Commit | Line | Data |
---|---|---|
09b55412 CM |
1 | /* |
2 | * SWIOTLB-based DMA API implementation | |
3 | * | |
4 | * Copyright (C) 2012 ARM Ltd. | |
5 | * Author: Catalin Marinas <catalin.marinas@arm.com> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | |
18 | */ | |
19 | ||
20 | #include <linux/gfp.h> | |
21 | #include <linux/export.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/dma-mapping.h> | |
6ac2104d | 24 | #include <linux/dma-contiguous.h> |
09b55412 CM |
25 | #include <linux/vmalloc.h> |
26 | #include <linux/swiotlb.h> | |
27 | ||
28 | #include <asm/cacheflush.h> | |
29 | ||
30 | struct dma_map_ops *dma_ops; | |
31 | EXPORT_SYMBOL(dma_ops); | |
32 | ||
bb10eb7b RH |
33 | static void *__dma_alloc_coherent(struct device *dev, size_t size, |
34 | dma_addr_t *dma_handle, gfp_t flags, | |
35 | struct dma_attrs *attrs) | |
09b55412 | 36 | { |
c666e8d5 LA |
37 | if (dev == NULL) { |
38 | WARN_ONCE(1, "Use an actual device structure for DMA allocation\n"); | |
39 | return NULL; | |
40 | } | |
41 | ||
19e7640d | 42 | if (IS_ENABLED(CONFIG_ZONE_DMA) && |
09b55412 | 43 | dev->coherent_dma_mask <= DMA_BIT_MASK(32)) |
19e7640d | 44 | flags |= GFP_DMA; |
6ac2104d LA |
45 | if (IS_ENABLED(CONFIG_DMA_CMA)) { |
46 | struct page *page; | |
47 | ||
ccc9e244 | 48 | size = PAGE_ALIGN(size); |
6ac2104d LA |
49 | page = dma_alloc_from_contiguous(dev, size >> PAGE_SHIFT, |
50 | get_order(size)); | |
51 | if (!page) | |
52 | return NULL; | |
53 | ||
54 | *dma_handle = phys_to_dma(dev, page_to_phys(page)); | |
55 | return page_address(page); | |
56 | } else { | |
57 | return swiotlb_alloc_coherent(dev, size, dma_handle, flags); | |
58 | } | |
09b55412 CM |
59 | } |
60 | ||
bb10eb7b RH |
61 | static void __dma_free_coherent(struct device *dev, size_t size, |
62 | void *vaddr, dma_addr_t dma_handle, | |
63 | struct dma_attrs *attrs) | |
09b55412 | 64 | { |
c666e8d5 LA |
65 | if (dev == NULL) { |
66 | WARN_ONCE(1, "Use an actual device structure for DMA allocation\n"); | |
67 | return; | |
68 | } | |
69 | ||
6ac2104d LA |
70 | if (IS_ENABLED(CONFIG_DMA_CMA)) { |
71 | phys_addr_t paddr = dma_to_phys(dev, dma_handle); | |
72 | ||
73 | dma_release_from_contiguous(dev, | |
74 | phys_to_page(paddr), | |
75 | size >> PAGE_SHIFT); | |
76 | } else { | |
77 | swiotlb_free_coherent(dev, size, vaddr, dma_handle); | |
78 | } | |
09b55412 CM |
79 | } |
80 | ||
7363590d CM |
81 | static void *__dma_alloc_noncoherent(struct device *dev, size_t size, |
82 | dma_addr_t *dma_handle, gfp_t flags, | |
83 | struct dma_attrs *attrs) | |
84 | { | |
85 | struct page *page, **map; | |
86 | void *ptr, *coherent_ptr; | |
87 | int order, i; | |
88 | ||
89 | size = PAGE_ALIGN(size); | |
90 | order = get_order(size); | |
91 | ||
92 | ptr = __dma_alloc_coherent(dev, size, dma_handle, flags, attrs); | |
93 | if (!ptr) | |
94 | goto no_mem; | |
95 | map = kmalloc(sizeof(struct page *) << order, flags & ~GFP_DMA); | |
