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 than 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 #include <asm/page.h>
41
42 #include <linux/memblock.h>
43 #include <linux/err.h>
44 #include <linux/sizes.h>
45 #include <linux/dma-map-ops.h>
46 #include <linux/cma.h>
47 #include <linux/nospec.h>
48
49 #ifdef CONFIG_CMA_SIZE_MBYTES
50 #define CMA_SIZE_MBYTES CONFIG_CMA_SIZE_MBYTES
51 #else
52 #define CMA_SIZE_MBYTES 0
53 #endif
54
55 static struct cma *dma_contiguous_areas[MAX_CMA_AREAS];
56 static unsigned int dma_contiguous_areas_num;
57
dma_contiguous_insert_area(struct cma * cma)58 static int dma_contiguous_insert_area(struct cma *cma)
59 {
60 if (dma_contiguous_areas_num >= ARRAY_SIZE(dma_contiguous_areas))
61 return -EINVAL;
62
63 dma_contiguous_areas[dma_contiguous_areas_num++] = cma;
64
65 return 0;
66 }
67
68 /**
69 * dma_contiguous_get_area_by_idx() - Get contiguous area at given index
70 * @idx: index of the area we query
71 *
72 * Queries for the contiguous area located at index @idx.
73 *
74 * Returns:
75 * A pointer to the requested contiguous area, or NULL otherwise.
76 */
dma_contiguous_get_area_by_idx(unsigned int idx)77 struct cma *dma_contiguous_get_area_by_idx(unsigned int idx)
78 {
79 if (idx >= dma_contiguous_areas_num)
80 return NULL;
81
82 return dma_contiguous_areas[idx];
83 }
84 EXPORT_SYMBOL_GPL(dma_contiguous_get_area_by_idx);
85
86 static struct cma *dma_contiguous_default_area;
87
88 /*
89 * Default global CMA area size can be defined in kernel's .config.
90 * This is useful mainly for distro maintainers to create a kernel
91 * that works correctly for most supported systems.
92 * The size can be set in bytes or as a percentage of the total memory
93 * in the system.
94 *
95 * Users, who want to set the size of global CMA area for their system
96 * should use cma= kernel parameter.
97 */
98 #define size_bytes ((phys_addr_t)CMA_SIZE_MBYTES * SZ_1M)
99 static phys_addr_t size_cmdline __initdata = -1;
100 static phys_addr_t base_cmdline __initdata;
101 static phys_addr_t limit_cmdline __initdata;
102
early_cma(char * p)103 static int __init early_cma(char *p)
104 {
105 if (!p) {
106 pr_err("Config string not provided\n");
107 return -EINVAL;
108 }
109
110 size_cmdline = memparse(p, &p);
111 if (*p != '@')
112 return 0;
113 base_cmdline = memparse(p + 1, &p);
114 if (*p != '-') {
115 limit_cmdline = base_cmdline + size_cmdline;
116 return 0;
117 }
118 limit_cmdline = memparse(p + 1, &p);
119
120 return 0;
121 }
122 early_param("cma", early_cma);
123
dev_get_cma_area(struct device * dev)124 struct cma *dev_get_cma_area(struct device *dev)
125 {
126 if (dev && dev->cma_area)
127 return dev->cma_area;
128
129 return dma_contiguous_default_area;
130 }
131 EXPORT_SYMBOL_GPL(dev_get_cma_area);
132
133 #ifdef CONFIG_DMA_NUMA_CMA
134
135 static struct cma *dma_contiguous_numa_area[MAX_NUMNODES];
136 static phys_addr_t numa_cma_size[MAX_NUMNODES] __initdata;
137 static phys_addr_t pernuma_size_bytes __initdata;
138 static bool numa_cma_configured __initdata;
139
early_numa_cma(char * p)140 static int __init early_numa_cma(char *p)
141 {
142 int nid, count = 0;
143 unsigned long node;
144 phys_addr_t size;
145 char *s = p;
146
147 while (*s) {
148 if (sscanf(s, "%lu%n", &node, &count) != 1)
149 break;
150
151 if (s[count] == ':') {
152 if (node >= MAX_NUMNODES)
153 break;
154 nid = array_index_nospec(node, MAX_NUMNODES);
155
156 s += count + 1;
157 size = memparse(s, &s);
158 numa_cma_size[nid] = size;
159
160 if (*s == ',')
161 s++;
162 else
163 break;
164 } else
165 break;
166 }
167
168 numa_cma_configured = true;
169 return 0;
170 }
171 early_param("numa_cma", early_numa_cma);
172
early_cma_pernuma(char * p)173 static int __init early_cma_pernuma(char *p)
