1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3 * Contiguous Memory Allocator for DMA mapping framework
4 * Copyright (c) 2010-2011 by Samsung Electronics.
5 * Written by:
6 * Marek Szyprowski <m.szyprowski@samsung.com>
7 * Michal Nazarewicz <mina86@mina86.com>
8 *
9 * Contiguous Memory Allocator
10 *
11 * The Contiguous Memory Allocator (CMA) makes it possible to
12 * allocate big contiguous chunks of memory after the system has
13 * booted.
14 *
15 * Why is it needed?
16 *
17 * Various devices on embedded systems have no scatter-getter and/or
18 * IO map support and require contiguous blocks of memory to
19 * operate. They include devices such as cameras, hardware video
20 * coders, etc.
21 *
22 * Such devices often require big memory buffers (a full HD frame
23 * is, for instance, more then 2 mega pixels large, i.e. more than 6
24 * MB of memory), which makes mechanisms such as kmalloc() or
25 * alloc_page() ineffective.
26 *
27 * At the same time, a solution where a big memory region is
28 * reserved for a device is suboptimal since often more memory is
29 * reserved then strictly required and, moreover, the memory is
30 * inaccessible to page system even if device drivers don't use it.
31 *
32 * CMA tries to solve this issue by operating on memory regions
33 * where only movable pages can be allocated from. This way, kernel
34 * can use the memory for pagecache and when device driver requests
35 * it, allocated pages can be migrated.
36 */
37
38 #define pr_fmt(fmt) "cma: " fmt
39
40 #ifdef CONFIG_CMA_DEBUG
41 #ifndef DEBUG
42 # define DEBUG
43 #endif
44 #endif
45
46 #include <asm/page.h>
47
48 #include <linux/memblock.h>
49 #include <linux/err.h>
50 #include <linux/sizes.h>
51 #include <linux/dma-map-ops.h>
52 #include <linux/cma.h>
53
54 #ifdef CONFIG_CMA_SIZE_MBYTES
55 #define CMA_SIZE_MBYTES CONFIG_CMA_SIZE_MBYTES
56 #else
57 #define CMA_SIZE_MBYTES 0
58 #endif
59
60 struct cma *dma_contiguous_default_area;
61 EXPORT_SYMBOL_GPL(dma_contiguous_default_area);
62
63 /*
64 * Default global CMA area size can be defined in kernel's .config.
65 * This is useful mainly for distro maintainers to create a kernel
66 * that works correctly for most supported systems.
67 * The size can be set in bytes or as a percentage of the total memory
68 * in the system.
69 *
70 * Users, who want to set the size of global CMA area for their system
71 * should use cma= kernel parameter.
72 */
73 static const phys_addr_t size_bytes __initconst =
74 (phys_addr_t)CMA_SIZE_MBYTES * SZ_1M;
75 static phys_addr_t size_cmdline __initdata = -1;
76 static phys_addr_t base_cmdline __initdata;
77 static phys_addr_t limit_cmdline __initdata;
78
early_cma(char * p)79 static int __init early_cma(char *p)
80 {
81 if (!p) {
82 pr_err("Config string not provided\n");
83 return -EINVAL;
84 }
85
86 size_cmdline = memparse(p, &p);
87 if (*p != '@')
88 return 0;
89 base_cmdline = memparse(p + 1, &p);
90 if (*p != '-') {
91 limit_cmdline = base_cmdline + size_cmdline;
92 return 0;
93 }
94 limit_cmdline = memparse(p + 1, &p);
95
96 return 0;
97 }
98 early_param("cma", early_cma);
99
100 #ifdef CONFIG_DMA_PERNUMA_CMA
101
102 static struct cma *dma_contiguous_pernuma_area[MAX_NUMNODES];
103 static phys_addr_t pernuma_size_bytes __initdata;
104
early_cma_pernuma(char * p)105 static int __init early_cma_pernuma(char *p)
106 {
107 pernuma_size_bytes = memparse(p, &p);
108 return 0;
109 }
110 early_param("cma_pernuma", early_cma_pernuma);
111 #endif
112
113 #ifdef CONFIG_CMA_SIZE_PERCENTAGE
114
cma_early_percent_memory(void)115 static phys_addr_t __init __maybe_unused cma_early_percent_memory(void)
116 {
117 unsigned long total_pages = PHYS_PFN(memblock_phys_mem_size());
