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 #include <trace/hooks/mm.h>
54
55 #ifdef CONFIG_CMA_SIZE_MBYTES
56 #define CMA_SIZE_MBYTES CONFIG_CMA_SIZE_MBYTES
57 #else
58 #define CMA_SIZE_MBYTES 0
59 #endif
60
61 struct cma *dma_contiguous_default_area;
62 EXPORT_SYMBOL_GPL(dma_contiguous_default_area);
63
64 /*
65 * Default global CMA area size can be defined in kernel's .config.
66 * This is useful mainly for distro maintainers to create a kernel
67 * that works correctly for most supported systems.
68 * The size can be set in bytes or as a percentage of the total memory
69 * in the system.
70 *
71 * Users, who want to set the size of global CMA area for their system
72 * should use cma= kernel parameter.
73 */
74 static const phys_addr_t size_bytes __initconst =
75 (phys_addr_t)CMA_SIZE_MBYTES * SZ_1M;
76 static phys_addr_t size_cmdline __initdata = -1;
77 static phys_addr_t base_cmdline __initdata;
78 static phys_addr_t limit_cmdline __initdata;
79
early_cma(char * p)80 static int __init early_cma(char *p)
81 {
82 if (!p) {
83 pr_err("Config string not provided\n");
84 return -EINVAL;
85 }
86
87 size_cmdline = memparse(p, &p);
88 if (*p != '@')
89 return 0;
90 base_cmdline = memparse(p + 1, &p);
91 if (*p != '-') {
92 limit_cmdline = base_cmdline + size_cmdline;
93 return 0;
94 }
95 limit_cmdline = memparse(p + 1, &p);
96
97 return 0;
98 }
99 early_param("cma", early_cma);
100
101 #ifdef CONFIG_DMA_PERNUMA_CMA
102
103 static struct cma *dma_contiguous_pernuma_area[MAX_NUMNODES];
104 static phys_addr_t pernuma_size_bytes __initdata;
105
early_cma_pernuma(char * p)106 static int __init early_cma_pernuma(char *p)
107 {
108 pernuma_size_bytes = memparse(p, &p);
109 return 0;
110 }
111 early_param("cma_pernuma", early_cma_pernuma);
112 #endif
113
114 #ifdef CONFIG_CMA_SIZE_PERCENTAGE
115
cma_early_percent_memory(void)116 static phys_addr_t __init __maybe_unused cma_early_percent_memory(void)
117 {
118 unsigned long total_pages = PHYS_PFN(memblock_phys_mem_size());
119
120 return (total_pages * CONFIG_CMA_SIZE_PERCENTAGE / 100) << PAGE_SHIFT;
121 }
122
123 #else
124
cma_early_percent_memory(void)125 static inline __maybe_unused phys_addr_t cma_early_percent_memory(void)
126 {
127 return 0;
128 }
129
130 #endif
131
132 #ifdef CONFIG_DMA_PERNUMA_CMA
dma_pernuma_cma_reserve(void)133 void __init dma_pernuma_cma_reserve(void)
134 {
135 int nid;
136
137 if (!pernuma_size_bytes)
138 return;
139
140 for_each_online_node(nid) {
141 int ret;
142 char name[CMA_MAX_NAME];
143 struct cma **cma = &dma_contiguous_pernuma_area[nid];
144
145 snprintf(name, sizeof(name), "pernuma%d", nid);
146 ret = cma_declare_contiguous_nid(0, pernuma_size_bytes, 0, 0,
147 0, false, name, cma, nid);
148 if (ret) {
149 pr_warn("%s: reservation failed: err %d, node %d", __func__,
150 ret, nid);
151 continue;
152 }
153
154 pr_debug("%s: reserved %llu MiB on node %d\n", __func__,
155 (unsigned long long)pernuma_size_bytes / SZ_1M, nid);
156 }
157 }
158 #endif
159
160 /**
161 * dma_contiguous_reserve() - reserve area(s) for contiguous memory handling
162 * @limit: End address of the reserved memory (optional, 0 for any).
