]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
0ddbccd1 | 2 | * linux/arch/arm/mm/dma-mapping.c |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 2000-2004 Russell King | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 as | |
8 | * published by the Free Software Foundation. | |
9 | * | |
10 | * DMA uncached mapping support. | |
11 | */ | |
12 | #include <linux/module.h> | |
13 | #include <linux/mm.h> | |
5a0e3ad6 | 14 | #include <linux/gfp.h> |
1da177e4 LT |
15 | #include <linux/errno.h> |
16 | #include <linux/list.h> | |
17 | #include <linux/init.h> | |
18 | #include <linux/device.h> | |
19 | #include <linux/dma-mapping.h> | |
c7909509 | 20 | #include <linux/dma-contiguous.h> |
39af22a7 | 21 | #include <linux/highmem.h> |
c7909509 | 22 | #include <linux/memblock.h> |
99d1717d | 23 | #include <linux/slab.h> |
4ce63fcd | 24 | #include <linux/iommu.h> |
e9da6e99 | 25 | #include <linux/io.h> |
4ce63fcd | 26 | #include <linux/vmalloc.h> |
158e8bfe | 27 | #include <linux/sizes.h> |
1da177e4 | 28 | |
23759dc6 | 29 | #include <asm/memory.h> |
43377453 | 30 | #include <asm/highmem.h> |
1da177e4 | 31 | #include <asm/cacheflush.h> |
1da177e4 | 32 | #include <asm/tlbflush.h> |
99d1717d | 33 | #include <asm/mach/arch.h> |
4ce63fcd | 34 | #include <asm/dma-iommu.h> |
c7909509 MS |
35 | #include <asm/mach/map.h> |
36 | #include <asm/system_info.h> | |
37 | #include <asm/dma-contiguous.h> | |
37134cd5 | 38 | |
022ae537 RK |
39 | #include "mm.h" |
40 | ||
15237e1f MS |
41 | /* |
42 | * The DMA API is built upon the notion of "buffer ownership". A buffer | |
43 | * is either exclusively owned by the CPU (and therefore may be accessed | |
44 | * by it) or exclusively owned by the DMA device. These helper functions | |
45 | * represent the transitions between these two ownership states. | |
46 | * | |
47 | * Note, however, that on later ARMs, this notion does not work due to | |
48 | * speculative prefetches. We model our approach on the assumption that | |
49 | * the CPU does do speculative prefetches, which means we clean caches | |
50 | * before transfers and delay cache invalidation until transfer completion. | |
51 | * | |
15237e1f | 52 | */ |
51fde349 | 53 | static void __dma_page_cpu_to_dev(struct page *, unsigned long, |
15237e1f | 54 | size_t, enum dma_data_direction); |
51fde349 | 55 | static void __dma_page_dev_to_cpu(struct page *, unsigned long, |
15237e1f MS |
56 | size_t, enum dma_data_direction); |
57 | ||
2dc6a016 MS |
58 | /** |
59 | * arm_dma_map_page - map a portion of a page for streaming DMA | |
60 | * @dev: valid struct device pointer, or NULL for ISA and EISA-like devices | |
61 | * @page: page that buffer resides in | |
62 | * @offset: offset into page for start of buffer | |
63 | * @size: size of buffer to map | |
64 | * @dir: DMA transfer direction | |
65 | * | |
66 | * Ensure that any data held in the cache is appropriately discarded | |
67 | * or written back. | |
68 | * | |
69 | * The device owns this memory once this call has completed. The CPU | |
70 | * can regain ownership by calling dma_unmap_page(). | |
71 | */ | |
51fde349 | 72 | static dma_addr_t arm_dma_map_page(struct device *dev, struct page *page, |
2dc6a016 MS |
73 | unsigned long offset, size_t size, enum dma_data_direction dir, |
74 | struct dma_attrs *attrs) | |
75 | { | |
dd37e940 | 76 | if (!dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) |
51fde349 MS |
77 | __dma_page_cpu_to_dev(page, offset, size, dir); |
78 | return pfn_to_dma(dev, page_to_pfn(page)) + offset; | |
2dc6a016 MS |
79 | } |
80 | ||
dd37e940 RH |
81 | static dma_addr_t arm_coherent_dma_map_page(struct device *dev, struct page *page, |
82 | unsigned long offset, size_t size, enum dma_data_direction dir, | |
83 | struct dma_attrs *attrs) | |
84 | { | |
85 | return pfn_to_dma(dev, page_to_pfn(page)) + offset; | |
86 | } | |
87 | ||
2dc6a016 MS |
88 | /** |
89 | * arm_dma_unmap_page - unmap a buffer previously mapped through dma_map_page() | |
90 | * @dev: valid struct device pointer, or NULL for ISA and EISA-like devices | |
91 | * @handle: DMA address of buffer | |
92 | * @size: size of buffer (same as passed to dma_map_page) | |
93 | * @dir: DMA transfer direction (same as passed to dma_map_page) | |
94 | * | |
95 | * Unmap a page streaming mode DMA translation. The handle and size | |
96 | * must match what was provided in the previous dma_map_page() call. | |
97 | * All other usages are undefined. | |
98 | * | |
99 | * After this call, reads by the CPU to the buffer are guaranteed to see | |
100 | * whatever the device wrote there. | |
101 | */ | |
51fde349 | 102 | static void arm_dma_unmap_page(struct device *dev, dma_addr_t handle, |
2dc6a016 MS |
103 | size_t size, enum dma_data_direction dir, |
104 | struct dma_attrs *attrs) | |
105 | { | |
dd37e940 | 106 | if (!dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) |
51fde349 MS |
107 | __dma_page_dev_to_cpu(pfn_to_page(dma_to_pfn(dev, handle)), |
108 | handle & ~PAGE_MASK, size, dir); | |
2dc6a016 MS |
109 | } |
110 | ||
51fde349 | 111 | static void arm_dma_sync_single_for_cpu(struct device *dev, |
2dc6a016 MS |
112 | dma_addr_t handle, size_t size, enum dma_data_direction dir) |
113 | { | |
114 | unsigned int offset = handle & (PAGE_SIZE - 1); | |
115 | struct page *page = pfn_to_page(dma_to_pfn(dev, handle-offset)); | |
dd37e940 | 116 | __dma_page_dev_to_cpu(page, offset, size, dir); |
2dc6a016 MS |
117 | } |
118 | ||
51fde349 | 119 | static void arm_dma_sync_single_for_device(struct device *dev, |
2dc6a016 MS |
120 | dma_addr_t handle, size_t size, enum dma_data_direction dir) |
121 | { | |
122 | unsigned int offset = handle & (PAGE_SIZE - 1); | |
123 | struct page *page = pfn_to_page(dma_to_pfn(dev, handle-offset)); | |