118
119 return (total_pages * CONFIG_CMA_SIZE_PERCENTAGE / 100) << PAGE_SHIFT;
120 }
121
122 #else
123
cma_early_percent_memory(void)124 static inline __maybe_unused phys_addr_t cma_early_percent_memory(void)
125 {
126 return 0;
127 }
128
129 #endif
130
131 #ifdef CONFIG_DMA_PERNUMA_CMA
dma_pernuma_cma_reserve(void)132 void __init dma_pernuma_cma_reserve(void)
133 {
134 int nid;
135
136 if (!pernuma_size_bytes)
137 return;
138
139 for_each_online_node(nid) {
140 int ret;
141 char name[CMA_MAX_NAME];
142 struct cma **cma = &dma_contiguous_pernuma_area[nid];
143
144 snprintf(name, sizeof(name), "pernuma%d", nid);
145 ret = cma_declare_contiguous_nid(0, pernuma_size_bytes, 0, 0,
146 0, false, name, cma, nid);
147 if (ret) {
148 pr_warn("%s: reservation failed: err %d, node %d", __func__,
149 ret, nid);
150 continue;
151 }
152
153 pr_debug("%s: reserved %llu MiB on node %d\n", __func__,
154 (unsigned long long)pernuma_size_bytes / SZ_1M, nid);
155 }
156 }
157 #endif
158
159 /**
160 * dma_contiguous_reserve() - reserve area(s) for contiguous memory handling
161 * @limit: End address of the reserved memory (optional, 0 for any).
162 *
163 * This function reserves memory from early allocator. It should be
164 * called by arch specific code once the early allocator (memblock or bootmem)
165 * has been activated and all other subsystems have already allocated/reserved
166 * memory.
167 */
dma_contiguous_reserve(phys_addr_t limit)168 void __init dma_contiguous_reserve(phys_addr_t limit)
169 {
170 phys_addr_t selected_size = 0;
171 phys_addr_t selected_base = 0;
172 phys_addr_t selected_limit = limit;
173 bool fixed = false;
174
175 pr_debug("%s(limit %08lx)\n", __func__, (unsigned long)limit);
176
177 if (size_cmdline != -1) {
178 selected_size = size_cmdline;
179 selected_base = base_cmdline;
180 selected_limit = min_not_zero(limit_cmdline, limit);
181 if (base_cmdline + size_cmdline == limit_cmdline)
182 fixed = true;
183 } else {
184 #ifdef CONFIG_CMA_SIZE_SEL_MBYTES
185 selected_size = size_bytes;
186 #elif defined(CONFIG_CMA_SIZE_SEL_PERCENTAGE)
187 selected_size = cma_early_percent_memory();
188 #elif defined(CONFIG_CMA_SIZE_SEL_MIN)
189 selected_size = min(size_bytes, cma_early_percent_memory());
190 #elif defined(CONFIG_CMA_SIZE_SEL_MAX)
191 selected_size = max(size_bytes, cma_early_percent_memory());
192 #endif
193 }
194
195 if (selected_size && !dma_contiguous_default_area) {
196 pr_debug("%s: reserving %ld MiB for global area\n", __func__,
197 (unsigned long)selected_size / SZ_1M);
198
199 dma_contiguous_reserve_area(selected_size, selected_base,
200 selected_limit,
201 &dma_contiguous_default_area,
202 fixed);
203 }
204 }
205
206 void __weak
dma_contiguous_early_fixup(phys_addr_t base,unsigned long size)207 dma_contiguous_early_fixup(phys_addr_t base, unsigned long size)
208 {
209 }
210
211 /**
212 * dma_contiguous_reserve_area() - reserve custom contiguous area
213 * @size: Size of the reserved area (in bytes),
214 * @base: Base address of the reserved area optional, use 0 for any
215 * @limit: End address of the reserved memory (optional, 0 for any).
216 * @res_cma: Pointer to store the created cma region.
217 * @fixed: hint about where to place the reserved area
218 *
219 * This function reserves memory from early allocator. It should be
220 * called by arch specific code once the early allocator (memblock or bootmem)
221 * has been activated and all other subsystems have already allocated/reserved
222 * memory. This function allows to create custom reserved areas for specific
223 * devices.
224 *
225 * If @fixed is true, reserve contiguous area at exactly @base. If false,
226 * reserve in range from @base to @limit.