163 *
164 * This function reserves memory from early allocator. It should be
165 * called by arch specific code once the early allocator (memblock or bootmem)
166 * has been activated and all other subsystems have already allocated/reserved
167 * memory.
168 */
dma_contiguous_reserve(phys_addr_t limit)169 void __init dma_contiguous_reserve(phys_addr_t limit)
170 {
171 phys_addr_t selected_size = 0;
172 phys_addr_t selected_base = 0;
173 phys_addr_t selected_limit = limit;
174 bool fixed = false;
175
176 pr_debug("%s(limit %08lx)\n", __func__, (unsigned long)limit);
177
178 if (size_cmdline != -1) {
179 selected_size = size_cmdline;
180 selected_base = base_cmdline;
181 selected_limit = min_not_zero(limit_cmdline, limit);
182 if (base_cmdline + size_cmdline == limit_cmdline)
183 fixed = true;
184 } else {
185 #ifdef CONFIG_CMA_SIZE_SEL_MBYTES
186 selected_size = size_bytes;
187 #elif defined(CONFIG_CMA_SIZE_SEL_PERCENTAGE)
188 selected_size = cma_early_percent_memory();
189 #elif defined(CONFIG_CMA_SIZE_SEL_MIN)
190 selected_size = min(size_bytes, cma_early_percent_memory());
191 #elif defined(CONFIG_CMA_SIZE_SEL_MAX)
192 selected_size = max(size_bytes, cma_early_percent_memory());
193 #endif
194 }
195
196 if (selected_size && !dma_contiguous_default_area) {
197 pr_debug("%s: reserving %ld MiB for global area\n", __func__,
198 (unsigned long)selected_size / SZ_1M);
199
200 dma_contiguous_reserve_area(selected_size, selected_base,
201 selected_limit,
202 &dma_contiguous_default_area,
203 fixed);
204 }
205 }
206
207 void __weak
dma_contiguous_early_fixup(phys_addr_t base,unsigned long size)208 dma_contiguous_early_fixup(phys_addr_t base, unsigned long size)
209 {
210 }
211
212 /**
213 * dma_contiguous_reserve_area() - reserve custom contiguous area
214 * @size: Size of the reserved area (in bytes),
215 * @base: Base address of the reserved area optional, use 0 for any
216 * @limit: End address of the reserved memory (optional, 0 for any).
217 * @res_cma: Pointer to store the created cma region.
218 * @fixed: hint about where to place the reserved area
219 *
220 * This function reserves memory from early allocator. It should be
221 * called by arch specific code once the early allocator (memblock or bootmem)
222 * has been activated and all other subsystems have already allocated/reserved
223 * memory. This function allows to create custom reserved areas for specific
224 * devices.
225 *
226 * If @fixed is true, reserve contiguous area at exactly @base. If false,
227 * reserve in range from @base to @limit.
228 */
dma_contiguous_reserve_area(phys_addr_t size,phys_addr_t base,phys_addr_t limit,struct cma ** res_cma,bool fixed)229 int __init dma_contiguous_reserve_area(phys_addr_t size, phys_addr_t base,
230 phys_addr_t limit, struct cma **res_cma,
231 bool fixed)
232 {
233 int ret;
234
235 ret = cma_declare_contiguous(base, size, limit, 0, 0, fixed,
236 "reserved", res_cma);
237 if (ret)
238 return ret;
239
240 /* Architecture specific contiguous memory fixup. */
241 dma_contiguous_early_fixup(cma_get_base(*res_cma),
242 cma_get_size(*res_cma));
243
244 return 0;
245 }
246
247 /**
248 * dma_alloc_from_contiguous() - allocate pages from contiguous area
249 * @dev: Pointer to device for which the allocation is performed.
250 * @count: Requested number of pages.
251 * @align: Requested alignment of pages (in PAGE_SIZE order).
252 * @no_warn: Avoid printing message about failed allocation.