dd37e940 | 124 | __dma_page_cpu_to_dev(page, offset, size, dir); |
2dc6a016 MS |
125 | } |
126 | ||
2dc6a016 | 127 | struct dma_map_ops arm_dma_ops = { |
f99d6034 MS |
128 | .alloc = arm_dma_alloc, |
129 | .free = arm_dma_free, | |
130 | .mmap = arm_dma_mmap, | |
dc2832e1 | 131 | .get_sgtable = arm_dma_get_sgtable, |
2dc6a016 MS |
132 | .map_page = arm_dma_map_page, |
133 | .unmap_page = arm_dma_unmap_page, | |
134 | .map_sg = arm_dma_map_sg, | |
135 | .unmap_sg = arm_dma_unmap_sg, | |
136 | .sync_single_for_cpu = arm_dma_sync_single_for_cpu, | |
137 | .sync_single_for_device = arm_dma_sync_single_for_device, | |
138 | .sync_sg_for_cpu = arm_dma_sync_sg_for_cpu, | |
139 | .sync_sg_for_device = arm_dma_sync_sg_for_device, | |
140 | .set_dma_mask = arm_dma_set_mask, | |
141 | }; | |
142 | EXPORT_SYMBOL(arm_dma_ops); | |
143 | ||
dd37e940 RH |
144 | static void *arm_coherent_dma_alloc(struct device *dev, size_t size, |
145 | dma_addr_t *handle, gfp_t gfp, struct dma_attrs *attrs); | |
146 | static void arm_coherent_dma_free(struct device *dev, size_t size, void *cpu_addr, | |
147 | dma_addr_t handle, struct dma_attrs *attrs); | |
148 | ||
149 | struct dma_map_ops arm_coherent_dma_ops = { | |
150 | .alloc = arm_coherent_dma_alloc, | |
151 | .free = arm_coherent_dma_free, | |
152 | .mmap = arm_dma_mmap, | |
153 | .get_sgtable = arm_dma_get_sgtable, | |
154 | .map_page = arm_coherent_dma_map_page, | |
155 | .map_sg = arm_dma_map_sg, | |
156 | .set_dma_mask = arm_dma_set_mask, | |
157 | }; | |
158 | EXPORT_SYMBOL(arm_coherent_dma_ops); | |
159 | ||
ab6494f0 CM |
160 | static u64 get_coherent_dma_mask(struct device *dev) |
161 | { | |
022ae537 | 162 | u64 mask = (u64)arm_dma_limit; |
ab6494f0 CM |
163 | |
164 | if (dev) { | |
165 | mask = dev->coherent_dma_mask; | |
166 | ||
167 | /* | |
168 | * Sanity check the DMA mask - it must be non-zero, and | |
169 | * must be able to be satisfied by a DMA allocation. | |
170 | */ | |
171 | if (mask == 0) { | |
172 | dev_warn(dev, "coherent DMA mask is unset\n"); | |
173 | return 0; | |
174 | } | |
175 | ||
022ae537 | 176 | if ((~mask) & (u64)arm_dma_limit) { |
ab6494f0 CM |
177 | dev_warn(dev, "coherent DMA mask %#llx is smaller " |
178 | "than system GFP_DMA mask %#llx\n", | |
022ae537 | 179 | mask, (u64)arm_dma_limit); |
ab6494f0 CM |
180 | return 0; |
181 | } | |
182 | } | |
1da177e4 | 183 | |
ab6494f0 CM |
184 | return mask; |
185 | } | |
186 | ||
c7909509 MS |
187 | static void __dma_clear_buffer(struct page *page, size_t size) |
188 | { | |
c7909509 MS |
189 | /* |
190 | * Ensure that the allocated pages are zeroed, and that any data | |
191 | * lurking in the kernel direct-mapped region is invalidated. | |
192 | */ | |
9848e48f MS |
193 | if (PageHighMem(page)) { |
194 | phys_addr_t base = __pfn_to_phys(page_to_pfn(page)); | |
195 | phys_addr_t end = base + size; | |
196 | while (size > 0) { | |
197 | void *ptr = kmap_atomic(page); | |
198 | memset(ptr, 0, PAGE_SIZE); | |
199 | dmac_flush_range(ptr, ptr + PAGE_SIZE); | |
200 | kunmap_atomic(ptr); | |
201 | page++; | |
202 | size -= PAGE_SIZE; | |
203 | } | |
204 | outer_flush_range(base, end); | |
205 | } else { | |
206 | void *ptr = page_address(page); | |
4ce63fcd MS |
207 | memset(ptr, 0, size); |
208 | dmac_flush_range(ptr, ptr + size); | |
209 | outer_flush_range(__pa(ptr), __pa(ptr) + size); | |
210 | } | |
c7909509 MS |
211 | } |
212 | ||
7a9a32a9 RK |
213 | /* |
214 | * Allocate a DMA buffer for 'dev' of size 'size' using the | |
215 | * specified gfp mask. Note that 'size' must be page aligned. | |
216 | */ | |
217 | static struct page *__dma_alloc_buffer(struct device *dev, size_t size, gfp_t gfp) | |
218 | { | |
219 | unsigned long order = get_order(size); | |
220 | struct page *page, *p, *e; | |
7a9a32a9 RK |
221 | |
222 | page = alloc_pages(gfp, order); | |
223 | if (!page) | |
224 | return NULL; | |
225 | ||
226 | /* | |
227 | * Now split the huge page and free the excess pages | |
228 | */ | |
229 | split_page(page, order); | |
230 | for (p = page + (size >> PAGE_SHIFT), e = page + (1 << order); p < e; p++) | |
231 | __free_page(p); | |
232 | ||
c7909509 | 233 | __dma_clear_buffer(page, size); |
7a9a32a9 RK |
234 | |
235 | return page; | |
236 | } | |
237 | ||
238 | /* | |
239 | * Free a DMA buffer. 'size' must be page aligned. | |
240 | */ | |
241 | static void __dma_free_buffer(struct page *page, size_t size) | |
242 | { | |
243 | struct page *e = page + (size >> PAGE_SHIFT); | |
244 | ||
245 | while (page < e) { | |
246 | __free_page(page); | |
247 | page++; | |
248 | } | |
249 | } | |
250 | ||
ab6494f0 | 251 | #ifdef CONFIG_MMU |
e9da6e99 | 252 | #ifdef CONFIG_HUGETLB_PAGE |
1355e2a6 | 253 | #warning ARM Coherent DMA allocator does not (yet) support huge TLB |
e9da6e99 | 254 | #endif |
a5e9d38b | 255 | |
e9da6e99 | 256 | static void *__alloc_from_contiguous(struct device *dev, size_t size, |
9848e48f MS |
257 | pgprot_t prot, struct page **ret_page, |
258 | const void *caller); | |
99d1717d | 259 | |
e9da6e99 MS |
260 | static void *__alloc_remap_buffer(struct device *dev, size_t size, gfp_t gfp, |
261 | pgprot_t prot, struct page **ret_page, | |
262 | const void *caller); | |
99d1717d | 263 | |
e9da6e99 MS |
264 | static void * |
265 | __dma_alloc_remap(struct page *page, size_t size, gfp_t gfp, pgprot_t prot, | |
266 | const void *caller) | |
99d1717d | 267 | { |
e9da6e99 MS |
268 | struct vm_struct *area; |
269 | unsigned long addr; | |
99d1717d | 270 | |
e9da6e99 MS |
271 | /* |
272 | * DMA allocation can be mapped to user space, so lets | |
273 | * set VM_USERMAP flags too. | |
274 | */ | |
275 | area = get_vm_area_caller(size, VM_ARM_DMA_CONSISTENT | VM_USERMAP, | |
276 | caller); | |
277 | if (!area) | |
278 | return NULL; | |
279 | addr = (unsigned long)area->addr; | |