96 | if (!map) | |
97 | goto no_map; | |
98 | ||
99 | /* remove any dirty cache lines on the kernel alias */ | |
100 | __dma_flush_range(ptr, ptr + size); | |
101 | ||
102 | /* create a coherent mapping */ | |
103 | page = virt_to_page(ptr); | |
104 | for (i = 0; i < (size >> PAGE_SHIFT); i++) | |
105 | map[i] = page + i; | |
106 | coherent_ptr = vmap(map, size >> PAGE_SHIFT, VM_MAP, | |
107 | pgprot_dmacoherent(pgprot_default)); | |
108 | kfree(map); | |
109 | if (!coherent_ptr) | |
110 | goto no_map; | |
111 | ||
112 | return coherent_ptr; | |
113 | ||
114 | no_map: | |
115 | __dma_free_coherent(dev, size, ptr, *dma_handle, attrs); | |
116 | no_mem: | |
117 | *dma_handle = ~0; | |
118 | return NULL; | |
119 | } | |
120 | ||
121 | static void __dma_free_noncoherent(struct device *dev, size_t size, | |
122 | void *vaddr, dma_addr_t dma_handle, | |
123 | struct dma_attrs *attrs) | |
124 | { | |
125 | void *swiotlb_addr = phys_to_virt(dma_to_phys(dev, dma_handle)); | |
126 | ||
127 | vunmap(vaddr); | |
128 | __dma_free_coherent(dev, size, swiotlb_addr, dma_handle, attrs); | |
129 | } | |
130 | ||
131 | static dma_addr_t __swiotlb_map_page(struct device *dev, struct page *page, | |
132 | unsigned long offset, size_t size, | |
133 | enum dma_data_direction dir, | |
134 | struct dma_attrs *attrs) | |
135 | { | |
136 | dma_addr_t dev_addr; | |
137 | ||
138 | dev_addr = swiotlb_map_page(dev, page, offset, size, dir, attrs); | |
139 | __dma_map_area(phys_to_virt(dma_to_phys(dev, dev_addr)), size, dir); | |
140 | ||
141 | return dev_addr; | |
142 | } | |
143 | ||
144 | ||
145 | static void __swiotlb_unmap_page(struct device *dev, dma_addr_t dev_addr, | |
146 | size_t size, enum dma_data_direction dir, | |
147 | struct dma_attrs *attrs) | |
148 | { | |
149 | __dma_unmap_area(phys_to_virt(dma_to_phys(dev, dev_addr)), size, dir); | |
150 | swiotlb_unmap_page(dev, dev_addr, size, dir, attrs); | |
151 | } | |
152 | ||
153 | static int __swiotlb_map_sg_attrs(struct device *dev, struct scatterlist *sgl, | |
154 | int nelems, enum dma_data_direction dir, | |
155 | struct dma_attrs *attrs) | |
156 | { | |
157 | struct scatterlist *sg; | |
158 | int i, ret; | |
159 | ||
160 | ret = swiotlb_map_sg_attrs(dev, sgl, nelems, dir, attrs); | |
161 | for_each_sg(sgl, sg, ret, i) | |
162 | __dma_map_area(phys_to_virt(dma_to_phys(dev, sg->dma_address)), | |
163 | sg->length, dir); | |
164 | ||
165 | return ret; | |
166 | } | |
167 | ||
168 | static void __swiotlb_unmap_sg_attrs(struct device *dev, | |
169 | struct scatterlist *sgl, int nelems, | |
170 | enum dma_data_direction dir, | |
171 | struct dma_attrs *attrs) | |
172 | { | |
173 | struct scatterlist *sg; | |
174 | int i; | |
175 | ||
176 | for_each_sg(sgl, sg, nelems, i) | |
177 | __dma_unmap_area(phys_to_virt(dma_to_phys(dev, sg->dma_address)), | |
178 | sg->length, dir); | |