253 *
254 * This function allocates memory buffer for specified device. It uses
255 * device specific contiguous memory area if available or the default
256 * global one. Requires architecture specific dev_get_cma_area() helper
257 * function.
258 */
dma_alloc_from_contiguous(struct device * dev,size_t count,unsigned int align,bool no_warn)259 struct page *dma_alloc_from_contiguous(struct device *dev, size_t count,
260 unsigned int align, bool no_warn)
261 {
262 if (align > CONFIG_CMA_ALIGNMENT)
263 align = CONFIG_CMA_ALIGNMENT;
264
265 return cma_alloc(dev_get_cma_area(dev), count, align, GFP_KERNEL |
266 (no_warn ? __GFP_NOWARN : 0));
267 }
268
269 /**
270 * dma_release_from_contiguous() - release allocated pages
271 * @dev: Pointer to device for which the pages were allocated.
272 * @pages: Allocated pages.
273 * @count: Number of allocated pages.
274 *
275 * This function releases memory allocated by dma_alloc_from_contiguous().
276 * It returns false when provided pages do not belong to contiguous area and
277 * true otherwise.
278 */
dma_release_from_contiguous(struct device * dev,struct page * pages,int count)279 bool dma_release_from_contiguous(struct device *dev, struct page *pages,
280 int count)
281 {
282 return cma_release(dev_get_cma_area(dev), pages, count);
283 }
284
cma_alloc_aligned(struct cma * cma,size_t size,gfp_t gfp)285 static struct page *cma_alloc_aligned(struct cma *cma, size_t size, gfp_t gfp)
286 {
287 unsigned int align = min(get_order(size), CONFIG_CMA_ALIGNMENT);
288
289 return cma_alloc(cma, size >> PAGE_SHIFT, align,
290 GFP_KERNEL | (gfp & __GFP_NOWARN));
291 }
292
293 /**
294 * dma_alloc_contiguous() - allocate contiguous pages
295 * @dev: Pointer to device for which the allocation is performed.
296 * @size: Requested allocation size.
297 * @gfp: Allocation flags.
298 *
299 * tries to use device specific contiguous memory area if available, or it
300 * tries to use per-numa cma, if the allocation fails, it will fallback to
301 * try default global one.
302 *
303 * Note that it bypass one-page size of allocations from the per-numa and
304 * global area as the addresses within one page are always contiguous, so
305 * there is no need to waste CMA pages for that kind; it also helps reduce
306 * fragmentations.
307 */
dma_alloc_contiguous(struct device * dev,size_t size,gfp_t gfp)308 struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
309 {
310 #ifdef CONFIG_DMA_PERNUMA_CMA
311 int nid = dev_to_node(dev);
312 #endif
313 bool allow_subpage_alloc = false;
314
315 /* CMA can be used only in the context which permits sleeping */
316 if (!gfpflags_allow_blocking(gfp))
317 return NULL;
318 if (dev->cma_area)
319 return cma_alloc_aligned(dev->cma_area, size, gfp);
320
321 if (size <= PAGE_SIZE) {
322 trace_android_vh_subpage_dma_contig_alloc(&allow_subpage_alloc, dev, &size);
323 if (!allow_subpage_alloc)
324 return NULL;
325 }
326
327 #ifdef CONFIG_DMA_PERNUMA_CMA
328 if (nid != NUMA_NO_NODE && !(gfp & (GFP_DMA | GFP_DMA32))) {
329 struct cma *cma = dma_contiguous_pernuma_area[nid];
330 struct page *page;
331
332 if (cma) {
333 page = cma_alloc_aligned(cma, size, gfp);
334 if (page)
335 return page;
336 }
337 }
338 #endif
339 if (!dma_contiguous_default_area)
340 return NULL;
341
342 return cma_alloc_aligned(dma_contiguous_default_area, size, gfp);
343 }
344
345 /**
346 * dma_free_contiguous() - release allocated pages
347 * @dev: Pointer to device for which the pages were allocated.
348 * @page: Pointer to the allocated pages.