174 {
175 pernuma_size_bytes = memparse(p, &p);
176 numa_cma_configured = true;
177 return 0;
178 }
179 early_param("cma_pernuma", early_cma_pernuma);
180 #endif
181
182 #ifdef CONFIG_CMA_SIZE_PERCENTAGE
183
cma_early_percent_memory(void)184 static phys_addr_t __init __maybe_unused cma_early_percent_memory(void)
185 {
186 unsigned long total_pages = PHYS_PFN(memblock_phys_mem_size());
187
188 return (total_pages * CONFIG_CMA_SIZE_PERCENTAGE / 100) << PAGE_SHIFT;
189 }
190
191 #else
192
cma_early_percent_memory(void)193 static inline __maybe_unused phys_addr_t cma_early_percent_memory(void)
194 {
195 return 0;
196 }
197
198 #endif
199
200 #ifdef CONFIG_DMA_NUMA_CMA
dma_numa_cma_reserve(void)201 static void __init dma_numa_cma_reserve(void)
202 {
203 int nid;
204
205 if (IS_ENABLED(CONFIG_CMA_SIZE_PERNUMA) &&
206 !numa_cma_configured && dma_contiguous_default_area &&
207 nr_online_nodes > 1)
208 pernuma_size_bytes = cma_get_size(dma_contiguous_default_area);
209
210 for_each_node(nid) {
211 phys_addr_t size;
212 char name[CMA_MAX_NAME];
213 struct cma **cma;
214 int ret;
215
216 if (!node_online(nid)) {
217 if (pernuma_size_bytes || numa_cma_size[nid])
218 pr_warn("invalid node %d specified\n", nid);
219 continue;
220 }
221
222 /* per-node numa setting has the priority */
223 size = numa_cma_size[nid] ?: pernuma_size_bytes;
224 if (!size)
225 continue;
226
227 cma = &dma_contiguous_numa_area[nid];
228 snprintf(name, sizeof(name), "numa%d", nid);
229 ret = cma_declare_contiguous_nid(0, size, 0, 0, 0, false, name, cma, nid);
230 if (ret)
231 pr_warn("%s: reservation failed: err %d, node %d", __func__,
232 ret, nid);
233 }
234 }
235 #else
dma_numa_cma_reserve(void)236 static inline void __init dma_numa_cma_reserve(void)
237 {
238 }
239 #endif
240
241 /**
242 * dma_contiguous_reserve() - reserve area(s) for contiguous memory handling
243 * @limit: End address of the reserved memory (optional, 0 for any).
244 *
245 * This function reserves memory from early allocator. It should be
246 * called by arch specific code once the early allocator (memblock or bootmem)
247 * has been activated and all other subsystems have already allocated/reserved
248 * memory.
249 */
dma_contiguous_reserve(phys_addr_t limit)250 void __init dma_contiguous_reserve(phys_addr_t limit)
251 {
252 phys_addr_t selected_size = 0;
253 phys_addr_t selected_base = 0;
254 phys_addr_t selected_limit = limit;
255 bool fixed = false;
256
257 pr_debug("%s(limit %08lx)\n", __func__, (unsigned long)limit);
258
259 if (size_cmdline != -1) {
260 selected_size = size_cmdline;
261 selected_base = base_cmdline;
262
263 /* Hornor the user setup dma address limit */
264 selected_limit = limit_cmdline ?: limit;
265
266 if (base_cmdline + size_cmdline == limit_cmdline)
267 fixed = true;
268 } else {
269 #ifdef CONFIG_CMA_SIZE_SEL_MBYTES
270 selected_size = size_bytes;
271 #elif defined(CONFIG_CMA_SIZE_SEL_PERCENTAGE)
272 selected_size = cma_early_percent_memory();
273 #elif defined(CONFIG_CMA_SIZE_SEL_MIN)
274 selected_size = min(size_bytes, cma_early_percent_memory());
275 #elif defined(CONFIG_CMA_SIZE_SEL_MAX)
276 selected_size = max(size_bytes, cma_early_percent_memory());
277 #endif
278 }
279
280 if (selected_size && !dma_contiguous_default_area) {
281 int ret;
282
283 pr_debug("%s: reserving %ld MiB for global area\n", __func__,
284 (unsigned long)selected_size / SZ_1M);
285
286 ret = dma_contiguous_reserve_area(selected_size, selected_base,
287 selected_limit,
288 &dma_contiguous_default_area,
289 fixed);
290 if (ret)
291 return;
292
293 /*
294 * We need to insert the new area in our list to avoid
295 * any inconsistencies between having the default area
296 * listed in the DT or not.