280 | area->phys_addr = __pfn_to_phys(page_to_pfn(page)); | |
99d1717d | 281 | |
e9da6e99 MS |
282 | if (ioremap_page_range(addr, addr + size, area->phys_addr, prot)) { |
283 | vunmap((void *)addr); | |
284 | return NULL; | |
285 | } | |
286 | return (void *)addr; | |
99d1717d | 287 | } |
1da177e4 | 288 | |
e9da6e99 | 289 | static void __dma_free_remap(void *cpu_addr, size_t size) |
88c58f3b | 290 | { |
e9da6e99 MS |
291 | unsigned int flags = VM_ARM_DMA_CONSISTENT | VM_USERMAP; |
292 | struct vm_struct *area = find_vm_area(cpu_addr); | |
293 | if (!area || (area->flags & flags) != flags) { | |
294 | WARN(1, "trying to free invalid coherent area: %p\n", cpu_addr); | |
295 | return; | |
99d1717d | 296 | } |
e9da6e99 MS |
297 | unmap_kernel_range((unsigned long)cpu_addr, size); |
298 | vunmap(cpu_addr); | |
88c58f3b | 299 | } |
88c58f3b | 300 | |
6e5267aa MS |
301 | #define DEFAULT_DMA_COHERENT_POOL_SIZE SZ_256K |
302 | ||
e9da6e99 MS |
303 | struct dma_pool { |
304 | size_t size; | |
305 | spinlock_t lock; | |
306 | unsigned long *bitmap; | |
307 | unsigned long nr_pages; | |
308 | void *vaddr; | |
6b3fe472 | 309 | struct page **pages; |
c7909509 MS |
310 | }; |
311 | ||
e9da6e99 | 312 | static struct dma_pool atomic_pool = { |
6e5267aa | 313 | .size = DEFAULT_DMA_COHERENT_POOL_SIZE, |
e9da6e99 | 314 | }; |
c7909509 MS |
315 | |
316 | static int __init early_coherent_pool(char *p) | |
317 | { | |
e9da6e99 | 318 | atomic_pool.size = memparse(p, &p); |
c7909509 MS |
319 | return 0; |
320 | } | |
321 | early_param("coherent_pool", early_coherent_pool); | |
322 | ||
6e5267aa MS |
323 | void __init init_dma_coherent_pool_size(unsigned long size) |
324 | { | |
325 | /* | |
326 | * Catch any attempt to set the pool size too late. | |
327 | */ | |
328 | BUG_ON(atomic_pool.vaddr); | |
329 | ||
330 | /* | |
331 | * Set architecture specific coherent pool size only if | |
332 | * it has not been changed by kernel command line parameter. | |
333 | */ | |
334 | if (atomic_pool.size == DEFAULT_DMA_COHERENT_POOL_SIZE) | |
335 | atomic_pool.size = size; | |
336 | } | |
337 | ||
c7909509 MS |
338 | /* |
339 | * Initialise the coherent pool for atomic allocations. | |
340 | */ | |
e9da6e99 | 341 | static int __init atomic_pool_init(void) |
c7909509 | 342 | { |
e9da6e99 | 343 | struct dma_pool *pool = &atomic_pool; |
c7909509 | 344 | pgprot_t prot = pgprot_dmacoherent(pgprot_kernel); |
9d1400cf | 345 | gfp_t gfp = GFP_KERNEL | GFP_DMA; |
e9da6e99 MS |
346 | unsigned long nr_pages = pool->size >> PAGE_SHIFT; |
347 | unsigned long *bitmap; | |
c7909509 | 348 | struct page *page; |
6b3fe472 | 349 | struct page **pages; |
c7909509 | 350 | void *ptr; |
e9da6e99 | 351 | int bitmap_size = BITS_TO_LONGS(nr_pages) * sizeof(long); |
c7909509 | 352 | |
e9da6e99 MS |
353 | bitmap = kzalloc(bitmap_size, GFP_KERNEL); |
354 | if (!bitmap) | |
355 | goto no_bitmap; | |
c7909509 | 356 | |
6b3fe472 HD |
357 | pages = kzalloc(nr_pages * sizeof(struct page *), GFP_KERNEL); |
358 | if (!pages) | |
359 | goto no_pages; | |
360 | ||
f825c736 | 361 | if (IS_ENABLED(CONFIG_DMA_CMA)) |
9848e48f MS |
362 | ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page, |
363 | atomic_pool_init); | |
e9da6e99 | 364 | else |
9d1400cf MS |
365 | ptr = __alloc_remap_buffer(NULL, pool->size, gfp, prot, &page, |
366 | atomic_pool_init); | |
c7909509 | 367 | if (ptr) { |
6b3fe472 HD |
368 | int i; |
369 | ||
370 | for (i = 0; i < nr_pages; i++) | |
371 | pages[i] = page + i; | |
372 | ||
e9da6e99 MS |
373 | spin_lock_init(&pool->lock); |
374 | pool->vaddr = ptr; | |
6b3fe472 | 375 | pool->pages = pages; |
e9da6e99 MS |
376 | pool->bitmap = bitmap; |
377 | pool->nr_pages = nr_pages; | |
378 | pr_info("DMA: preallocated %u KiB pool for atomic coherent allocations\n", | |
379 | (unsigned)pool->size / 1024); | |
c7909509 MS |
380 | return 0; |
381 | } | |
ec10665c SK |
382 | |
383 | kfree(pages); | |
6b3fe472 | 384 | no_pages: |
e9da6e99 MS |
385 | kfree(bitmap); |
386 | no_bitmap: | |
387 | pr_err("DMA: failed to allocate %u KiB pool for atomic coherent allocation\n", | |
388 | (unsigned)pool->size / 1024); | |
c7909509 MS |
389 | return -ENOMEM; |
390 | } | |
391 | /* | |
392 | * CMA is activated by core_initcall, so we must be called after it. | |
393 | */ | |
e9da6e99 | 394 | postcore_initcall(atomic_pool_init); |
c7909509 MS |
395 | |
396 | struct dma_contig_early_reserve { | |
397 | phys_addr_t base; | |
398 | unsigned long size; | |
399 | }; | |
400 | ||
401 | static struct dma_contig_early_reserve dma_mmu_remap[MAX_CMA_AREAS] __initdata; | |
402 | ||
403 | static int dma_mmu_remap_num __initdata; | |
404 | ||
405 | void __init dma_contiguous_early_fixup(phys_addr_t base, unsigned long size) | |
406 | { | |
407 | dma_mmu_remap[dma_mmu_remap_num].base = base; | |
408 | dma_mmu_remap[dma_mmu_remap_num].size = size; | |
409 | dma_mmu_remap_num++; | |
410 | } | |
411 | ||
412 | void __init dma_contiguous_remap(void) | |
413 | { | |
414 | int i; | |
415 | for (i = 0; i < dma_mmu_remap_num; i++) { | |
416 | phys_addr_t start = dma_mmu_remap[i].base; | |
417 | phys_addr_t end = start + dma_mmu_remap[i].size; | |
418 | struct map_desc map; | |
419 | unsigned long addr; | |
420 | ||
421 | if (end > arm_lowmem_limit) | |
422 | end = arm_lowmem_limit; | |
423 | if (start >= end) | |
39f78e70 | 424 | continue; |
c7909509 MS |
425 | |
426 | map.pfn = __phys_to_pfn(start); | |
427 | map.virtual = __phys_to_virt(start); | |
428 | map.length = end - start; | |
429 | map.type = MT_MEMORY_DMA_READY; | |
430 | ||
431 | /* | |
432 | * Clear previous low-memory mapping | |
433 | */ | |
434 | for (addr = __phys_to_virt(start); addr < __phys_to_virt(end); | |
61f6c7a4 | 435 | addr += PMD_SIZE) |
c7909509 MS |
436 | pmd_clear(pmd_off_k(addr)); |
