]>
Commit | Line | Data |
---|---|---|
a254129e JK |
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 <[email protected]> | |
9 | * Michal Nazarewicz <[email protected]> | |
10 | * Aneesh Kumar K.V <[email protected]> | |
11 | * Joonsoo Kim <[email protected]> | |
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> | |
f7426b98 | 35 | #include <linux/highmem.h> |
a254129e JK |
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 | { | |
68faed63 WY |
61 | if (align_order <= cma->order_per_bit) |
62 | return 0; | |
63 | return (1UL << (align_order - cma->order_per_bit)) - 1; | |
a254129e JK |
64 | } |
65 | ||
66 | static unsigned long cma_bitmap_maxno(struct cma *cma) | |
67 | { | |
68 | return cma->count >> cma->order_per_bit; | |
69 | } | |
70 | ||
71 | static unsigned long cma_bitmap_pages_to_bits(struct cma *cma, | |
72 | unsigned long pages) | |
73 | { | |
74 | return ALIGN(pages, 1UL << cma->order_per_bit) >> cma->order_per_bit; | |
75 | } | |
76 | ||
77 | static void cma_clear_bitmap(struct cma *cma, unsigned long pfn, int count) | |
78 | { | |
79 | unsigned long bitmap_no, bitmap_count; | |
80 | ||
81 | bitmap_no = (pfn - cma->base_pfn) >> cma->order_per_bit; | |
82 | bitmap_count = cma_bitmap_pages_to_bits(cma, count); | |
83 | ||
84 | mutex_lock(&cma->lock); | |
85 | bitmap_clear(cma->bitmap, bitmap_no, bitmap_count); | |
86 | mutex_unlock(&cma->lock); | |
87 | } | |
88 | ||
89 | static int __init cma_activate_area(struct cma *cma) | |
90 | { | |
91 | int bitmap_size = BITS_TO_LONGS(cma_bitmap_maxno(cma)) * sizeof(long); | |
92 | unsigned long base_pfn = cma->base_pfn, pfn = base_pfn; | |
93 | unsigned i = cma->count >> pageblock_order; | |
94 | struct zone *zone; | |
95 | ||
96 | cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL); | |
97 | ||
98 | if (!cma->bitmap) | |
99 | return -ENOMEM; | |
100 | ||
101 | WARN_ON_ONCE(!pfn_valid(pfn)); | |
102 | zone = page_zone(pfn_to_page(pfn)); | |
103 | ||
104 | do { | |
105 | unsigned j; | |
106 | ||
107 | base_pfn = pfn; | |
108 | for (j = pageblock_nr_pages; j; --j, pfn++) { | |
109 | WARN_ON_ONCE(!pfn_valid(pfn)); | |
110 | /* | |
111 | * alloc_contig_range requires the pfn range | |
112 | * specified to be in the same zone. Make this | |
113 | * simple by forcing the entire CMA resv range | |
114 | * to be in the same zone. | |
115 | */ | |
116 | if (page_zone(pfn_to_page(pfn)) != zone) | |
117 | goto err; | |
118 | } | |
119 | init_cma_reserved_pageblock(pfn_to_page(base_pfn)); | |
120 | } while (--i); | |
121 | ||
122 | mutex_init(&cma->lock); | |
123 | return 0; | |
124 | ||
125 | err: | |
126 | kfree(cma->bitmap); | |
f022d8cb | 127 | cma->count = 0; |
a254129e JK |
128 | return -EINVAL; |
129 | } | |
130 | ||
131 | static int __init cma_init_reserved_areas(void) | |
132 | { | |
133 | int i; | |
134 | ||
135 | for (i = 0; i < cma_area_count; i++) { | |
136 | int ret = cma_activate_area(&cma_areas[i]); | |
137 | ||
138 | if (ret) | |
139 | return ret; | |
140 | } | |
141 | ||
142 | return 0; | |
143 | } | |
144 | core_initcall(cma_init_reserved_areas); | |
145 | ||
de9e14ee MS |
146 | /** |
147 | * cma_init_reserved_mem() - create custom contiguous area from reserved memory | |
148 | * @base: Base address of the reserved area | |
149 | * @size: Size of the reserved area (in bytes), | |
150 | * @order_per_bit: Order of pages represented by one bit on bitmap. | |
