xref: /openbmc/linux/mm/cma.c (revision e9246c87)
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 	if (align_order <= cma->order_per_bit)
62 		return 0;
63 	return (1UL << (align_order - cma->order_per_bit)) - 1;
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);
127 	return -EINVAL;
128 }
129 
130 static int __init cma_init_reserved_areas(void)
131 {
132 	int i;
133 
134 	for (i = 0; i < cma_area_count; i++) {
135 		int ret = cma_activate_area(&cma_areas[i]);
136 
137 		if (ret)
138 			return ret;
139 	}
140 
141 	return 0;
142 }
143 core_initcall(cma_init_reserved_areas);
144 
145 /**
146  * cma_init_reserved_mem() - create custom contiguous area from reserved memory
147  * @base: Base address of the reserved area
148  * @size: Size of the reserved area (in bytes),
149  * @order_per_bit: Order of pages represented by one bit on bitmap.
150  * @res_cma: Pointer to store the created cma region.
151  *
152  * This function creates custom contiguous area from already reserved memory.
153  */
154 int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
155 				 int order_per_bit, struct cma **res_cma)
156 {
157 	struct cma *cma;
158 	phys_addr_t alignment;
159 
160 	/* Sanity checks */
161 	if (cma_area_count == ARRAY_SIZE(cma_areas)) {
162 		pr_err("Not enough slots for CMA reserved regions!\n");
163 		return -ENOSPC;
164 	}
165 
166 	if (!size || !memblock_is_region_reserved(base, size))
167 		return -EINVAL;
168 
169 	/* ensure minimal alignment requied by mm core */
170 	alignment = PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order);
171 
172 	/* alignment should be aligned with order_per_bit */
173 	if (!IS_ALIGNED(alignment >> PAGE_SHIFT, 1 << order_per_bit))
174 		return -EINVAL;
175 
176 	if (ALIGN(base, alignment) != base || ALIGN(size, alignment) != size)
177 		return -EINVAL;
178 
179 	/*
180 	 * Each reserved area must be initialised later, when more kernel
181 	 * subsystems (like slab allocator) are available.
182 	 */
183 	cma = &cma_areas[cma_area_count];
184 	cma->base_pfn = PFN_DOWN(base);
185 	cma->count = size >> PAGE_SHIFT;
186 	cma->order_per_bit = order_per_bit;
187 	*res_cma = cma;
188 	cma_area_count++;
189 
190 	return 0;
191 }
192 
193 /**
194  * cma_declare_contiguous() - reserve custom contiguous area
195  * @base: Base address of the reserved area optional, use 0 for any
196  * @size: Size of the reserved area (in bytes),
197  * @limit: End address of the reserved memory (optional, 0 for any).
198  * @alignment: Alignment for the CMA area, should be power of 2 or zero
199  * @order_per_bit: Order of pages represented by one bit on bitmap.
200  * @fixed: hint about where to place the reserved area
201  * @res_cma: Pointer to store the created cma region.
202  *
203  * This function reserves memory from early allocator. It should be
204  * called by arch specific code once the early allocator (memblock or bootmem)
205  * has been activated and all other subsystems have already allocated/reserved
206  * memory. This function allows to create custom reserved areas.
207  *
208  * If @fixed is true, reserve contiguous area at exactly @base.  If false,
209  * reserve in range from @base to @limit.
210  */
211 int __init cma_declare_contiguous(phys_addr_t base,
212 			phys_addr_t size, phys_addr_t limit,
213 			phys_addr_t alignment, unsigned int order_per_bit,
214 			bool fixed, struct cma **res_cma)
215 {
216 	phys_addr_t memblock_end = memblock_end_of_DRAM();
217 	phys_addr_t highmem_start = __pa(high_memory);
218 	int ret = 0;
219 
220 	pr_debug("%s(size %lx, base %08lx, limit %08lx alignment %08lx)\n",
221 		__func__, (unsigned long)size, (unsigned long)base,
222 		(unsigned long)limit, (unsigned long)alignment);
223 
224 	if (cma_area_count == ARRAY_SIZE(cma_areas)) {
225 		pr_err("Not enough slots for CMA reserved regions!\n");
226 		return -ENOSPC;
227 	}
228 
229 	if (!size)
230 		return -EINVAL;
231 
232 	if (alignment && !is_power_of_2(alignment))
233 		return -EINVAL;
234 
235 	/*
236 	 * Sanitise input arguments.
237 	 * Pages both ends in CMA area could be merged into adjacent unmovable
238 	 * migratetype page by page allocator's buddy algorithm. In the case,
239 	 * you couldn't get a contiguous memory, which is not what we want.