437 | ||
438 | iotable_init(&map, 1); | |
439 | } | |
440 | } | |
441 | ||
c7909509 MS |
442 | static int __dma_update_pte(pte_t *pte, pgtable_t token, unsigned long addr, |
443 | void *data) | |
444 | { | |
445 | struct page *page = virt_to_page(addr); | |
446 | pgprot_t prot = *(pgprot_t *)data; | |
447 | ||
448 | set_pte_ext(pte, mk_pte(page, prot), 0); | |
449 | return 0; | |
450 | } | |
451 | ||
452 | static void __dma_remap(struct page *page, size_t size, pgprot_t prot) | |
453 | { | |
454 | unsigned long start = (unsigned long) page_address(page); | |
455 | unsigned end = start + size; | |
456 | ||
457 | apply_to_page_range(&init_mm, start, size, __dma_update_pte, &prot); | |
c7909509 MS |
458 | flush_tlb_kernel_range(start, end); |
459 | } | |
460 | ||
461 | static void *__alloc_remap_buffer(struct device *dev, size_t size, gfp_t gfp, | |
462 | pgprot_t prot, struct page **ret_page, | |
463 | const void *caller) | |
464 | { | |
465 | struct page *page; | |
466 | void *ptr; | |
467 | page = __dma_alloc_buffer(dev, size, gfp); | |
468 | if (!page) | |
469 | return NULL; | |
470 | ||
471 | ptr = __dma_alloc_remap(page, size, gfp, prot, caller); | |
472 | if (!ptr) { | |
473 | __dma_free_buffer(page, size); | |
474 | return NULL; | |
475 | } | |
476 | ||
477 | *ret_page = page; | |
478 | return ptr; | |
479 | } | |
480 | ||
e9da6e99 | 481 | static void *__alloc_from_pool(size_t size, struct page **ret_page) |
c7909509 | 482 | { |
e9da6e99 MS |
483 | struct dma_pool *pool = &atomic_pool; |
484 | unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT; | |
485 | unsigned int pageno; | |
486 | unsigned long flags; | |
487 | void *ptr = NULL; | |
e4ea6918 | 488 | unsigned long align_mask; |
c7909509 | 489 | |
e9da6e99 MS |
490 | if (!pool->vaddr) { |
491 | WARN(1, "coherent pool not initialised!\n"); | |
c7909509 MS |
492 | return NULL; |
493 | } | |
494 | ||
495 | /* | |
496 | * Align the region allocation - allocations from pool are rather | |
497 | * small, so align them to their order in pages, minimum is a page | |
498 | * size. This helps reduce fragmentation of the DMA space. | |
499 | */ | |
e4ea6918 | 500 | align_mask = (1 << get_order(size)) - 1; |
e9da6e99 MS |
501 | |
502 | spin_lock_irqsave(&pool->lock, flags); | |
503 | pageno = bitmap_find_next_zero_area(pool->bitmap, pool->nr_pages, | |
e4ea6918 | 504 | 0, count, align_mask); |
e9da6e99 MS |
505 | if (pageno < pool->nr_pages) { |
506 | bitmap_set(pool->bitmap, pageno, count); | |
507 | ptr = pool->vaddr + PAGE_SIZE * pageno; | |
6b3fe472 | 508 | *ret_page = pool->pages[pageno]; |
fb71285f MS |
509 | } else { |
510 | pr_err_once("ERROR: %u KiB atomic DMA coherent pool is too small!\n" | |
511 | "Please increase it with coherent_pool= kernel parameter!\n", | |
512 | (unsigned)pool->size / 1024); | |
c7909509 | 513 | } |
e9da6e99 MS |
514 | spin_unlock_irqrestore(&pool->lock, flags); |
515 | ||
516 | return ptr; | |
c7909509 MS |
517 | } |
518 | ||
21d0a759 HD |
519 | static bool __in_atomic_pool(void *start, size_t size) |
520 | { | |
521 | struct dma_pool *pool = &atomic_pool; | |
522 | void *end = start + size; | |
523 | void *pool_start = pool->vaddr; | |
524 | void *pool_end = pool->vaddr + pool->size; | |
525 | ||
f3d87524 | 526 | if (start < pool_start || start >= pool_end) |
21d0a759 HD |
527 | return false; |
528 | ||
529 | if (end <= pool_end) | |
530 | return true; | |
531 | ||
532 | WARN(1, "Wrong coherent size(%p-%p) from atomic pool(%p-%p)\n", | |
533 | start, end - 1, pool_start, pool_end - 1); | |
534 | ||
535 | return false; | |
536 | } | |
537 | ||
e9da6e99 | 538 | static int __free_from_pool(void *start, size_t size) |
c7909509 | 539 | { |
e9da6e99 MS |
540 | struct dma_pool *pool = &atomic_pool; |
541 | unsigned long pageno, count; | |
542 | unsigned long flags; | |
c7909509 | 543 | |
21d0a759 | 544 | if (!__in_atomic_pool(start, size)) |
c7909509 MS |
545 | return 0; |
546 | ||
e9da6e99 MS |
547 | pageno = (start - pool->vaddr) >> PAGE_SHIFT; |
548 | count = size >> PAGE_SHIFT; | |
549 | ||
550 | spin_lock_irqsave(&pool->lock, flags); | |
551 | bitmap_clear(pool->bitmap, pageno, count); | |
552 | spin_unlock_irqrestore(&pool->lock, flags); | |
553 | ||
c7909509 MS |
554 | return 1; |
555 | } | |
556 | ||
557 | static void *__alloc_from_contiguous(struct device *dev, size_t size, | |
9848e48f MS |
558 | pgprot_t prot, struct page **ret_page, |
559 | const void *caller) | |
c7909509 MS |
560 | { |
561 | unsigned long order = get_order(size); | |
562 | size_t count = size >> PAGE_SHIFT; | |
563 | struct page *page; | |
9848e48f | 564 | void *ptr; |
c7909509 MS |
565 | |
566 | page = dma_alloc_from_contiguous(dev, count, order); | |
567 | if (!page) | |
568 | return NULL; | |
569 | ||
570 | __dma_clear_buffer(page, size); | |
c7909509 | 571 | |
9848e48f MS |
572 | if (PageHighMem(page)) { |
573 | ptr = __dma_alloc_remap(page, size, GFP_KERNEL, prot, caller); | |
574 | if (!ptr) { | |
575 | dma_release_from_contiguous(dev, page, count); | |
576 | return NULL; | |
577 | } | |
578 | } else { | |
579 | __dma_remap(page, size, prot); | |
580 | ptr = page_address(page); | |
581 | } | |
c7909509 | 582 | *ret_page = page; |
9848e48f | 583 | return ptr; |
c7909509 MS |
584 | } |
585 | ||
586 | static void __free_from_contiguous(struct device *dev, struct page *page, | |
9848e48f | 587 | void *cpu_addr, size_t size) |
c7909509 | 588 | { |
9848e48f MS |
589 | if (PageHighMem(page)) |
590 | __dma_free_remap(cpu_addr, size); | |
591 | else | |
592 | __dma_remap(page, size, pgprot_kernel); | |
c7909509 MS |
593 | dma_release_from_contiguous(dev, page, size >> PAGE_SHIFT); |
594 | } | |
595 | ||
f99d6034 MS |
596 | static inline pgprot_t __get_dma_pgprot(struct dma_attrs *attrs, pgprot_t prot) |
