xref: /openbmc/linux/kernel/dma/contiguous.c (revision e0d07278)
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 
10 #define pr_fmt(fmt) "cma: " fmt
11 
12 #ifdef CONFIG_CMA_DEBUG
13 #ifndef DEBUG
14 #  define DEBUG
15 #endif
16 #endif
17 
18 #include <asm/page.h>
19 #include <asm/dma-contiguous.h>
20 
21 #include <linux/memblock.h>
22 #include <linux/err.h>
23 #include <linux/sizes.h>
24 #include <linux/dma-contiguous.h>
25 #include <linux/cma.h>
26 
27 #ifdef CONFIG_CMA_SIZE_MBYTES
28 #define CMA_SIZE_MBYTES CONFIG_CMA_SIZE_MBYTES
29 #else
30 #define CMA_SIZE_MBYTES 0
31 #endif
32 
33 struct cma *dma_contiguous_default_area;
34 
35 /*
36  * Default global CMA area size can be defined in kernel's .config.
37  * This is useful mainly for distro maintainers to create a kernel
38  * that works correctly for most supported systems.
39  * The size can be set in bytes or as a percentage of the total memory
40  * in the system.
41  *
42  * Users, who want to set the size of global CMA area for their system
43  * should use cma= kernel parameter.
44  */
45 static const phys_addr_t size_bytes __initconst =
46 	(phys_addr_t)CMA_SIZE_MBYTES * SZ_1M;
47 static phys_addr_t  size_cmdline __initdata = -1;
48 static phys_addr_t base_cmdline __initdata;
49 static phys_addr_t limit_cmdline __initdata;
50 
51 static int __init early_cma(char *p)
52 {
53 	if (!p) {
54 		pr_err("Config string not provided\n");
55 		return -EINVAL;
56 	}
57 
58 	size_cmdline = memparse(p, &p);
59 	if (*p != '@')
60 		return 0;
61 	base_cmdline = memparse(p + 1, &p);
62 	if (*p != '-') {
63 		limit_cmdline = base_cmdline + size_cmdline;
64 		return 0;
65 	}
66 	limit_cmdline = memparse(p + 1, &p);
67 
68 	return 0;
69 }
70 early_param("cma", early_cma);
71 
72 #ifdef CONFIG_DMA_PERNUMA_CMA
73 
74 static struct cma *dma_contiguous_pernuma_area[MAX_NUMNODES];
75 static phys_addr_t pernuma_size_bytes __initdata;
76 
77 static int __init early_cma_pernuma(char *p)
78 {
79 	pernuma_size_bytes = memparse(p, &p);
80 	return 0;
81 }
82 early_param("cma_pernuma", early_cma_pernuma);
83 #endif
84 
85 #ifdef CONFIG_CMA_SIZE_PERCENTAGE
86 
87 static phys_addr_t __init __maybe_unused cma_early_percent_memory(void)
88 {
89 	struct memblock_region *reg;
90 	unsigned long total_pages = 0;
91 
92 	/*
93 	 * We cannot use memblock_phys_mem_size() here, because
94 	 * memblock_analyze() has not been called yet.
95 	 */
96 	for_each_memblock(memory, reg)
97 		total_pages += memblock_region_memory_end_pfn(reg) -
98 			       memblock_region_memory_base_pfn(reg);
99 
100 	return (total_pages * CONFIG_CMA_SIZE_PERCENTAGE / 100) << PAGE_SHIFT;
101 }
102 
103 #else
104 
105 static inline __maybe_unused phys_addr_t cma_early_percent_memory(void)
106 {
107 	return 0;
108 }
109 
110 #endif
111 
112 #ifdef CONFIG_DMA_PERNUMA_CMA
113 void __init dma_pernuma_cma_reserve(void)
114 {
115 	int nid;
116 
117 	if (!pernuma_size_bytes)
118 		return;
119 
120 	for_each_online_node(nid) {
121 		int ret;
122 		char name[CMA_MAX_NAME];
123 		struct cma **cma = &dma_contiguous_pernuma_area[nid];
124 
125 		snprintf(name, sizeof(name), "pernuma%d", nid);
126 		ret = cma_declare_contiguous_nid(0, pernuma_size_bytes, 0, 0,
127 						 0, false, name, cma, nid);
128 		if (ret) {
129 			pr_warn("%s: reservation failed: err %d, node %d", __func__,
130 				ret, nid);
131 			continue;
132 		}
133 
134 		pr_debug("%s: reserved %llu MiB on node %d\n", __func__,
135 			(unsigned long long)pernuma_size_bytes / SZ_1M, nid);
136 	}
137 }
138 #endif
139 
140 /**
141  * dma_contiguous_reserve() - reserve area(s) for contiguous memory handling
142  * @limit: End address of the reserved memory (optional, 0 for any).
