2 * Contiguous Memory Allocator
4 * Copyright (c) 2010-2011 by Samsung Electronics.
5 * Copyright IBM Corporation, 2013
6 * Copyright LG Electronics Inc., 2014
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>
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.
19 #define pr_fmt(fmt) "cma: " fmt
21 #ifdef CONFIG_CMA_DEBUG
26 #define CREATE_TRACE_POINTS
28 #include <linux/memblock.h>
29 #include <linux/err.h>
31 #include <linux/mutex.h>
32 #include <linux/sizes.h>
33 #include <linux/slab.h>
34 #include <linux/log2.h>
35 #include <linux/cma.h>
36 #include <linux/highmem.h>
38 #include <linux/delay.h>
39 #include <trace/events/cma.h>
43 struct cma cma_areas[MAX_CMA_AREAS];
44 unsigned cma_area_count;
45 static DEFINE_MUTEX(cma_mutex);
47 phys_addr_t cma_get_base(const struct cma *cma)
49 return PFN_PHYS(cma->base_pfn);
52 unsigned long cma_get_size(const struct cma *cma)
54 return cma->count << PAGE_SHIFT;
57 static unsigned long cma_bitmap_aligned_mask(const struct cma *cma,
58 unsigned int align_order)
60 if (align_order <= cma->order_per_bit)
62 return (1UL << (align_order - cma->order_per_bit)) - 1;
66 * Find the offset of the base PFN from the specified align_order.
67 * The value returned is represented in order_per_bits.
69 static unsigned long cma_bitmap_aligned_offset(const struct cma *cma,
70 unsigned int align_order)
72 return (cma->base_pfn & ((1UL << align_order) - 1))
73 >> cma->order_per_bit;
76 static unsigned long cma_bitmap_pages_to_bits(const struct cma *cma,
79 return ALIGN(pages, 1UL << cma->order_per_bit) >> cma->order_per_bit;
82 static void cma_clear_bitmap(struct cma *cma, unsigned long pfn,
85 unsigned long bitmap_no, bitmap_count;
87 bitmap_no = (pfn - cma->base_pfn) >> cma->order_per_bit;
88 bitmap_count = cma_bitmap_pages_to_bits(cma, count);
90 mutex_lock(&cma->lock);
91 bitmap_clear(cma->bitmap, bitmap_no, bitmap_count);
92 mutex_unlock(&cma->lock);
95 static int __init cma_activate_area(struct cma *cma)
97 int bitmap_size = BITS_TO_LONGS(cma_bitmap_maxno(cma)) * sizeof(long);
98 unsigned long base_pfn = cma->base_pfn, pfn = base_pfn;
99 unsigned i = cma->count >> pageblock_order;
102 cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL);
107 WARN_ON_ONCE(!pfn_valid(pfn));
108 zone = page_zone(pfn_to_page(pfn));
114 for (j = pageblock_nr_pages; j; --j, pfn++) {
115 WARN_ON_ONCE(!pfn_valid(pfn));
117 * alloc_contig_range requires the pfn range
118 * specified to be in the same zone. Make this
119 * simple by forcing the entire CMA resv range
120 * to be in the same zone.
122 if (page_zone(pfn_to_page(pfn)) != zone)
125 init_cma_reserved_pageblock(pfn_to_page(base_pfn));
128 mutex_init(&cma->lock);
130 #ifdef CONFIG_CMA_DEBUGFS
131 INIT_HLIST_HEAD(&cma->mem_head);
132 spin_lock_init(&cma->mem_head_lock);
135 if (!PageHighMem(pfn_to_page(cma->base_pfn)))
136 kmemleak_free_part(__va(cma->base_pfn << PAGE_SHIFT),
137 cma->count << PAGE_SHIFT);
147 static int __init cma_init_reserved_areas(void)
151 for (i = 0; i < cma_area_count; i++) {
152 int ret = cma_activate_area(&cma_areas[i]);
160 core_initcall(cma_init_reserved_areas);
163 * cma_init_reserved_mem() - create custom contiguous area from reserved memory
164 * @base: Base address of the reserved area
165 * @size: Size of the reserved area (in bytes),
166 * @order_per_bit: Order of pages represented by one bit on bitmap.
167 * @res_cma: Pointer to store the created cma region.
169 * This function creates custom contiguous area from already reserved memory.