597 | { | |
598 | prot = dma_get_attr(DMA_ATTR_WRITE_COMBINE, attrs) ? | |
599 | pgprot_writecombine(prot) : | |
600 | pgprot_dmacoherent(prot); | |
601 | return prot; | |
602 | } | |
603 | ||
c7909509 MS |
604 | #define nommu() 0 |
605 | ||
ab6494f0 | 606 | #else /* !CONFIG_MMU */ |
695ae0af | 607 | |
c7909509 MS |
608 | #define nommu() 1 |
609 | ||
f99d6034 | 610 | #define __get_dma_pgprot(attrs, prot) __pgprot(0) |
c7909509 | 611 | #define __alloc_remap_buffer(dev, size, gfp, prot, ret, c) NULL |
e9da6e99 | 612 | #define __alloc_from_pool(size, ret_page) NULL |
9848e48f | 613 | #define __alloc_from_contiguous(dev, size, prot, ret, c) NULL |
c7909509 | 614 | #define __free_from_pool(cpu_addr, size) 0 |
9848e48f | 615 | #define __free_from_contiguous(dev, page, cpu_addr, size) do { } while (0) |
c7909509 | 616 | #define __dma_free_remap(cpu_addr, size) do { } while (0) |
31ebf944 RK |
617 | |
618 | #endif /* CONFIG_MMU */ | |
619 | ||
c7909509 MS |
620 | static void *__alloc_simple_buffer(struct device *dev, size_t size, gfp_t gfp, |
621 | struct page **ret_page) | |
ab6494f0 | 622 | { |
c7909509 MS |
623 | struct page *page; |
624 | page = __dma_alloc_buffer(dev, size, gfp); | |
625 | if (!page) | |
626 | return NULL; | |
627 | ||
628 | *ret_page = page; | |
629 | return page_address(page); | |
630 | } | |
631 | ||
632 | ||
633 | ||
634 | static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle, | |
dd37e940 | 635 | gfp_t gfp, pgprot_t prot, bool is_coherent, const void *caller) |
c7909509 MS |
636 | { |
637 | u64 mask = get_coherent_dma_mask(dev); | |
3dd7ea92 | 638 | struct page *page = NULL; |
31ebf944 | 639 | void *addr; |
ab6494f0 | 640 | |
c7909509 MS |
641 | #ifdef CONFIG_DMA_API_DEBUG |
642 | u64 limit = (mask + 1) & ~mask; | |
643 | if (limit && size >= limit) { | |
644 | dev_warn(dev, "coherent allocation too big (requested %#x mask %#llx)\n", | |
645 | size, mask); | |
646 | return NULL; | |
647 | } | |
648 | #endif | |
649 | ||
650 | if (!mask) | |
651 | return NULL; | |
652 | ||
653 | if (mask < 0xffffffffULL) | |
654 | gfp |= GFP_DMA; | |
655 | ||
ea2e7057 SB |
656 | /* |
657 | * Following is a work-around (a.k.a. hack) to prevent pages | |
658 | * with __GFP_COMP being passed to split_page() which cannot | |
659 | * handle them. The real problem is that this flag probably | |
660 | * should be 0 on ARM as it is not supported on this | |
661 | * platform; see CONFIG_HUGETLBFS. | |
662 | */ | |
663 | gfp &= ~(__GFP_COMP); | |
664 | ||
553ac788 | 665 | *handle = DMA_ERROR_CODE; |
04da5694 | 666 | size = PAGE_ALIGN(size); |
ab6494f0 | 667 | |
dd37e940 | 668 | if (is_coherent || nommu()) |
c7909509 | 669 | addr = __alloc_simple_buffer(dev, size, gfp, &page); |
633dc92a | 670 | else if (!(gfp & __GFP_WAIT)) |
e9da6e99 | 671 | addr = __alloc_from_pool(size, &page); |
f825c736 | 672 | else if (!IS_ENABLED(CONFIG_DMA_CMA)) |
c7909509 | 673 | addr = __alloc_remap_buffer(dev, size, gfp, prot, &page, caller); |
31ebf944 | 674 | else |
9848e48f | 675 | addr = __alloc_from_contiguous(dev, size, prot, &page, caller); |
695ae0af | 676 | |
31ebf944 | 677 | if (addr) |
9eedd963 | 678 | *handle = pfn_to_dma(dev, page_to_pfn(page)); |
695ae0af | 679 | |
31ebf944 RK |
680 | return addr; |
681 | } | |
1da177e4 LT |
682 | |
683 | /* | |
684 | * Allocate DMA-coherent memory space and return both the kernel remapped | |
685 | * virtual and bus address for that space. | |
686 | */ | |
f99d6034 MS |
687 | void *arm_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle, |
688 | gfp_t gfp, struct dma_attrs *attrs) | |
1da177e4 | 689 | { |
f99d6034 | 690 | pgprot_t prot = __get_dma_pgprot(attrs, pgprot_kernel); |
1fe53268 DB |
691 | void *memory; |
692 | ||
693 | if (dma_alloc_from_coherent(dev, size, handle, &memory)) | |
694 | return memory; | |
695 | ||
dd37e940 RH |
696 | return __dma_alloc(dev, size, handle, gfp, prot, false, |
697 | __builtin_return_address(0)); | |
698 | } | |
699 | ||
700 | static void *arm_coherent_dma_alloc(struct device *dev, size_t size, | |
701 | dma_addr_t *handle, gfp_t gfp, struct dma_attrs *attrs) | |
702 | { | |
703 | pgprot_t prot = __get_dma_pgprot(attrs, pgprot_kernel); | |
704 | void *memory; | |
705 | ||
706 | if (dma_alloc_from_coherent(dev, size, handle, &memory)) | |
707 | return memory; | |
708 | ||
709 | return __dma_alloc(dev, size, handle, gfp, prot, true, | |
45cd5290 | 710 | __builtin_return_address(0)); |
1da177e4 | 711 | } |
1da177e4 LT |
712 | |
713 | /* | |
f99d6034 | 714 | * Create userspace mapping for the DMA-coherent memory. |
1da177e4 | 715 | */ |
f99d6034 MS |
716 | int arm_dma_mmap(struct device *dev, struct vm_area_struct *vma, |
717 | void *cpu_addr, dma_addr_t dma_addr, size_t size, | |
718 | struct dma_attrs *attrs) | |
1da177e4 | 719 | { |
ab6494f0 CM |
720 | int ret = -ENXIO; |
721 | #ifdef CONFIG_MMU | |
50262a4b MS |
722 | unsigned long nr_vma_pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; |
723 | unsigned long nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT; | |
c7909509 | 724 | unsigned long pfn = dma_to_pfn(dev, dma_addr); |
50262a4b MS |
725 | unsigned long off = vma->vm_pgoff; |
726 | ||
f99d6034 MS |
727 | vma->vm_page_prot = __get_dma_pgprot(attrs, vma->vm_page_prot); |
728 | ||
47142f07 MS |
729 | if (dma_mmap_from_coherent(dev, vma, cpu_addr, size, &ret)) |
730 | return ret; | |
731 | ||
50262a4b MS |
732 | if (off < nr_pages && nr_vma_pages <= (nr_pages - off)) { |
733 | ret = remap_pfn_range(vma, vma->vm_start, | |
734 | pfn + off, | |
735 | vma->vm_end - vma->vm_start, | |
736 | vma->vm_page_prot); | |
737 | } | |
ab6494f0 | 738 | #endif /* CONFIG_MMU */ |
1da177e4 LT |
739 | |
740 | return ret; | |
741 | } | |
742 | ||
1da177e4 | 743 | /* |
c7909509 | 744 | * Free a buffer as defined by the above mapping. |