297 *
298 * The DT case is handled by rmem_cma_setup() and will
299 * always insert all its areas in our list. However, if
300 * it didn't run (because OF_RESERVED_MEM isn't set, or
301 * there's no DT region specified), then we don't have a
302 * default area yet, and no area in our list.
303 *
304 * This block creates the default area in such a case,
305 * but we also need to insert it in our list to avoid
306 * having a default area but an empty list.
307 */
308 ret = dma_contiguous_insert_area(dma_contiguous_default_area);
309 if (ret)
310 pr_warn("Couldn't queue default CMA region for heap creation.");
311 }
312
313 dma_numa_cma_reserve();
314 }
315
316 void __weak
dma_contiguous_early_fixup(phys_addr_t base,unsigned long size)317 dma_contiguous_early_fixup(phys_addr_t base, unsigned long size)
318 {
319 }
320
321 /**
322 * dma_contiguous_reserve_area() - reserve custom contiguous area
323 * @size: Size of the reserved area (in bytes),
324 * @base: Base address of the reserved area optional, use 0 for any
325 * @limit: End address of the reserved memory (optional, 0 for any).
326 * @res_cma: Pointer to store the created cma region.
327 * @fixed: hint about where to place the reserved area
328 *
329 * This function reserves memory from early allocator. It should be
330 * called by arch specific code once the early allocator (memblock or bootmem)
331 * has been activated and all other subsystems have already allocated/reserved
332 * memory. This function allows to create custom reserved areas for specific
333 * devices.
334 *
335 * If @fixed is true, reserve contiguous area at exactly @base. If false,
336 * reserve in range from @base to @limit.
337 */
dma_contiguous_reserve_area(phys_addr_t size,phys_addr_t base,phys_addr_t limit,struct cma ** res_cma,bool fixed)338 int __init dma_contiguous_reserve_area(phys_addr_t size, phys_addr_t base,
339 phys_addr_t limit, struct cma **res_cma,
340 bool fixed)
341 {
342 int ret;
343
344 ret = cma_declare_contiguous(base, size, limit, 0, 0, fixed,
345 "reserved", res_cma);
346 if (ret)
347 return ret;
348
349 /* Architecture specific contiguous memory fixup. */
350 dma_contiguous_early_fixup(cma_get_base(*res_cma),
351 cma_get_size(*res_cma));
352
353 return 0;
354 }
355
356 /**
357 * dma_alloc_from_contiguous() - allocate pages from contiguous area
358 * @dev: Pointer to device for which the allocation is performed.
359 * @count: Requested number of pages.
360 * @align: Requested alignment of pages (in PAGE_SIZE order).
361 * @no_warn: Avoid printing message about failed allocation.
362 *
363 * This function allocates memory buffer for specified device. It uses
364 * device specific contiguous memory area if available or the default
365 * global one. Requires architecture specific dev_get_cma_area() helper
366 * function.
367 */
dma_alloc_from_contiguous(struct device * dev,size_t count,unsigned int align,bool no_warn)368 struct page *dma_alloc_from_contiguous(struct device *dev, size_t count,
369 unsigned int align, bool no_warn)
370 {
371 if (align > CONFIG_CMA_ALIGNMENT)
372 align = CONFIG_CMA_ALIGNMENT;
373
374 return cma_alloc(dev_get_cma_area(dev), count, align, no_warn);
375 }
376
377 /**
378 * dma_release_from_contiguous() - release allocated pages
379 * @dev: Pointer to device for which the pages were allocated.