240 	 */
241 	alignment = max(alignment,
242 		(phys_addr_t)PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order));
243 	base = ALIGN(base, alignment);
244 	size = ALIGN(size, alignment);
245 	limit &= ~(alignment - 1);
246 
247 	/* size should be aligned with order_per_bit */
248 	if (!IS_ALIGNED(size >> PAGE_SHIFT, 1 << order_per_bit))
249 		return -EINVAL;
250 
251 	/*
252 	 * adjust limit to avoid crossing low/high memory boundary for
253 	 * automatically allocated regions
254 	 */
255 	if (((limit == 0 || limit > memblock_end) &&
256 	     (memblock_end - size < highmem_start &&
257 	      memblock_end > highmem_start)) ||
258 	    (!fixed && limit > highmem_start && limit - size < highmem_start)) {
259 		limit = highmem_start;
260 	}
261 
262 	if (fixed && base < highmem_start && base+size > highmem_start) {
263 		ret = -EINVAL;
264 		pr_err("Region at %08lx defined on low/high memory boundary (%08lx)\n",
265 			(unsigned long)base, (unsigned long)highmem_start);
266 		goto err;
267 	}
268 
269 	/* Reserve memory */
270 	if (base && fixed) {
271 		if (memblock_is_region_reserved(base, size) ||
272 		    memblock_reserve(base, size) < 0) {
273 			ret = -EBUSY;
274 			goto err;
275 		}
276 	} else {
277 		phys_addr_t addr = memblock_alloc_range(size, alignment, base,
278 							limit);
279 		if (!addr) {
280 			ret = -ENOMEM;
281 			goto err;
282 		} else {
283 			base = addr;
284 		}
285 	}
286 
287 	ret = cma_init_reserved_mem(base, size, order_per_bit, res_cma);
288 	if (ret)
289 		goto err;
290 
291 	pr_info("Reserved %ld MiB at %08lx\n", (unsigned long)size / SZ_1M,
292 		(unsigned long)base);
293 	return 0;
294 
295 err:
296 	pr_err("Failed to reserve %ld MiB\n", (unsigned long)size / SZ_1M);
297 	return ret;
298 }
299 
300 /**
301  * cma_alloc() - allocate pages from contiguous area
302  * @cma:   Contiguous memory region for which the allocation is performed.
303  * @count: Requested number of pages.
304  * @align: Requested alignment of pages (in PAGE_SIZE order).
305  *
306  * This function allocates part of contiguous memory on specific
307  * contiguous memory area.
308  */
309 struct page *cma_alloc(struct cma *cma, int count, unsigned int align)
310 {
311 	unsigned long mask, pfn, start = 0;
312 	unsigned long bitmap_maxno, bitmap_no, bitmap_count;
313 	struct page *page = NULL;
314 	int ret;
315 
316 	if (!cma || !cma->count)
317 		return NULL;
318 
319 	pr_debug("%s(cma %p, count %d, align %d)\n", __func__, (void *)cma,
320 		 count, align);
321 
322 	if (!count)
323 		return NULL;
324 
325 	mask = cma_bitmap_aligned_mask(cma, align);
326 	bitmap_maxno = cma_bitmap_maxno(cma);
327 	bitmap_count = cma_bitmap_pages_to_bits(cma, count);
328 
329 	for (;;) {
330 		mutex_lock(&cma->lock);
331 		bitmap_no = bitmap_find_next_zero_area(cma->bitmap,
332 				bitmap_maxno, start, bitmap_count, mask);
333 		if (bitmap_no >= bitmap_maxno) {
334 			mutex_unlock(&cma->lock);
335 			break;
336 		}
337 		bitmap_set(cma->bitmap, bitmap_no, bitmap_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 + (bitmap_no << cma->order_per_bit);
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 		}
353 
354 		cma_clear_bitmap(cma, pfn, count);
355 		if (ret != -EBUSY)
356 			break;
357 
358 		pr_debug("%s(): memory range at %p is busy, retrying\n",
359 			 __func__, pfn_to_page(pfn));
360 		/* try again with a bit different memory target */
361 		start = bitmap_no + mask + 1;
362 	}
363 
364 	pr_debug("%s(): returned %p\n", __func__, page);
365 	return page;
366 }
367 
368 /**
369  * cma_release() - release allocated pages
370  * @cma:   Contiguous memory region for which the allocation is performed.
371  * @pages: Allocated pages.
372  * @count: Number of allocated pages.
373  *
374  * This function releases memory allocated by alloc_cma().
375  * It returns false when provided pages do not belong to contiguous area and
376  * true otherwise.
377  */
378 bool cma_release(struct cma *cma, struct page *pages, int count)
379 {
380 	unsigned long pfn;
381 
382 	if (!cma || !pages)
383 		return false;
384 
385 	pr_debug("%s(page %p)\n", __func__, (void *)pages);
386 
387 	pfn = page_to_pfn(pages);
388 
389 	if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count)
390 		return false;
391 
392 	VM_BUG_ON(pfn + count > cma->base_pfn + cma->count);
393 
394 	free_contig_range(pfn, count);
395 	cma_clear_bitmap(cma, pfn, count);
396 
397 	return true;
398 }
399