1da177e4 | 745 | */ |
dd37e940 RH |
746 | static void __arm_dma_free(struct device *dev, size_t size, void *cpu_addr, |
747 | dma_addr_t handle, struct dma_attrs *attrs, | |
748 | bool is_coherent) | |
1da177e4 | 749 | { |
c7909509 | 750 | struct page *page = pfn_to_page(dma_to_pfn(dev, handle)); |
5edf71ae | 751 | |
1fe53268 DB |
752 | if (dma_release_from_coherent(dev, get_order(size), cpu_addr)) |
753 | return; | |
754 | ||
3e82d012 RK |
755 | size = PAGE_ALIGN(size); |
756 | ||
dd37e940 | 757 | if (is_coherent || nommu()) { |
c7909509 | 758 | __dma_free_buffer(page, size); |
d9e0d149 AK |
759 | } else if (__free_from_pool(cpu_addr, size)) { |
760 | return; | |
f825c736 | 761 | } else if (!IS_ENABLED(CONFIG_DMA_CMA)) { |
695ae0af | 762 | __dma_free_remap(cpu_addr, size); |
c7909509 MS |
763 | __dma_free_buffer(page, size); |
764 | } else { | |
c7909509 MS |
765 | /* |
766 | * Non-atomic allocations cannot be freed with IRQs disabled | |
767 | */ | |
768 | WARN_ON(irqs_disabled()); | |
9848e48f | 769 | __free_from_contiguous(dev, page, cpu_addr, size); |
c7909509 | 770 | } |
1da177e4 | 771 | } |
afd1a321 | 772 | |
dd37e940 RH |
773 | void arm_dma_free(struct device *dev, size_t size, void *cpu_addr, |
774 | dma_addr_t handle, struct dma_attrs *attrs) | |
775 | { | |
776 | __arm_dma_free(dev, size, cpu_addr, handle, attrs, false); | |
777 | } | |
778 | ||
779 | static void arm_coherent_dma_free(struct device *dev, size_t size, void *cpu_addr, | |
780 | dma_addr_t handle, struct dma_attrs *attrs) | |
781 | { | |
782 | __arm_dma_free(dev, size, cpu_addr, handle, attrs, true); | |
783 | } | |
784 | ||
dc2832e1 MS |
785 | int arm_dma_get_sgtable(struct device *dev, struct sg_table *sgt, |
786 | void *cpu_addr, dma_addr_t handle, size_t size, | |
787 | struct dma_attrs *attrs) | |
788 | { | |
789 | struct page *page = pfn_to_page(dma_to_pfn(dev, handle)); | |
790 | int ret; | |
791 | ||
792 | ret = sg_alloc_table(sgt, 1, GFP_KERNEL); | |
793 | if (unlikely(ret)) | |
794 | return ret; | |
795 | ||
796 | sg_set_page(sgt->sgl, page, PAGE_ALIGN(size), 0); | |
797 | return 0; | |
798 | } | |
799 | ||
4ea0d737 | 800 | static void dma_cache_maint_page(struct page *page, unsigned long offset, |
a9c9147e RK |
801 | size_t size, enum dma_data_direction dir, |
802 | void (*op)(const void *, size_t, int)) | |
43377453 | 803 | { |
15653371 RK |
804 | unsigned long pfn; |
805 | size_t left = size; | |
806 | ||
807 | pfn = page_to_pfn(page) + offset / PAGE_SIZE; | |
808 | offset %= PAGE_SIZE; | |
809 | ||
43377453 NP |
810 | /* |
811 | * A single sg entry may refer to multiple physically contiguous | |
812 | * pages. But we still need to process highmem pages individually. | |
813 | * If highmem is not configured then the bulk of this loop gets | |
814 | * optimized out. | |
815 | */ | |
43377453 NP |
816 | do { |
817 | size_t len = left; | |
93f1d629 RK |
818 | void *vaddr; |
819 | ||
15653371 RK |
820 | page = pfn_to_page(pfn); |
821 | ||
93f1d629 | 822 | if (PageHighMem(page)) { |
15653371 | 823 | if (len + offset > PAGE_SIZE) |
93f1d629 | 824 | len = PAGE_SIZE - offset; |
dd0f67f4 JK |
825 | |
826 | if (cache_is_vipt_nonaliasing()) { | |
39af22a7 | 827 | vaddr = kmap_atomic(page); |
7e5a69e8 | 828 | op(vaddr + offset, len, dir); |
39af22a7 | 829 | kunmap_atomic(vaddr); |
dd0f67f4 JK |
830 | } else { |
831 | vaddr = kmap_high_get(page); | |
832 | if (vaddr) { | |
833 | op(vaddr + offset, len, dir); | |
834 | kunmap_high(page); | |
835 | } | |
43377453 | 836 | } |
93f1d629 RK |
837 | } else { |
838 | vaddr = page_address(page) + offset; | |
a9c9147e | 839 | op(vaddr, len, dir); |
43377453 | 840 | } |
43377453 | 841 | offset = 0; |
15653371 | 842 | pfn++; |
43377453 NP |
843 | left -= len; |
844 | } while (left); | |
845 | } | |
4ea0d737 | 846 | |
51fde349 MS |
847 | /* |
848 | * Make an area consistent for devices. | |
849 | * Note: Drivers should NOT use this function directly, as it will break | |
850 | * platforms with CONFIG_DMABOUNCE. | |
851 | * Use the driver DMA support - see dma-mapping.h (dma_sync_*) | |
852 | */ | |
853 | static void __dma_page_cpu_to_dev(struct page *page, unsigned long off, | |
4ea0d737 RK |
854 | size_t size, enum dma_data_direction dir) |
855 | { | |
65af191a | 856 | unsigned long paddr; |
65af191a | 857 | |
a9c9147e | 858 | dma_cache_maint_page(page, off, size, dir, dmac_map_area); |
65af191a RK |
859 | |
860 | paddr = page_to_phys(page) + off; | |
2ffe2da3 RK |
861 | if (dir == DMA_FROM_DEVICE) { |
862 | outer_inv_range(paddr, paddr + size); | |
863 | } else { | |
864 | outer_clean_range(paddr, paddr + size); | |
865 | } | |
866 | /* FIXME: non-speculating: flush on bidirectional mappings? */ | |
4ea0d737 | 867 | } |
4ea0d737 | 868 | |
51fde349 | 869 | static void __dma_page_dev_to_cpu(struct page *page, unsigned long off, |
4ea0d737 RK |
870 | size_t size, enum dma_data_direction dir) |
871 | { | |
2ffe2da3 RK |
872 | unsigned long paddr = page_to_phys(page) + off; |
873 | ||
874 | /* FIXME: non-speculating: not required */ | |
875 | /* don't bother invalidating if DMA to device */ | |
876 | if (dir != DMA_TO_DEVICE) | |
877 | outer_inv_range(paddr, paddr + size); | |
878 | ||
a9c9147e | 879 | dma_cache_maint_page(page, off, size, dir, dmac_unmap_area); |
c0177800 CM |
880 | |
881 | /* | |
b2a234ed | 882 | * Mark the D-cache clean for these pages to avoid extra flushing. |
c0177800 | 883 | */ |
b2a234ed ML |
884 | if (dir != DMA_TO_DEVICE && size >= PAGE_SIZE) { |
885 | unsigned long pfn; | |
886 | size_t left = size; | |
887 | ||
888 | pfn = page_to_pfn(page) + off / PAGE_SIZE; | |
889 | off %= PAGE_SIZE; | |
890 | if (off) { | |
891 | pfn++; | |
892 | left -= PAGE_SIZE - off; | |
893 | } | |
894 | while (left >= PAGE_SIZE) { | |