380 * @pages: Allocated pages.
381 * @count: Number of allocated pages.
382 *
383 * This function releases memory allocated by dma_alloc_from_contiguous().
384 * It returns false when provided pages do not belong to contiguous area and
385 * true otherwise.
386 */
dma_release_from_contiguous(struct device * dev,struct page * pages,int count)387 bool dma_release_from_contiguous(struct device *dev, struct page *pages,
388 int count)
389 {
390 return cma_release(dev_get_cma_area(dev), pages, count);
391 }
392
cma_alloc_aligned(struct cma * cma,size_t size,gfp_t gfp)393 static struct page *cma_alloc_aligned(struct cma *cma, size_t size, gfp_t gfp)
394 {
395 unsigned int align = min(get_order(size), CONFIG_CMA_ALIGNMENT);
396
397 return cma_alloc(cma, size >> PAGE_SHIFT, align, gfp & __GFP_NOWARN);
398 }
399
400 /**
401 * dma_alloc_contiguous() - allocate contiguous pages
402 * @dev: Pointer to device for which the allocation is performed.
403 * @size: Requested allocation size.
404 * @gfp: Allocation flags.
405 *
406 * tries to use device specific contiguous memory area if available, or it
407 * tries to use per-numa cma, if the allocation fails, it will fallback to
408 * try default global one.
409 *
410 * Note that it bypass one-page size of allocations from the per-numa and
411 * global area as the addresses within one page are always contiguous, so
412 * there is no need to waste CMA pages for that kind; it also helps reduce
413 * fragmentations.
414 */
dma_alloc_contiguous(struct device * dev,size_t size,gfp_t gfp)415 struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
416 {
417 #ifdef CONFIG_DMA_NUMA_CMA
418 int nid = dev_to_node(dev);
419 #endif
420
421 /* CMA can be used only in the context which permits sleeping */
422 if (!gfpflags_allow_blocking(gfp))
423 return NULL;
424 if (dev->cma_area)
425 return cma_alloc_aligned(dev->cma_area, size, gfp);
426 if (size <= PAGE_SIZE)
427 return NULL;
428
429 #ifdef CONFIG_DMA_NUMA_CMA
430 if (nid != NUMA_NO_NODE && !(gfp & (GFP_DMA | GFP_DMA32))) {
431 struct cma *cma = dma_contiguous_numa_area[nid];
432 struct page *page;
433 if (cma) {
434 page = cma_alloc_aligned(cma, size, gfp);
435 if (page)
436 return page;
437 }
438 }
439 #endif
440 if (!dma_contiguous_default_area)
441 return NULL;
442
443 return cma_alloc_aligned(dma_contiguous_default_area, size, gfp);
444 }
445
446 /**
447 * dma_free_contiguous() - release allocated pages
448 * @dev: Pointer to device for which the pages were allocated.
449 * @page: Pointer to the allocated pages.
450 * @size: Size of allocated pages.
451 *
452 * This function releases memory allocated by dma_alloc_contiguous(). As the
453 * cma_release returns false when provided pages do not belong to contiguous
454 * area and true otherwise, this function then does a fallback __free_pages()
455 * upon a false-return.
