  * If @fixed is true, reserve contiguous area at exactly @base.  If false,
 159  * reserve in range from @base to @limit.
 160  */
 161 int __init cma_declare_contiguous(phys_addr_t base,
 162                         phys_addr_t size, phys_addr_t limit,
 163                         phys_addr_t alignment, unsigned int order_per_bit,
 164                         bool fixed, struct cma **res_cma)
 165 {
 166         struct cma *cma;
 167         phys_addr_t memblock_end = memblock_end_of_DRAM();
 168         phys_addr_t highmem_start = __pa(high_memory);
 169         int ret = 0;
 170
 171         pr_debug("%s(size %lx, base %08lx, limit %08lx alignment %08lx)\n",
 172                 __func__, (unsigned long)size, (unsigned long)base,
 173                 (unsigned long)limit, (unsigned long)alignment);
 174
 175         if (cma_area_count == ARRAY_SIZE(cma_areas)) {
 176                 pr_err("Not enough slots for CMA reserved regions!\n");
 177                 return -ENOSPC;
 178         }
 179
 180         if (!size)
 181                 return -EINVAL;
 182
 183         if (alignment && !is_power_of_2(alignment))
 184                 return -EINVAL;
 185
 186         /*
 187          * Sanitise input arguments.
 188          * Pages both ends in CMA area could be merged into adjacent unmovable
 189          * migratetype page by page allocator's buddy algorithm. In the case,
 190          * you couldn't get a contiguous memory, which is not what we want.
 191          */
 192         alignment = max(alignment,
 193                 (phys_addr_t)PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order));
 194         base = ALIGN(base, alignment);
 195         size = ALIGN(size, alignment);
 196         limit &= ~(alignment - 1);
 197
 198         /* size should be aligned with order_per_bit */
 199         if (!IS_ALIGNED(size >> PAGE_SHIFT, 1 << order_per_bit))
 200                 return -EINVAL;
 201
 202         /*
 203          * adjust limit to avoid crossing low/high memory boundary for
 204          * automatically allocated regions
 205          */
 206         if (((limit == 0 || limit > memblock_end) &&
 207              (memblock_end - size < highmem_start &&
 208               memblock_end > highmem_start)) ||
 209             (!fixed && limit > highmem_start && limit - size < highmem_start)) {
 210                 limit = highmem_start;
 211         }
 212
 213         if (fixed && base < highmem_start && base+size > highmem_start) {
 214                 ret = -EINVAL;
 215                 pr_err("Region at %08lx defined on low/high memory boundary (%08lx)\n",
 216                         (unsigned long)base, (unsigned long)highmem_start);
 217                 goto err;
 218         }
 219
 220         /* Reserve memory */
 221         if (base && fixed) {
 222                 if (memblock_is_region_reserved(base, size) ||
 223                     memblock_reserve(base, size) < 0) {
 224                         ret = -EBUSY;
 225                         goto err;
 226                 }
 227         } else {
 228                 phys_addr_t addr = memblock_alloc_range(size, alignment, base,
 229                                                         limit);
 230                 if (!addr) {
 231                         ret = -ENOMEM;
 232                         goto err;
 233                 } else {
 234                         base = addr;
 235                 }
 236         }
 237
 238         /*
 239          * Each reserved area must be initialised later, when more kernel
 240          * subsystems (like slab allocator) are available.
 241          */
 242         cma = &cma_areas[cma_area_count];
 243         cma->base_pfn = PFN_DOWN(base);
 244         cma->count = size >> PAGE_SHIFT;
 245         cma->order_per_bit = order_per_bit;
 246         *res_cma = cma;
 247         cma_area_count++;
 248
 249         pr_info("Reserved %ld MiB at %08lx\n", (unsigned long)size / SZ_1M,
 250                 (unsigned long)base);
 251         return 0;
 252
 253 err:
 254         pr_err("Failed to reserve %ld MiB\n", (unsigned long)size / SZ_1M);
 255         return ret;
 256 }
 257
 258 /**
 259  * cma_alloc() - allocate pages from contiguous area
 260  * @cma:   Contiguous memory region for which the allocation is performed.
 261  * @count: Requested number of pages.
 262  * @align: Requested alignment of pages (in PAGE_SIZE order).
 263  *
 264  * This function allocates part of contiguous memory on specific
 265  * contiguous memory area.
 266  */
 267 struct page *cma_alloc(struct cma *cma, int count, unsigned int align)
 268 {
 269         unsigned long mask, pfn, start = 0;
 270         unsigned long bitmap_maxno, bitmap_no, bitmap_count;
 271         struct page *page = NULL;
 272         int ret;
 273
 274         if (!cma || !cma->count)
 275                 return NULL;
 276
 277         pr_debug("%s(cma %p, count %d, align %d)\n", __func__, (void *)cma,
 278                  count, align);
 279
 280         if (!count)
 281                 return NULL;
 282
 283         mask = cma_bitmap_aligned_mask(cma, align);
 284         bitmap_maxno = cma_bitmap_maxno(cma);
 285         bitmap_count = cma_bitmap_pages_to_bits(cma, count);
 286
 287         for (;;) {
 288                 mutex_lock(&cma->lock);
 289                 bitmap_no = bitmap_find_next_zero_area(cma->bitmap,
 290                                 bitmap_maxno, start, bitmap_count, mask);
 291                 if (bitmap_no >= bitmap_maxno) {
 292                         mutex_unlock(&cma->lock);
 293                         break;
 294                 }
 295                 bitmap_set(cma->bitmap, bitmap_no, bitmap_count);
 296                 /*
 297                  * It's safe to drop the lock here. We've marked this region for
 298                  * our exclusive use. If the migration fails we will take the
 299                  * lock again and unmark it.
 300                  */
 301                 mutex_unlock(&cma->lock);
 302
 303                 pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit);
 304                 mutex_lock(&cma_mutex);
 305                 ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA);
 306                 mutex_unlock(&cma_mutex);
 307                 if (ret == 0) {
 308                         page = pfn_to_page(pfn);
 309                         break;
 310                 }
 311
 312                 cma_clear_bitmap(cma, pfn, count);
 313                 if (ret != -EBUSY)
 314                         break;
 315
 316                 pr_debug("%s(): memory range at %p is busy, retrying\n",
 317                          __func__, pfn_to_page(pfn));
 318                 /* try again with a bit different memory target */
 319                 start = bitmap_no + mask + 1;
 320         }
 321
 322         pr_debug("%s(): returned %p\n", __func__, page);
 323         return page;
 324 }
 325
 326 /**
 327  * cma_release() - release allocated pages
 328  * @cma:   Contiguous memory region for which the allocation is performed.
 329  * @pages: Allocated pages.
 330  * @count: Number of allocated pages.
 331  *
 332  * This function releases memory allocated by alloc_cma().
 333  * It returns false when provided pages do not belong to contiguous area and
 334  * true otherwise.
 335  */
 336 bool cma_release(struct cma *cma, struct page *pages, int count)
 337 {
 338         unsigned long pfn;
 339
 340         if (!cma || !pages)
 341                 return false;
 342
 343         pr_debug("%s(page %p)\n", __func__, (void *)pages);
 344
 345         pfn = page_to_pfn(pages);
 346
 347         if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count)
 348                 return false;
 349
 350         VM_BUG_ON(pfn + count > cma->base_pfn + cma->count);
 351
 352         free_contig_range(pfn, count);
 353         cma_clear_bitmap(cma, pfn, count);
 354
 355         return true;
 356 }