895 | page = pfn_to_page(pfn++); | |
896 | set_bit(PG_dcache_clean, &page->flags); | |
897 | left -= PAGE_SIZE; | |
898 | } | |
899 | } | |
4ea0d737 | 900 | } |
43377453 | 901 | |
afd1a321 | 902 | /** |
2a550e73 | 903 | * arm_dma_map_sg - map a set of SG buffers for streaming mode DMA |
afd1a321 RK |
904 | * @dev: valid struct device pointer, or NULL for ISA and EISA-like devices |
905 | * @sg: list of buffers | |
906 | * @nents: number of buffers to map | |
907 | * @dir: DMA transfer direction | |
908 | * | |
909 | * Map a set of buffers described by scatterlist in streaming mode for DMA. | |
910 | * This is the scatter-gather version of the dma_map_single interface. | |
911 | * Here the scatter gather list elements are each tagged with the | |
912 | * appropriate dma address and length. They are obtained via | |
913 | * sg_dma_{address,length}. | |
914 | * | |
915 | * Device ownership issues as mentioned for dma_map_single are the same | |
916 | * here. | |
917 | */ | |
2dc6a016 MS |
918 | int arm_dma_map_sg(struct device *dev, struct scatterlist *sg, int nents, |
919 | enum dma_data_direction dir, struct dma_attrs *attrs) | |
afd1a321 | 920 | { |
2a550e73 | 921 | struct dma_map_ops *ops = get_dma_ops(dev); |
afd1a321 | 922 | struct scatterlist *s; |
01135d92 | 923 | int i, j; |
afd1a321 RK |
924 | |
925 | for_each_sg(sg, s, nents, i) { | |
4ce63fcd MS |
926 | #ifdef CONFIG_NEED_SG_DMA_LENGTH |
927 | s->dma_length = s->length; | |
928 | #endif | |
2a550e73 MS |
929 | s->dma_address = ops->map_page(dev, sg_page(s), s->offset, |
930 | s->length, dir, attrs); | |
01135d92 RK |
931 | if (dma_mapping_error(dev, s->dma_address)) |
932 | goto bad_mapping; | |
afd1a321 | 933 | } |
afd1a321 | 934 | return nents; |
01135d92 RK |
935 | |
936 | bad_mapping: | |
937 | for_each_sg(sg, s, i, j) | |
2a550e73 | 938 | ops->unmap_page(dev, sg_dma_address(s), sg_dma_len(s), dir, attrs); |
01135d92 | 939 | return 0; |
afd1a321 | 940 | } |
afd1a321 RK |
941 | |
942 | /** | |
2a550e73 | 943 | * arm_dma_unmap_sg - unmap a set of SG buffers mapped by dma_map_sg |
afd1a321 RK |
944 | * @dev: valid struct device pointer, or NULL for ISA and EISA-like devices |
945 | * @sg: list of buffers | |
0adfca6f | 946 | * @nents: number of buffers to unmap (same as was passed to dma_map_sg) |
afd1a321 RK |
947 | * @dir: DMA transfer direction (same as was passed to dma_map_sg) |
948 | * | |
949 | * Unmap a set of streaming mode DMA translations. Again, CPU access | |
950 | * rules concerning calls here are the same as for dma_unmap_single(). | |
951 | */ | |
2dc6a016 MS |
952 | void arm_dma_unmap_sg(struct device *dev, struct scatterlist *sg, int nents, |
953 | enum dma_data_direction dir, struct dma_attrs *attrs) | |
afd1a321 | 954 | { |
2a550e73 | 955 | struct dma_map_ops *ops = get_dma_ops(dev); |
01135d92 | 956 | struct scatterlist *s; |
01135d92 | 957 | |
01135d92 | 958 | int i; |
24056f52 | 959 | |
01135d92 | 960 | for_each_sg(sg, s, nents, i) |
2a550e73 | 961 | ops->unmap_page(dev, sg_dma_address(s), sg_dma_len(s), dir, attrs); |
afd1a321 | 962 | } |
afd1a321 RK |
963 | |
964 | /** | |
2a550e73 | 965 | * arm_dma_sync_sg_for_cpu |
afd1a321 RK |
966 | * @dev: valid struct device pointer, or NULL for ISA and EISA-like devices |
967 | * @sg: list of buffers | |
968 | * @nents: number of buffers to map (returned from dma_map_sg) | |
969 | * @dir: DMA transfer direction (same as was passed to dma_map_sg) | |
970 | */ | |
2dc6a016 | 971 | void arm_dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg, |
afd1a321 RK |
972 | int nents, enum dma_data_direction dir) |
973 | { | |
2a550e73 | 974 | struct dma_map_ops *ops = get_dma_ops(dev); |
afd1a321 RK |
975 | struct scatterlist *s; |
976 | int i; | |
977 | ||
2a550e73 MS |
978 | for_each_sg(sg, s, nents, i) |
979 | ops->sync_single_for_cpu(dev, sg_dma_address(s), s->length, | |
980 | dir); | |
afd1a321 | 981 | } |
afd1a321 RK |
982 | |
983 | /** | |
2a550e73 | 984 | * arm_dma_sync_sg_for_device |
afd1a321 RK |
985 | * @dev: valid struct device pointer, or NULL for ISA and EISA-like devices |
986 | * @sg: list of buffers | |
987 | * @nents: number of buffers to map (returned from dma_map_sg) | |
988 | * @dir: DMA transfer direction (same as was passed to dma_map_sg) | |
989 | */ | |
2dc6a016 | 990 | void arm_dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg, |
afd1a321 RK |
991 | int nents, enum dma_data_direction dir) |
992 | { | |
2a550e73 | 993 | struct dma_map_ops *ops = get_dma_ops(dev); |
afd1a321 RK |
994 | struct scatterlist *s; |
995 | int i; | |
996 | ||
2a550e73 MS |
997 | for_each_sg(sg, s, nents, i) |
998 | ops->sync_single_for_device(dev, sg_dma_address(s), s->length, | |
999 | dir); | |
afd1a321 | 1000 | } |
24056f52 | 1001 | |
022ae537 RK |
1002 | /* |
1003 | * Return whether the given device DMA address mask can be supported | |
1004 | * properly. For example, if your device can only drive the low 24-bits | |
1005 | * during bus mastering, then you would pass 0x00ffffff as the mask | |
1006 | * to this function. | |
1007 | */ | |
1008 | int dma_supported(struct device *dev, u64 mask) | |
1009 | { | |
1010 | if (mask < (u64)arm_dma_limit) | |
1011 | return 0; | |
1012 | return 1; | |
1013 | } | |
1014 | EXPORT_SYMBOL(dma_supported); | |
1015 | ||
87b54e78 | 1016 | int arm_dma_set_mask(struct device *dev, u64 dma_mask) |
022ae537 RK |
1017 | { |
1018 | if (!dev->dma_mask || !dma_supported(dev, dma_mask)) | |
1019 | return -EIO; | |
1020 | ||
022ae537 | 1021 | *dev->dma_mask = dma_mask; |
022ae537 RK |
1022 | |
1023 | return 0; | |
1024 | } | |
022ae537 | 1025 | |
24056f52 RK |
1026 | #define PREALLOC_DMA_DEBUG_ENTRIES 4096 |
1027 | ||
1028 | static int __init dma_debug_do_init(void) | |
1029 | { | |