171 int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
172 unsigned int order_per_bit,
173 struct cma **res_cma)
176 phys_addr_t alignment;
179 if (cma_area_count == ARRAY_SIZE(cma_areas)) {
180 pr_err("Not enough slots for CMA reserved regions!\n");
184 if (!size || !memblock_is_region_reserved(base, size))
187 /* ensure minimal alignment required by mm core */
188 alignment = PAGE_SIZE <<
189 max_t(unsigned long, MAX_ORDER - 1, pageblock_order);
191 /* alignment should be aligned with order_per_bit */
192 if (!IS_ALIGNED(alignment >> PAGE_SHIFT, 1 << order_per_bit))
195 if (ALIGN(base, alignment) != base || ALIGN(size, alignment) != size)
199 * Each reserved area must be initialised later, when more kernel
200 * subsystems (like slab allocator) are available.
202 cma = &cma_areas[cma_area_count];
203 cma->base_pfn = PFN_DOWN(base);
204 cma->count = size >> PAGE_SHIFT;
205 cma->order_per_bit = order_per_bit;
208 totalcma_pages += (size / PAGE_SIZE);
214 * cma_declare_contiguous() - reserve custom contiguous area
215 * @base: Base address of the reserved area optional, use 0 for any
216 * @size: Size of the reserved area (in bytes),
217 * @limit: End address of the reserved memory (optional, 0 for any).
218 * @alignment: Alignment for the CMA area, should be power of 2 or zero
219 * @order_per_bit: Order of pages represented by one bit on bitmap.
220 * @fixed: hint about where to place the reserved area
221 * @res_cma: Pointer to store the created cma region.
223 * This function reserves memory from early allocator. It should be
224 * called by arch specific code once the early allocator (memblock or bootmem)
225 * has been activated and all other subsystems have already allocated/reserved
226 * memory. This function allows to create custom reserved areas.
228 * If @fixed is true, reserve contiguous area at exactly @base. If false,
229 * reserve in range from @base to @limit.
231 int __init cma_declare_contiguous(phys_addr_t base,
232 phys_addr_t size, phys_addr_t limit,
233 phys_addr_t alignment, unsigned int order_per_bit,
234 bool fixed, struct cma **res_cma)
236 phys_addr_t memblock_end = memblock_end_of_DRAM();
237 phys_addr_t highmem_start;
242 * high_memory isn't direct mapped memory so retrieving its physical
243 * address isn't appropriate. But it would be useful to check the
244 * physical address of the highmem boundary so it's justifiable to get
245 * the physical address from it. On x86 there is a validation check for
246 * this case, so the following workaround is needed to avoid it.
248 highmem_start = __pa_nodebug(high_memory);
250 highmem_start = __pa(high_memory);
252 pr_debug("%s(size %pa, base %pa, limit %pa alignment %pa)\n",
253 __func__, &size, &base, &limit, &alignment);
255 if (cma_area_count == ARRAY_SIZE(cma_areas)) {
256 pr_err("Not enough slots for CMA reserved regions!\n");
263 if (alignment && !is_power_of_2(alignment))
267 * Sanitise input arguments.
268 * Pages both ends in CMA area could be merged into adjacent unmovable
269 * migratetype page by page allocator's buddy algorithm. In the case,
270 * you couldn't get a contiguous memory, which is not what we want.
272 alignment = max(alignment, (phys_addr_t)PAGE_SIZE <<
273 max_t(unsigned long, MAX_ORDER - 1, pageblock_order));
274 base = ALIGN(base, alignment);
275 size = ALIGN(size, alignment);
276 limit &= ~(alignment - 1);
281 /* size should be aligned with order_per_bit */
282 if (!IS_ALIGNED(size >> PAGE_SHIFT, 1 << order_per_bit))
286 * If allocating at a fixed base the request region must not cross the
287 * low/high memory boundary.
289 if (fixed && base < highmem_start && base + size > highmem_start) {
291 pr_err("Region at %pa defined on low/high memory boundary (%pa)\n",
292 &base, &highmem_start);
297 * If the limit is unspecified or above the memblock end, its effective
298 * value will be the memblock end. Set it explicitly to simplify further
301 if (limit == 0 || limit > memblock_end)
302 limit = memblock_end;
306 if (memblock_is_region_reserved(base, size) ||
307 memblock_reserve(base, size) < 0) {
312 phys_addr_t addr = 0;
315 * All pages in the reserved area must come from the same zone.