349 * @size: Size of allocated pages.
350 *
351 * This function releases memory allocated by dma_alloc_contiguous(). As the
352 * cma_release returns false when provided pages do not belong to contiguous
353 * area and true otherwise, this function then does a fallback __free_pages()
354 * upon a false-return.
355 */
dma_free_contiguous(struct device * dev,struct page * page,size_t size)356 void dma_free_contiguous(struct device *dev, struct page *page, size_t size)
357 {
358 unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT;
359
360 /* if dev has its own cma, free page from there */
361 if (dev->cma_area) {
362 if (cma_release(dev->cma_area, page, count))
363 return;
364 } else {
365 /*
366 * otherwise, page is from either per-numa cma or default cma
367 */
368 #ifdef CONFIG_DMA_PERNUMA_CMA
369 if (cma_release(dma_contiguous_pernuma_area[page_to_nid(page)],
370 page, count))
371 return;
372 #endif
373 if (cma_release(dma_contiguous_default_area, page, count))
374 return;
375 }
376
377 /* not in any cma, free from buddy */
378 __free_pages(page, get_order(size));
379 }
380
381 /*
382 * Support for reserved memory regions defined in device tree
383 */
384 #ifdef CONFIG_OF_RESERVED_MEM
385 #include <linux/of.h>
386 #include <linux/of_fdt.h>
387 #include <linux/of_reserved_mem.h>
388
389 #undef pr_fmt
390 #define pr_fmt(fmt) fmt
391
rmem_cma_device_init(struct reserved_mem * rmem,struct device * dev)392 static int rmem_cma_device_init(struct reserved_mem *rmem, struct device *dev)
393 {
394 dev->cma_area = rmem->priv;
395 return 0;
396 }
397
rmem_cma_device_release(struct reserved_mem * rmem,struct device * dev)398 static void rmem_cma_device_release(struct reserved_mem *rmem,
399 struct device *dev)
400 {
401 dev->cma_area = NULL;
402 }
403
404 static const struct reserved_mem_ops rmem_cma_ops = {
405 .device_init = rmem_cma_device_init,
406 .device_release = rmem_cma_device_release,
407 };
408
rmem_cma_setup(struct reserved_mem * rmem)409 static int __init rmem_cma_setup(struct reserved_mem *rmem)
410 {
411 phys_addr_t align = PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order);
412 phys_addr_t mask = align - 1;
413 unsigned long node = rmem->fdt_node;
414 bool default_cma = of_get_flat_dt_prop(node, "linux,cma-default", NULL);
415 struct cma *cma;
416 int err;
417
418 if (size_cmdline != -1 && default_cma) {
419 pr_info("Reserved memory: bypass %s node, using cmdline CMA params instead\n",
420 rmem->name);
421 return -EBUSY;
422 }
423
424 if (!of_get_flat_dt_prop(node, "reusable", NULL) ||
425 of_get_flat_dt_prop(node, "no-map", NULL))
426 return -EINVAL;
427
428 if ((rmem->base & mask) || (rmem->size & mask)) {
429 pr_err("Reserved memory: incorrect alignment of CMA region\n");
430 return -EINVAL;
431 }
432
433 err = cma_init_reserved_mem(rmem->base, rmem->size, 0, rmem->name, &cma);
434 if (err) {
435 pr_err("Reserved memory: unable to setup CMA region\n");
436 return err;
437 }
438 /* Architecture specific contiguous memory fixup. */
439 dma_contiguous_early_fixup(rmem->base, rmem->size);
440
441 if (default_cma)
442 dma_contiguous_default_area = cma;
443
444 rmem->ops = &rmem_cma_ops;
445 rmem->priv = cma;
446
447 pr_info("Reserved memory: created CMA memory pool at %pa, size %ld MiB\n",
448 &rmem->base, (unsigned long)rmem->size / SZ_1M);
449
450 return 0;
451 }
452 RESERVEDMEM_OF_DECLARE(cma, "shared-dma-pool", rmem_cma_setup);
453 #endif
454