227 */
dma_contiguous_reserve_area(phys_addr_t size,phys_addr_t base,phys_addr_t limit,struct cma ** res_cma,bool fixed)228 int __init dma_contiguous_reserve_area(phys_addr_t size, phys_addr_t base,
229 phys_addr_t limit, struct cma **res_cma,
230 bool fixed)
231 {
232 int ret;
233
234 ret = cma_declare_contiguous(base, size, limit, 0, 0, fixed,
235 "reserved", res_cma);
236 if (ret)
237 return ret;
238
239 /* Architecture specific contiguous memory fixup. */
240 dma_contiguous_early_fixup(cma_get_base(*res_cma),
241 cma_get_size(*res_cma));
242
243 return 0;
244 }
245
246 /**
247 * dma_alloc_from_contiguous() - allocate pages from contiguous area
248 * @dev: Pointer to device for which the allocation is performed.
249 * @count: Requested number of pages.
250 * @align: Requested alignment of pages (in PAGE_SIZE order).
251 * @no_warn: Avoid printing message about failed allocation.
252 *
253 * This function allocates memory buffer for specified device. It uses
254 * device specific contiguous memory area if available or the default
255 * global one. Requires architecture specific dev_get_cma_area() helper
256 * function.
257 */
dma_alloc_from_contiguous(struct device * dev,size_t count,unsigned int align,bool no_warn)258 struct page *dma_alloc_from_contiguous(struct device *dev, size_t count,
259 unsigned int align, bool no_warn)
260 {
261 if (align > CONFIG_CMA_ALIGNMENT)
262 align = CONFIG_CMA_ALIGNMENT;
263
264 return cma_alloc(dev_get_cma_area(dev), count, align, GFP_KERNEL |
265 (no_warn ? __GFP_NOWARN : 0));
266 }
267
268 /**
269 * dma_release_from_contiguous() - release allocated pages
270 * @dev: Pointer to device for which the pages were allocated.
271 * @pages: Allocated pages.
272 * @count: Number of allocated pages.
273 *
274 * This function releases memory allocated by dma_alloc_from_contiguous().
275 * It returns false when provided pages do not belong to contiguous area and
276 * true otherwise.
277 */
dma_release_from_contiguous(struct device * dev,struct page * pages,int count)278 bool dma_release_from_contiguous(struct device *dev, struct page *pages,
279 int count)
280 {
281 return cma_release(dev_get_cma_area(dev), pages, count);
282 }
283
cma_alloc_aligned(struct cma * cma,size_t size,gfp_t gfp)284 static struct page *cma_alloc_aligned(struct cma *cma, size_t size, gfp_t gfp)
285 {
286 unsigned int align = min(get_order(size), CONFIG_CMA_ALIGNMENT);
287
288 return cma_alloc(cma, size >> PAGE_SHIFT, align,
289 GFP_KERNEL | (gfp & __GFP_NOWARN));
290 }
291
292 /**
293 * dma_alloc_contiguous() - allocate contiguous pages
294 * @dev: Pointer to device for which the allocation is performed.
295 * @size: Requested allocation size.
296 * @gfp: Allocation flags.
297 *
298 * tries to use device specific contiguous memory area if available, or it
299 * tries to use per-numa cma, if the allocation fails, it will fallback to
300 * try default global one.
301 *
302 * Note that it bypass one-page size of allocations from the per-numa and
303 * global area as the addresses within one page are always contiguous, so
304 * there is no need to waste CMA pages for that kind; it also helps reduce
305 * fragmentations.
306 */
dma_alloc_contiguous(struct device * dev,size_t size,gfp_t gfp)307 struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
308 {
309 #ifdef CONFIG_DMA_PERNUMA_CMA
310 int nid = dev_to_node(dev);
311 #endif
312
313 /* CMA can be used only in the context which permits sleeping */
314 if (!gfpflags_allow_blocking(gfp))
315 return NULL;
316 if (dev->cma_area)
317 return cma_alloc_aligned(dev->cma_area, size, gfp);
318 if (size <= PAGE_SIZE)
319 return NULL;
320
321 #ifdef CONFIG_DMA_PERNUMA_CMA
322 if (nid != NUMA_NO_NODE && !(gfp & (GFP_DMA | GFP_DMA32))) {
323 struct cma *cma = dma_contiguous_pernuma_area[nid];
324 struct page *page;
325
326 if (cma) {
327 page = cma_alloc_aligned(cma, size, gfp);
328 if (page)
329 return page;
330 }
331 }
332 #endif
333 if (!dma_contiguous_default_area)
334 return NULL;
335
336 return cma_alloc_aligned(dma_contiguous_default_area, size, gfp);
337 }
338
339 /**
340 * dma_free_contiguous() - release allocated pages
341 * @dev: Pointer to device for which the pages were allocated.