316 * If the requested region crosses the low/high memory boundary,
317 * try allocating from high memory first and fall back to low
318 * memory in case of failure.
320 if (base < highmem_start && limit > highmem_start) {
321 addr = memblock_alloc_range(size, alignment,
322 highmem_start, limit,
324 limit = highmem_start;
328 addr = memblock_alloc_range(size, alignment, base,
338 * kmemleak scans/reads tracked objects for pointers to other
339 * objects but this address isn't mapped and accessible
341 kmemleak_ignore(phys_to_virt(addr));
345 ret = cma_init_reserved_mem(base, size, order_per_bit, res_cma);
349 pr_info("Reserved %ld MiB at %pa\n", (unsigned long)size / SZ_1M,
354 pr_err("Failed to reserve %ld MiB\n", (unsigned long)size / SZ_1M);
359 * cma_alloc() - allocate pages from contiguous area
360 * @cma: Contiguous memory region for which the allocation is performed.
361 * @count: Requested number of pages.
362 * @align: Requested alignment of pages (in PAGE_SIZE order).
364 * This function allocates part of contiguous memory on specific
365 * contiguous memory area.
367 struct page *cma_alloc(struct cma *cma, size_t count, unsigned int align)
369 unsigned long mask, offset;
370 unsigned long pfn = -1;
371 unsigned long start = 0;
372 unsigned long bitmap_maxno, bitmap_no, bitmap_count;
373 struct page *page = NULL;
375 int retry_after_sleep = 0;
377 if (!cma || !cma->count)
380 pr_debug("%s(cma %p, count %zu, align %d)\n", __func__, (void *)cma,
386 trace_cma_alloc_start(count, align);
388 mask = cma_bitmap_aligned_mask(cma, align);
389 offset = cma_bitmap_aligned_offset(cma, align);
390 bitmap_maxno = cma_bitmap_maxno(cma);
391 bitmap_count = cma_bitmap_pages_to_bits(cma, count);
393 if (bitmap_count > bitmap_maxno)
397 mutex_lock(&cma->lock);
398 bitmap_no = bitmap_find_next_zero_area_off(cma->bitmap,
399 bitmap_maxno, start, bitmap_count, mask,
401 if (bitmap_no >= bitmap_maxno) {
402 if (retry_after_sleep < 2) {
405 * Page may be momentarily pinned by some other
406 * process which has been scheduled out, eg.
407 * in exit path, during unmap call, or process
408 * fork and so cannot be freed there. Sleep
409 * for 100ms and retry twice to see if it has
412 mutex_unlock(&cma->lock);
417 mutex_unlock(&cma->lock);
421 bitmap_set(cma->bitmap, bitmap_no, bitmap_count);
423 * It's safe to drop the lock here. We've marked this region for
424 * our exclusive use. If the migration fails we will take the
425 * lock again and unmark it.
427 mutex_unlock(&cma->lock);
429 pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit);
430 mutex_lock(&cma_mutex);
431 ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA);
432 mutex_unlock(&cma_mutex);
434 page = pfn_to_page(pfn);
438 cma_clear_bitmap(cma, pfn, count);
442 pr_debug("%s(): memory range at %p is busy, retrying\n",
443 __func__, pfn_to_page(pfn));
445 trace_cma_alloc_busy_retry(pfn, pfn_to_page(pfn), count, align);
446 /* try again with a bit different memory target */
447 start = bitmap_no + mask + 1;
450 trace_cma_alloc(pfn, page, count, align);
452 pr_debug("%s(): returned %p\n", __func__, page);
457 * cma_release() - release allocated pages
458 * @cma: Contiguous memory region for which the allocation is performed.
459 * @pages: Allocated pages.
460 * @count: Number of allocated pages.
462 * This function releases memory allocated by alloc_cma().
463 * It returns false when provided pages do not belong to contiguous area and
466 bool cma_release(struct cma *cma, const struct page *pages, unsigned int count)
473 pr_debug("%s(page %p)\n", __func__, (void *)pages);
475 pfn = page_to_pfn(pages);
477 if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count)
480 VM_BUG_ON(pfn + count > cma->base_pfn + cma->count);
482 free_contig_range(pfn, count);
483 cma_clear_bitmap(cma, pfn, count);
484 trace_cma_release(pfn, pages, count);