456 */
dma_free_contiguous(struct device * dev,struct page * page,size_t size)457 void dma_free_contiguous(struct device *dev, struct page *page, size_t size)
458 {
459 unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT;
460
461 /* if dev has its own cma, free page from there */
462 if (dev->cma_area) {
463 if (cma_release(dev->cma_area, page, count))
464 return;
465 } else {
466 /*
467 * otherwise, page is from either per-numa cma or default cma
468 */
469 #ifdef CONFIG_DMA_NUMA_CMA
470 if (cma_release(dma_contiguous_numa_area[page_to_nid(page)],
471 page, count))
472 return;
473 #endif
474 if (cma_release(dma_contiguous_default_area, page, count))
475 return;
476 }
477
478 /* not in any cma, free from buddy */
479 __free_pages(page, get_order(size));
480 }
481
482 /*
483 * Support for reserved memory regions defined in device tree
484 */
485 #ifdef CONFIG_OF_RESERVED_MEM
486 #include <linux/of.h>
487 #include <linux/of_fdt.h>
488 #include <linux/of_reserved_mem.h>
489
490 #undef pr_fmt
491 #define pr_fmt(fmt) fmt
492
rmem_cma_device_init(struct reserved_mem * rmem,struct device * dev)493 static int rmem_cma_device_init(struct reserved_mem *rmem, struct device *dev)
494 {
495 dev->cma_area = rmem->priv;
496 return 0;
497 }
498
rmem_cma_device_release(struct reserved_mem * rmem,struct device * dev)499 static void rmem_cma_device_release(struct reserved_mem *rmem,
500 struct device *dev)
501 {
502 dev->cma_area = NULL;
503 }
504
__rmem_cma_verify_node(unsigned long node)505 static int __init __rmem_cma_verify_node(unsigned long node)
506 {
507 if (!of_get_flat_dt_prop(node, "reusable", NULL) ||
508 of_get_flat_dt_prop(node, "no-map", NULL))
509 return -ENODEV;
510
511 if (size_cmdline != -1 &&
512 of_get_flat_dt_prop(node, "linux,cma-default", NULL)) {
513 pr_err("Skipping dt linux,cma-default node in favor for \"cma=\" kernel param.\n");
514 return -EBUSY;
515 }
516 return 0;
517 }
518
rmem_cma_validate(unsigned long node,phys_addr_t * align)519 static int __init rmem_cma_validate(unsigned long node, phys_addr_t *align)
520 {
521 int ret = __rmem_cma_verify_node(node);
522
523 if (ret)
524 return ret;
525
526 if (align)
527 *align = max_t(phys_addr_t, *align, CMA_MIN_ALIGNMENT_BYTES);
528
529 return 0;
530 }
531
rmem_cma_fixup(unsigned long node,phys_addr_t base,phys_addr_t size)532 static int __init rmem_cma_fixup(unsigned long node, phys_addr_t base,
533 phys_addr_t size)
534 {
535 int ret = __rmem_cma_verify_node(node);
536
537 if (ret)
538 return ret;
539
540 /* Architecture specific contiguous memory fixup. */
541 dma_contiguous_early_fixup(base, size);
542 return 0;
543 }
544
rmem_cma_setup(unsigned long node,struct reserved_mem * rmem)545 static int __init rmem_cma_setup(unsigned long node, struct reserved_mem *rmem)
546 {
547 bool default_cma = of_get_flat_dt_prop(node, "linux,cma-default", NULL);
548 struct cma *cma;
549 int ret;
550
551 ret = __rmem_cma_verify_node(node);
552 if (ret)
553 return ret;
554
555 if (!IS_ALIGNED(rmem->base | rmem->size, CMA_MIN_ALIGNMENT_BYTES)) {
556 pr_err("Reserved memory: incorrect alignment of CMA region\n");
557 return -EINVAL;
558 }
559
560 ret = cma_init_reserved_mem(rmem->base, rmem->size, 0, rmem->name, &cma);
561 if (ret) {
562 pr_err("Reserved memory: unable to setup CMA region\n");
563 return ret;
564 }
565
566 if (default_cma)
567 dma_contiguous_default_area = cma;
568
569 rmem->priv = cma;
570
571 pr_info("Reserved memory: created CMA memory pool at %pa, size %ld MiB\n",
572 &rmem->base, (unsigned long)rmem->size / SZ_1M);
573
574 ret = dma_contiguous_insert_area(cma);
575 if (ret)
576 pr_warn("Couldn't store CMA reserved area.");
577
578 return 0;
579 }
580
581 static const struct reserved_mem_ops rmem_cma_ops = {
582 .node_validate = rmem_cma_validate,
583 .node_fixup = rmem_cma_fixup,
584 .node_init = rmem_cma_setup,
585 .device_init = rmem_cma_device_init,
586 .device_release = rmem_cma_device_release,
587 };
588
589 RESERVEDMEM_OF_DECLARE(cma, "shared-dma-pool", &rmem_cma_ops);
590 #endif
591