342 * @page: Pointer to the allocated pages.
343 * @size: Size of allocated pages.
344 *
345 * This function releases memory allocated by dma_alloc_contiguous(). As the
346 * cma_release returns false when provided pages do not belong to contiguous
347 * area and true otherwise, this function then does a fallback __free_pages()
348 * upon a false-return.
349 */
dma_free_contiguous(struct device * dev,struct page * page,size_t size)350 void dma_free_contiguous(struct device *dev, struct page *page, size_t size)
351 {
352 unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT;
353
354 /* if dev has its own cma, free page from there */
355 if (dev->cma_area) {
356 if (cma_release(dev->cma_area, page, count))
357 return;
358 } else {
359 /*
360 * otherwise, page is from either per-numa cma or default cma
361 */
362 #ifdef CONFIG_DMA_PERNUMA_CMA
363 if (cma_release(dma_contiguous_pernuma_area[page_to_nid(page)],
364 page, count))
365 return;
366 #endif
367 if (cma_release(dma_contiguous_default_area, page, count))
368 return;
369 }
370
371 /* not in any cma, free from buddy */
372 __free_pages(page, get_order(size));
373 }
374
375 /*
376 * Support for reserved memory regions defined in device tree
377 */
378 #ifdef CONFIG_OF_RESERVED_MEM
379 #include <linux/of.h>
380 #include <linux/of_fdt.h>
381 #include <linux/of_reserved_mem.h>
382
383 #undef pr_fmt
384 #define pr_fmt(fmt) fmt
385
rmem_cma_device_init(struct reserved_mem * rmem,struct device * dev)386 static int rmem_cma_device_init(struct reserved_mem *rmem, struct device *dev)
387 {
388 dev->cma_area = rmem->priv;
389 return 0;
390 }
391
rmem_cma_device_release(struct reserved_mem * rmem,struct device * dev)392 static void rmem_cma_device_release(struct reserved_mem *rmem,
393 struct device *dev)
394 {
395 dev->cma_area = NULL;
396 }
397
398 static const struct reserved_mem_ops rmem_cma_ops = {
399 .device_init = rmem_cma_device_init,
400 .device_release = rmem_cma_device_release,
401 };
402
rmem_cma_setup(struct reserved_mem * rmem)403 static int __init rmem_cma_setup(struct reserved_mem *rmem)
404 {
405 phys_addr_t align = PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order);
406 phys_addr_t mask = align - 1;
407 unsigned long node = rmem->fdt_node;
408 bool default_cma = of_get_flat_dt_prop(node, "linux,cma-default", NULL);
409 struct cma *cma;
410 int err;
411
412 if (size_cmdline != -1 && default_cma) {
413 pr_info("Reserved memory: bypass %s node, using cmdline CMA params instead\n",
414 rmem->name);
415 return -EBUSY;
416 }
417
418 if (!of_get_flat_dt_prop(node, "reusable", NULL) ||
419 of_get_flat_dt_prop(node, "no-map", NULL))
420 return -EINVAL;
421
422 if ((rmem->base & mask) || (rmem->size & mask)) {
423 pr_err("Reserved memory: incorrect alignment of CMA region\n");
424 return -EINVAL;
425 }
426
427 err = cma_init_reserved_mem(rmem->base, rmem->size, 0, rmem->name, &cma);
428 if (err) {
429 pr_err("Reserved memory: unable to setup CMA region\n");
430 return err;
431 }
432 /* Architecture specific contiguous memory fixup. */
433 dma_contiguous_early_fixup(rmem->base, rmem->size);
434
435 if (default_cma)
436 dma_contiguous_default_area = cma;
437
438 rmem->ops = &rmem_cma_ops;
439 rmem->priv = cma;
440
441 pr_info("Reserved memory: created CMA memory pool at %pa, size %ld MiB\n",
442 &rmem->base, (unsigned long)rmem->size / SZ_1M);
443
444 return 0;
445 }
446 RESERVEDMEM_OF_DECLARE(cma, "shared-dma-pool", rmem_cma_setup);
447 #endif
448