151 | * @res_cma: Pointer to store the created cma region. | |
152 | * | |
153 | * This function creates custom contiguous area from already reserved memory. | |
154 | */ | |
155 | int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size, | |
156 | int order_per_bit, struct cma **res_cma) | |
157 | { | |
158 | struct cma *cma; | |
159 | phys_addr_t alignment; | |
160 | ||
161 | /* Sanity checks */ | |
162 | if (cma_area_count == ARRAY_SIZE(cma_areas)) { | |
163 | pr_err("Not enough slots for CMA reserved regions!\n"); | |
164 | return -ENOSPC; | |
165 | } | |
166 | ||
167 | if (!size || !memblock_is_region_reserved(base, size)) | |
168 | return -EINVAL; | |
169 | ||
170 | /* ensure minimal alignment requied by mm core */ | |
171 | alignment = PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order); | |
172 | ||
173 | /* alignment should be aligned with order_per_bit */ | |
174 | if (!IS_ALIGNED(alignment >> PAGE_SHIFT, 1 << order_per_bit)) | |
175 | return -EINVAL; | |
176 | ||
177 | if (ALIGN(base, alignment) != base || ALIGN(size, alignment) != size) | |
178 | return -EINVAL; | |
179 | ||
180 | /* | |
181 | * Each reserved area must be initialised later, when more kernel | |
182 | * subsystems (like slab allocator) are available. | |
183 | */ | |
184 | cma = &cma_areas[cma_area_count]; | |
185 | cma->base_pfn = PFN_DOWN(base); | |
186 | cma->count = size >> PAGE_SHIFT; | |
187 | cma->order_per_bit = order_per_bit; | |
188 | *res_cma = cma; | |
189 | cma_area_count++; | |
190 | ||
191 | return 0; | |
192 | } | |
193 | ||
a254129e JK |
194 | /** |
195 | * cma_declare_contiguous() - reserve custom contiguous area | |
a254129e | 196 | * @base: Base address of the reserved area optional, use 0 for any |
c1f733aa | 197 | * @size: Size of the reserved area (in bytes), |
a254129e JK |
198 | * @limit: End address of the reserved memory (optional, 0 for any). |
199 | * @alignment: Alignment for the CMA area, should be power of 2 or zero | |
200 | * @order_per_bit: Order of pages represented by one bit on bitmap. | |
a254129e | 201 | * @fixed: hint about where to place the reserved area |
c1f733aa | 202 | * @res_cma: Pointer to store the created cma region. |
a254129e JK |
203 | * |
204 | * This function reserves memory from early allocator. It should be | |
205 | * called by arch specific code once the early allocator (memblock or bootmem) | |
206 | * has been activated and all other subsystems have already allocated/reserved | |
207 | * memory. This function allows to create custom reserved areas. | |
208 | * | |
209 | * If @fixed is true, reserve contiguous area at exactly @base. If false, | |
210 | * reserve in range from @base to @limit. | |
211 | */ | |
c1f733aa JK |
212 | int __init cma_declare_contiguous(phys_addr_t base, |
213 | phys_addr_t size, phys_addr_t limit, | |
a254129e | 214 | phys_addr_t alignment, unsigned int order_per_bit, |
c1f733aa | 215 | bool fixed, struct cma **res_cma) |
a254129e | 216 | { |
f7426b98 | 217 | phys_addr_t memblock_end = memblock_end_of_DRAM(); |
6b101e2a | 218 | phys_addr_t highmem_start; |
a254129e JK |
219 | int ret = 0; |
220 | ||
6b101e2a JK |
221 | #ifdef CONFIG_X86 |
222 | /* | |
223 | * high_memory isn't direct mapped memory so retrieving its physical | |
224 | * address isn't appropriate. But it would be useful to check the | |
225 | * physical address of the highmem boundary so it's justfiable to get | |