179 | swiotlb_unmap_sg_attrs(dev, sgl, nelems, dir, attrs); | |
180 | } | |
181 | ||
182 | static void __swiotlb_sync_single_for_cpu(struct device *dev, | |
183 | dma_addr_t dev_addr, size_t size, | |
184 | enum dma_data_direction dir) | |
185 | { | |
186 | __dma_unmap_area(phys_to_virt(dma_to_phys(dev, dev_addr)), size, dir); | |
187 | swiotlb_sync_single_for_cpu(dev, dev_addr, size, dir); | |
188 | } | |
189 | ||
190 | static void __swiotlb_sync_single_for_device(struct device *dev, | |
191 | dma_addr_t dev_addr, size_t size, | |
192 | enum dma_data_direction dir) | |
193 | { | |
194 | swiotlb_sync_single_for_device(dev, dev_addr, size, dir); | |
195 | __dma_map_area(phys_to_virt(dma_to_phys(dev, dev_addr)), size, dir); | |
196 | } | |
197 | ||
198 | static void __swiotlb_sync_sg_for_cpu(struct device *dev, | |
199 | struct scatterlist *sgl, int nelems, | |
200 | enum dma_data_direction dir) | |
201 | { | |
202 | struct scatterlist *sg; | |
203 | int i; | |
204 | ||
205 | for_each_sg(sgl, sg, nelems, i) | |
206 | __dma_unmap_area(phys_to_virt(dma_to_phys(dev, sg->dma_address)), | |
207 | sg->length, dir); | |
208 | swiotlb_sync_sg_for_cpu(dev, sgl, nelems, dir); | |
209 | } | |
210 | ||
211 | static void __swiotlb_sync_sg_for_device(struct device *dev, | |
212 | struct scatterlist *sgl, int nelems, | |
213 | enum dma_data_direction dir) | |
214 | { | |
215 | struct scatterlist *sg; | |
216 | int i; | |
217 | ||
218 | swiotlb_sync_sg_for_device(dev, sgl, nelems, dir); | |
219 | for_each_sg(sgl, sg, nelems, i) | |
220 | __dma_map_area(phys_to_virt(dma_to_phys(dev, sg->dma_address)), | |
221 | sg->length, dir); | |
222 | } | |
223 | ||
6e8d7968 LA |
224 | /* vma->vm_page_prot must be set appropriately before calling this function */ |
225 | static int __dma_common_mmap(struct device *dev, struct vm_area_struct *vma, | |
226 | void *cpu_addr, dma_addr_t dma_addr, size_t size) | |
227 | { | |
228 | int ret = -ENXIO; | |
229 | unsigned long nr_vma_pages = (vma->vm_end - vma->vm_start) >> | |
230 | PAGE_SHIFT; | |
231 | unsigned long nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT; | |
232 | unsigned long pfn = dma_to_phys(dev, dma_addr) >> PAGE_SHIFT; | |
233 | unsigned long off = vma->vm_pgoff; | |
234 | ||
235 | if (dma_mmap_from_coherent(dev, vma, cpu_addr, size, &ret)) | |
236 | return ret; | |
237 | ||
238 | if (off < nr_pages && nr_vma_pages <= (nr_pages - off)) { | |
239 | ret = remap_pfn_range(vma, vma->vm_start, | |
240 | pfn + off, | |
241 | vma->vm_end - vma->vm_start, | |
242 | vma->vm_page_prot); | |
243 | } | |
244 | ||
245 | return ret; | |
246 | } | |
247 | ||
248 | static int __swiotlb_mmap_noncoherent(struct device *dev, | |
249 | struct vm_area_struct *vma, | |
250 | void *cpu_addr, dma_addr_t dma_addr, size_t size, | |
251 | struct dma_attrs *attrs) | |
252 | { | |
253 | vma->vm_page_prot = pgprot_dmacoherent(vma->vm_page_prot); | |