143  *
144  * This function reserves memory from early allocator. It should be
145  * called by arch specific code once the early allocator (memblock or bootmem)
146  * has been activated and all other subsystems have already allocated/reserved
147  * memory.
148  */
149 void __init dma_contiguous_reserve(phys_addr_t limit)
150 {
151 	phys_addr_t selected_size = 0;
152 	phys_addr_t selected_base = 0;
153 	phys_addr_t selected_limit = limit;
154 	bool fixed = false;
155 
156 	pr_debug("%s(limit %08lx)\n", __func__, (unsigned long)limit);
157 
158 	if (size_cmdline != -1) {
159 		selected_size = size_cmdline;
160 		selected_base = base_cmdline;
161 		selected_limit = min_not_zero(limit_cmdline, limit);
162 		if (base_cmdline + size_cmdline == limit_cmdline)
163 			fixed = true;
164 	} else {
165 #ifdef CONFIG_CMA_SIZE_SEL_MBYTES
166 		selected_size = size_bytes;
167 #elif defined(CONFIG_CMA_SIZE_SEL_PERCENTAGE)
168 		selected_size = cma_early_percent_memory();
169 #elif defined(CONFIG_CMA_SIZE_SEL_MIN)
170 		selected_size = min(size_bytes, cma_early_percent_memory());
171 #elif defined(CONFIG_CMA_SIZE_SEL_MAX)
172 		selected_size = max(size_bytes, cma_early_percent_memory());
173 #endif
174 	}
175 
176 	if (selected_size && !dma_contiguous_default_area) {
177 		pr_debug("%s: reserving %ld MiB for global area\n", __func__,
178 			 (unsigned long)selected_size / SZ_1M);
179 
180 		dma_contiguous_reserve_area(selected_size, selected_base,
181 					    selected_limit,
182 					    &dma_contiguous_default_area,
183 					    fixed);
184 	}
185 }
186 
187 /**
188  * dma_contiguous_reserve_area() - reserve custom contiguous area
189  * @size: Size of the reserved area (in bytes),
190  * @base: Base address of the reserved area optional, use 0 for any
191  * @limit: End address of the reserved memory (optional, 0 for any).
192  * @res_cma: Pointer to store the created cma region.
193  * @fixed: hint about where to place the reserved area
194  *
195  * This function reserves memory from early allocator. It should be
196  * called by arch specific code once the early allocator (memblock or bootmem)
197  * has been activated and all other subsystems have already allocated/reserved
198  * memory. This function allows to create custom reserved areas for specific
199  * devices.
200  *
201  * If @fixed is true, reserve contiguous area at exactly @base.  If false,
202  * reserve in range from @base to @limit.
203  */
204 int __init dma_contiguous_reserve_area(phys_addr_t size, phys_addr_t base,
205 				       phys_addr_t limit, struct cma **res_cma,
206 				       bool fixed)
207 {
208 	int ret;
209 
210 	ret = cma_declare_contiguous(base, size, limit, 0, 0, fixed,
211 					"reserved", res_cma);
212 	if (ret)
213 		return ret;
214 
215 	/* Architecture specific contiguous memory fixup. */
216 	dma_contiguous_early_fixup(cma_get_base(*res_cma),
217 				cma_get_size(*res_cma));
218 
219 	return 0;
220 }
221 
222 /**
223  * dma_alloc_from_contiguous() - allocate pages from contiguous area
224  * @dev:   Pointer to device for which the allocation is performed.