226 | * the physical address from it. On x86 there is a validation check for | |
227 | * this case, so the following workaround is needed to avoid it. | |
228 | */ | |
229 | highmem_start = __pa_nodebug(high_memory); | |
230 | #else | |
231 | highmem_start = __pa(high_memory); | |
232 | #endif | |
56fa4f60 LP |
233 | pr_debug("%s(size %pa, base %pa, limit %pa alignment %pa)\n", |
234 | __func__, &size, &base, &limit, &alignment); | |
a254129e JK |
235 | |
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 | if (alignment && !is_power_of_2(alignment)) | |
245 | return -EINVAL; | |
246 | ||
247 | /* | |
248 | * Sanitise input arguments. | |
249 | * Pages both ends in CMA area could be merged into adjacent unmovable | |
250 | * migratetype page by page allocator's buddy algorithm. In the case, | |
251 | * you couldn't get a contiguous memory, which is not what we want. | |
252 | */ | |
253 | alignment = max(alignment, | |
254 | (phys_addr_t)PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order)); | |
255 | base = ALIGN(base, alignment); | |
256 | size = ALIGN(size, alignment); | |
257 | limit &= ~(alignment - 1); | |
258 | ||
800a85d3 LP |
259 | if (!base) |
260 | fixed = false; | |
261 | ||
a254129e JK |
262 | /* size should be aligned with order_per_bit */ |
263 | if (!IS_ALIGNED(size >> PAGE_SHIFT, 1 << order_per_bit)) | |
264 | return -EINVAL; | |
265 | ||
f7426b98 | 266 | /* |
16195ddd LP |
267 | * If allocating at a fixed base the request region must not cross the |
268 | * low/high memory boundary. | |
f7426b98 | 269 | */ |
16195ddd | 270 | if (fixed && base < highmem_start && base + size > highmem_start) { |
f7426b98 | 271 | ret = -EINVAL; |
56fa4f60 LP |
272 | pr_err("Region at %pa defined on low/high memory boundary (%pa)\n", |
273 | &base, &highmem_start); | |
f7426b98 MS |
274 | goto err; |
275 | } | |
276 | ||
16195ddd LP |
277 | /* |
278 | * If the limit is unspecified or above the memblock end, its effective | |
279 | * value will be the memblock end. Set it explicitly to simplify further | |
280 | * checks. | |
281 | */ | |
282 | if (limit == 0 || limit > memblock_end) | |
283 | limit = memblock_end; | |
284 | ||
a254129e | 285 | /* Reserve memory */ |
800a85d3 | 286 | if (fixed) { |
a254129e JK |
287 | if (memblock_is_region_reserved(base, size) || |
288 | memblock_reserve(base, size) < 0) { | |
289 | ret = -EBUSY; | |
290 | goto err; | |
291 | } | |
292 | } else { | |
16195ddd LP |
293 | phys_addr_t addr = 0; |
294 | ||
295 | /* | |
296 | * All pages in the reserved area must come from the same zone. | |
297 | * If the requested region crosses the low/high memory boundary, | |
298 | * try allocating from high memory first and fall back to low | |
299 | * memory in case of failure. | |
300 | */ | |
301 | if (base < highmem_start && limit > highmem_start) { | |
302 | addr = memblock_alloc_range(size, alignment, | |
303 | highmem_start, limit); | |
304 | limit = highmem_start; | |
305 | } | |
306 | ||
a254129e | 307 | if (!addr) { |
16195ddd LP |
308 | addr = memblock_alloc_range(size, alignment, base, |
309 | limit); | |
310 | if (!addr) { | |
311 | ret = -ENOMEM; | |
312 | goto err; | |
313 | } | |
a254129e | 314 | } |
16195ddd LP |
315 | |
316 | base = addr; | |
a254129e JK |
317 | } |
318 | ||
de9e14ee MS |
319 | ret = cma_init_reserved_mem(base, size, order_per_bit, res_cma); |
320 | if (ret) | |
321 | goto err; | |
a254129e | 322 | |