254 | return __dma_common_mmap(dev, vma, cpu_addr, dma_addr, size); | |
255 | } | |
256 | ||
257 | static int __swiotlb_mmap_coherent(struct device *dev, | |
258 | struct vm_area_struct *vma, | |
259 | void *cpu_addr, dma_addr_t dma_addr, size_t size, | |
260 | struct dma_attrs *attrs) | |
261 | { | |
262 | /* Just use whatever page_prot attributes were specified */ | |
263 | return __dma_common_mmap(dev, vma, cpu_addr, dma_addr, size); | |
264 | } | |
265 | ||
7363590d CM |
266 | struct dma_map_ops noncoherent_swiotlb_dma_ops = { |
267 | .alloc = __dma_alloc_noncoherent, | |
268 | .free = __dma_free_noncoherent, | |
6e8d7968 | 269 | .mmap = __swiotlb_mmap_noncoherent, |
7363590d CM |
270 | .map_page = __swiotlb_map_page, |
271 | .unmap_page = __swiotlb_unmap_page, | |
272 | .map_sg = __swiotlb_map_sg_attrs, | |
273 | .unmap_sg = __swiotlb_unmap_sg_attrs, | |
274 | .sync_single_for_cpu = __swiotlb_sync_single_for_cpu, | |
275 | .sync_single_for_device = __swiotlb_sync_single_for_device, | |
276 | .sync_sg_for_cpu = __swiotlb_sync_sg_for_cpu, | |
277 | .sync_sg_for_device = __swiotlb_sync_sg_for_device, | |
278 | .dma_supported = swiotlb_dma_supported, | |
279 | .mapping_error = swiotlb_dma_mapping_error, | |
280 | }; | |
281 | EXPORT_SYMBOL(noncoherent_swiotlb_dma_ops); | |
282 | ||
283 | struct dma_map_ops coherent_swiotlb_dma_ops = { | |
bb10eb7b RH |
284 | .alloc = __dma_alloc_coherent, |
285 | .free = __dma_free_coherent, | |
6e8d7968 | 286 | .mmap = __swiotlb_mmap_coherent, |
09b55412 CM |
287 | .map_page = swiotlb_map_page, |
288 | .unmap_page = swiotlb_unmap_page, | |
289 | .map_sg = swiotlb_map_sg_attrs, | |
290 | .unmap_sg = swiotlb_unmap_sg_attrs, | |
291 | .sync_single_for_cpu = swiotlb_sync_single_for_cpu, | |
292 | .sync_single_for_device = swiotlb_sync_single_for_device, | |
293 | .sync_sg_for_cpu = swiotlb_sync_sg_for_cpu, | |
294 | .sync_sg_for_device = swiotlb_sync_sg_for_device, | |
295 | .dma_supported = swiotlb_dma_supported, | |
296 | .mapping_error = swiotlb_dma_mapping_error, | |
297 | }; | |
7363590d | 298 | EXPORT_SYMBOL(coherent_swiotlb_dma_ops); |
09b55412 | 299 | |
3690951f CM |
300 | extern int swiotlb_late_init_with_default_size(size_t default_size); |
301 | ||
302 | static int __init swiotlb_late_init(void) | |
09b55412 | 303 | { |
3690951f CM |
304 | size_t swiotlb_size = min(SZ_64M, MAX_ORDER_NR_PAGES << PAGE_SHIFT); |
305 | ||
bb10eb7b | 306 | dma_ops = &coherent_swiotlb_dma_ops; |
3690951f CM |
307 | |
308 | return swiotlb_late_init_with_default_size(swiotlb_size); | |
09b55412 | 309 | } |
3690951f | 310 | subsys_initcall(swiotlb_late_init); |
09b55412 CM |
311 | |
312 | #define PREALLOC_DMA_DEBUG_ENTRIES 4096 | |
313 | ||
314 | static int __init dma_debug_do_init(void) | |
315 | { | |
316 | dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES); | |
317 | return 0; | |
318 | } | |
319 | fs_initcall(dma_debug_do_init); |