225  * @count: Requested number of pages.
226  * @align: Requested alignment of pages (in PAGE_SIZE order).
227  * @no_warn: Avoid printing message about failed allocation.
228  *
229  * This function allocates memory buffer for specified device. It uses
230  * device specific contiguous memory area if available or the default
231  * global one. Requires architecture specific dev_get_cma_area() helper
232  * function.
233  */
234 struct page *dma_alloc_from_contiguous(struct device *dev, size_t count,
235 				       unsigned int align, bool no_warn)
236 {
237 	if (align > CONFIG_CMA_ALIGNMENT)
238 		align = CONFIG_CMA_ALIGNMENT;
239 
240 	return cma_alloc(dev_get_cma_area(dev), count, align, no_warn);
241 }
242 
243 /**
244  * dma_release_from_contiguous() - release allocated pages
245  * @dev:   Pointer to device for which the pages were allocated.
246  * @pages: Allocated pages.
247  * @count: Number of allocated pages.
248  *
249  * This function releases memory allocated by dma_alloc_from_contiguous().
250  * It returns false when provided pages do not belong to contiguous area and
251  * true otherwise.
252  */
253 bool dma_release_from_contiguous(struct device *dev, struct page *pages,
254 				 int count)
255 {
256 	return cma_release(dev_get_cma_area(dev), pages, count);
257 }
258 
259 static struct page *cma_alloc_aligned(struct cma *cma, size_t size, gfp_t gfp)
260 {
261 	unsigned int align = min(get_order(size), CONFIG_CMA_ALIGNMENT);
262 
263 	return cma_alloc(cma, size >> PAGE_SHIFT, align, gfp & __GFP_NOWARN);
264 }
265 
266 /**
267  * dma_alloc_contiguous() - allocate contiguous pages
268  * @dev:   Pointer to device for which the allocation is performed.
269  * @size:  Requested allocation size.
270  * @gfp:   Allocation flags.
271  *
272  * tries to use device specific contiguous memory area if available, or it
273  * tries to use per-numa cma, if the allocation fails, it will fallback to
274  * try default global one.
275  *
276  * Note that it bypass one-page size of allocations from the per-numa and
277  * global area as the addresses within one page are always contiguous, so
278  * there is no need to waste CMA pages for that kind; it also helps reduce
279  * fragmentations.
280  */
281 struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
282 {
283 #ifdef CONFIG_DMA_PERNUMA_CMA
284 	int nid = dev_to_node(dev);
285 #endif
286 
287 	/* CMA can be used only in the context which permits sleeping */
288 	if (!gfpflags_allow_blocking(gfp))
289 		return NULL;
290 	if (dev->cma_area)
291 		return cma_alloc_aligned(dev->cma_area, size, gfp);
292 	if (size <= PAGE_SIZE)
293 		return NULL;
294 
295 #ifdef CONFIG_DMA_PERNUMA_CMA
296 	if (nid != NUMA_NO_NODE && !(gfp & (GFP_DMA | GFP_DMA32))) {
297 		struct cma *cma = dma_contiguous_pernuma_area[nid];
298 		struct page *page;
299 
300 		if (cma) {
301 			page = cma_alloc_aligned(cma, size, gfp);
302 			if (page)
303 				return page;
304 		}
305 	}
306 #endif
307 	if (!dma_contiguous_default_area)
308 		return NULL;
309 
310 	return cma_alloc_aligned(dma_contiguous_default_area, size, gfp);
311 }
312 
313 /**
314  * dma_free_contiguous() - release allocated pages
315  * @dev:   Pointer to device for which the pages were allocated.
316  * @page:  Pointer to the allocated pages.