56fa4f60 LP |
323 | pr_info("Reserved %ld MiB at %pa\n", (unsigned long)size / SZ_1M, |
324 | &base); | |
a254129e JK |
325 | return 0; |
326 | ||
327 | err: | |
0de9d2eb | 328 | pr_err("Failed to reserve %ld MiB\n", (unsigned long)size / SZ_1M); |
a254129e JK |
329 | return ret; |
330 | } | |
331 | ||
332 | /** | |
333 | * cma_alloc() - allocate pages from contiguous area | |
334 | * @cma: Contiguous memory region for which the allocation is performed. | |
335 | * @count: Requested number of pages. | |
336 | * @align: Requested alignment of pages (in PAGE_SIZE order). | |
337 | * | |
338 | * This function allocates part of contiguous memory on specific | |
339 | * contiguous memory area. | |
340 | */ | |
341 | struct page *cma_alloc(struct cma *cma, int count, unsigned int align) | |
342 | { | |
343 | unsigned long mask, pfn, start = 0; | |
344 | unsigned long bitmap_maxno, bitmap_no, bitmap_count; | |
345 | struct page *page = NULL; | |
346 | int ret; | |
347 | ||
348 | if (!cma || !cma->count) | |
349 | return NULL; | |
350 | ||
351 | pr_debug("%s(cma %p, count %d, align %d)\n", __func__, (void *)cma, | |
352 | count, align); | |
353 | ||
354 | if (!count) | |
355 | return NULL; | |
356 | ||
357 | mask = cma_bitmap_aligned_mask(cma, align); | |
358 | bitmap_maxno = cma_bitmap_maxno(cma); | |
359 | bitmap_count = cma_bitmap_pages_to_bits(cma, count); | |
360 | ||
361 | for (;;) { | |
362 | mutex_lock(&cma->lock); | |
363 | bitmap_no = bitmap_find_next_zero_area(cma->bitmap, | |
364 | bitmap_maxno, start, bitmap_count, mask); | |
365 | if (bitmap_no >= bitmap_maxno) { | |
366 | mutex_unlock(&cma->lock); | |
367 | break; | |
368 | } | |
369 | bitmap_set(cma->bitmap, bitmap_no, bitmap_count); | |
370 | /* | |
371 | * It's safe to drop the lock here. We've marked this region for | |
372 | * our exclusive use. If the migration fails we will take the | |
373 | * lock again and unmark it. | |
374 | */ | |
375 | mutex_unlock(&cma->lock); | |
376 | ||
377 | pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit); | |
378 | mutex_lock(&cma_mutex); | |
379 | ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA); | |
380 | mutex_unlock(&cma_mutex); | |
381 | if (ret == 0) { | |
382 | page = pfn_to_page(pfn); | |
383 | break; | |
a254129e | 384 | } |
b7155e76 | 385 | |
a254129e | 386 | cma_clear_bitmap(cma, pfn, count); |
b7155e76 JK |
387 | if (ret != -EBUSY) |
388 | break; | |
389 | ||
a254129e JK |
390 | pr_debug("%s(): memory range at %p is busy, retrying\n", |
391 | __func__, pfn_to_page(pfn)); | |
392 | /* try again with a bit different memory target */ | |
393 | start = bitmap_no + mask + 1; | |
394 | } | |
395 | ||
396 | pr_debug("%s(): returned %p\n", __func__, page); | |
397 | return page; | |
398 | } | |
399 | ||
400 | /** | |
401 | * cma_release() - release allocated pages | |
402 | * @cma: Contiguous memory region for which the allocation is performed. | |
403 | * @pages: Allocated pages. | |
404 | * @count: Number of allocated pages. | |
405 | * | |
406 | * This function releases memory allocated by alloc_cma(). | |
407 | * It returns false when provided pages do not belong to contiguous area and | |
408 | * true otherwise. | |
409 | */ | |
410 | bool cma_release(struct cma *cma, struct page *pages, 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 | cma_clear_bitmap(cma, pfn, count); | |
428 | ||
429 | return true; | |
430 | } |