317  * @size:  Size of allocated pages.
318  *
319  * This function releases memory allocated by dma_alloc_contiguous(). As the
320  * cma_release returns false when provided pages do not belong to contiguous
321  * area and true otherwise, this function then does a fallback __free_pages()
322  * upon a false-return.
323  */
324 void dma_free_contiguous(struct device *dev, struct page *page, size_t size)
325 {
326 	unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT;
327 
328 	/* if dev has its own cma, free page from there */
329 	if (dev->cma_area) {
330 		if (cma_release(dev->cma_area, page, count))
331 			return;
332 	} else {
333 		/*
334 		 * otherwise, page is from either per-numa cma or default cma
335 		 */
336 #ifdef CONFIG_DMA_PERNUMA_CMA
337 		if (cma_release(dma_contiguous_pernuma_area[page_to_nid(page)],
338 					page, count))
339 			return;
340 #endif
341 		if (cma_release(dma_contiguous_default_area, page, count))
342 			return;
343 	}
344 
345 	/* not in any cma, free from buddy */
346 	__free_pages(page, get_order(size));
347 }
348 
349 /*
350  * Support for reserved memory regions defined in device tree
351  */
352 #ifdef CONFIG_OF_RESERVED_MEM
353 #include <linux/of.h>
354 #include <linux/of_fdt.h>
355 #include <linux/of_reserved_mem.h>
356 
357 #undef pr_fmt
358 #define pr_fmt(fmt) fmt
359 
360 static int rmem_cma_device_init(struct reserved_mem *rmem, struct device *dev)
361 {
362 	dev_set_cma_area(dev, rmem->priv);
363 	return 0;
364 }
365 
366 static void rmem_cma_device_release(struct reserved_mem *rmem,
367 				    struct device *dev)
368 {
369 	dev_set_cma_area(dev, NULL);
370 }
371 
372 static const struct reserved_mem_ops rmem_cma_ops = {
373 	.device_init	= rmem_cma_device_init,
374 	.device_release = rmem_cma_device_release,
375 };
376 
377 static int __init rmem_cma_setup(struct reserved_mem *rmem)
378 {
379 	phys_addr_t align = PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order);
380 	phys_addr_t mask = align - 1;
381 	unsigned long node = rmem->fdt_node;
382 	bool default_cma = of_get_flat_dt_prop(node, "linux,cma-default", NULL);
383 	struct cma *cma;
384 	int err;
385 
386 	if (size_cmdline != -1 && default_cma) {
387 		pr_info("Reserved memory: bypass %s node, using cmdline CMA params instead\n",
388 			rmem->name);
389 		return -EBUSY;
390 	}
391 
392 	if (!of_get_flat_dt_prop(node, "reusable", NULL) ||
393 	    of_get_flat_dt_prop(node, "no-map", NULL))
394 		return -EINVAL;
395 
396 	if ((rmem->base & mask) || (rmem->size & mask)) {
397 		pr_err("Reserved memory: incorrect alignment of CMA region\n");
398 		return -EINVAL;
399 	}
400 
401 	err = cma_init_reserved_mem(rmem->base, rmem->size, 0, rmem->name, &cma);
402 	if (err) {
403 		pr_err("Reserved memory: unable to setup CMA region\n");
404 		return err;
405 	}
406 	/* Architecture specific contiguous memory fixup. */
407 	dma_contiguous_early_fixup(rmem->base, rmem->size);
408 
409 	if (default_cma)
410 		dma_contiguous_set_default(cma);
411 
412 	rmem->ops = &rmem_cma_ops;
413 	rmem->priv = cma;
414 
415 	pr_info("Reserved memory: created CMA memory pool at %pa, size %ld MiB\n",
416 		&rmem->base, (unsigned long)rmem->size / SZ_1M);
417 
418 	return 0;
419 }
420 RESERVEDMEM_OF_DECLARE(cma, "shared-dma-pool", rmem_cma_setup);
421 #endif
422