1030 | dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES); | |
1031 | return 0; | |
1032 | } | |
1033 | fs_initcall(dma_debug_do_init); | |
4ce63fcd MS |
1034 | |
1035 | #ifdef CONFIG_ARM_DMA_USE_IOMMU | |
1036 | ||
1037 | /* IOMMU */ | |
1038 | ||
1039 | static inline dma_addr_t __alloc_iova(struct dma_iommu_mapping *mapping, | |
1040 | size_t size) | |
1041 | { | |
1042 | unsigned int order = get_order(size); | |
1043 | unsigned int align = 0; | |
1044 | unsigned int count, start; | |
1045 | unsigned long flags; | |
1046 | ||
60460abf SWK |
1047 | if (order > CONFIG_ARM_DMA_IOMMU_ALIGNMENT) |
1048 | order = CONFIG_ARM_DMA_IOMMU_ALIGNMENT; | |
1049 | ||
4ce63fcd MS |
1050 | count = ((PAGE_ALIGN(size) >> PAGE_SHIFT) + |
1051 | (1 << mapping->order) - 1) >> mapping->order; | |
1052 | ||
1053 | if (order > mapping->order) | |
1054 | align = (1 << (order - mapping->order)) - 1; | |
1055 | ||
1056 | spin_lock_irqsave(&mapping->lock, flags); | |
1057 | start = bitmap_find_next_zero_area(mapping->bitmap, mapping->bits, 0, | |
1058 | count, align); | |
1059 | if (start > mapping->bits) { | |
1060 | spin_unlock_irqrestore(&mapping->lock, flags); | |
1061 | return DMA_ERROR_CODE; | |
1062 | } | |
1063 | ||
1064 | bitmap_set(mapping->bitmap, start, count); | |
1065 | spin_unlock_irqrestore(&mapping->lock, flags); | |
1066 | ||
1067 | return mapping->base + (start << (mapping->order + PAGE_SHIFT)); | |
1068 | } | |
1069 | ||
1070 | static inline void __free_iova(struct dma_iommu_mapping *mapping, | |
1071 | dma_addr_t addr, size_t size) | |
1072 | { | |
1073 | unsigned int start = (addr - mapping->base) >> | |
1074 | (mapping->order + PAGE_SHIFT); | |
1075 | unsigned int count = ((size >> PAGE_SHIFT) + | |
1076 | (1 << mapping->order) - 1) >> mapping->order; | |
1077 | unsigned long flags; | |
1078 | ||
1079 | spin_lock_irqsave(&mapping->lock, flags); | |
1080 | bitmap_clear(mapping->bitmap, start, count); | |
1081 | spin_unlock_irqrestore(&mapping->lock, flags); | |
1082 | } | |
1083 | ||
549a17e4 MS |
1084 | static struct page **__iommu_alloc_buffer(struct device *dev, size_t size, |
1085 | gfp_t gfp, struct dma_attrs *attrs) | |
4ce63fcd MS |
1086 | { |
1087 | struct page **pages; | |
1088 | int count = size >> PAGE_SHIFT; | |
1089 | int array_size = count * sizeof(struct page *); | |
1090 | int i = 0; | |
1091 | ||
1092 | if (array_size <= PAGE_SIZE) | |
1093 | pages = kzalloc(array_size, gfp); | |
1094 | else | |
1095 | pages = vzalloc(array_size); | |
1096 | if (!pages) | |
1097 | return NULL; | |
1098 | ||
549a17e4 MS |
1099 | if (dma_get_attr(DMA_ATTR_FORCE_CONTIGUOUS, attrs)) |
1100 | { | |
1101 | unsigned long order = get_order(size); | |
1102 | struct page *page; | |
1103 | ||
1104 | page = dma_alloc_from_contiguous(dev, count, order); | |
1105 | if (!page) | |
1106 | goto error; | |
1107 | ||
1108 | __dma_clear_buffer(page, size); | |
1109 | ||
1110 | for (i = 0; i < count; i++) | |
1111 | pages[i] = page + i; | |
1112 | ||
1113 | return pages; | |
1114 | } | |
1115 | ||
f8669bef MS |
1116 | /* |
1117 | * IOMMU can map any pages, so himem can also be used here | |
1118 | */ | |
1119 | gfp |= __GFP_NOWARN | __GFP_HIGHMEM; | |
1120 | ||
4ce63fcd | 1121 | while (count) { |
593f4735 | 1122 | int j, order = __fls(count); |
4ce63fcd | 1123 | |
f8669bef | 1124 | pages[i] = alloc_pages(gfp, order); |
4ce63fcd | 1125 | while (!pages[i] && order) |
f8669bef | 1126 | pages[i] = alloc_pages(gfp, --order); |
4ce63fcd MS |
1127 | if (!pages[i]) |
1128 | goto error; | |
1129 | ||
5a796eeb | 1130 | if (order) { |
4ce63fcd | 1131 | split_page(pages[i], order); |
5a796eeb HD |
1132 | j = 1 << order; |
1133 | while (--j) | |
1134 | pages[i + j] = pages[i] + j; | |
1135 | } | |
4ce63fcd MS |
1136 | |
1137 | __dma_clear_buffer(pages[i], PAGE_SIZE << order); | |
1138 | i += 1 << order; | |
1139 | count -= 1 << order; | |
1140 | } | |
1141 | ||
1142 | return pages; | |
1143 | error: | |
9fa8af91 | 1144 | while (i--) |
4ce63fcd MS |
1145 | if (pages[i]) |
1146 | __free_pages(pages[i], 0); | |
46c87852 | 1147 | if (array_size <= PAGE_SIZE) |
4ce63fcd MS |
1148 | kfree(pages); |
1149 | else | |
1150 | vfree(pages); | |
1151 | return NULL; | |
1152 | } | |
1153 | ||
549a17e4 MS |
1154 | static int __iommu_free_buffer(struct device *dev, struct page **pages, |
1155 | size_t size, struct dma_attrs *attrs) | |
4ce63fcd MS |
1156 | { |
1157 | int count = size >> PAGE_SHIFT; | |
1158 | int array_size = count * sizeof(struct page *); | |
1159 | int i; | |
549a17e4 MS |
1160 | |
1161 | if (dma_get_attr(DMA_ATTR_FORCE_CONTIGUOUS, attrs)) { | |
1162 | dma_release_from_contiguous(dev, pages[0], count); | |
1163 | } else { | |
1164 | for (i = 0; i < count; i++) | |
1165 | if (pages[i]) | |
1166 | __free_pages(pages[i], 0); | |
1167 | } | |
1168 | ||
46c87852 | 1169 | if (array_size <= PAGE_SIZE) |
4ce63fcd MS |
1170 | kfree(pages); |
1171 | else | |
1172 | vfree(pages); | |
1173 | return 0; | |
1174 | } | |
1175 | ||
1176 | /* | |
1177 | * Create a CPU mapping for a specified pages | |
1178 | */ | |
1179 | static void * | |
e9da6e99 MS |
1180 | __iommu_alloc_remap(struct page **pages, size_t size, gfp_t gfp, pgprot_t prot, |
1181 | const void *caller) | |
4ce63fcd | 1182 | { |
e9da6e99 MS |
1183 | unsigned int i, nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT; |
1184 | struct vm_struct *area; | |
1185 | unsigned long p; | |
4ce63fcd | 1186 | |
e9da6e99 MS |
1187 | area = get_vm_area_caller(size, VM_ARM_DMA_CONSISTENT | VM_USERMAP, |
1188 | caller); | |
1189 | if (!area) | |
4ce63fcd | 1190 | return NULL; |
4ce63fcd | 1191 | |
e9da6e99 MS |
1192 | area->pages = pages; |
1193 | area->nr_pages = nr_pages; | |
1194 | p = (unsigned long)area->addr; | |
4ce63fcd | 1195 | |
e9da6e99 MS |
1196 | for (i = 0; i < nr_pages; i++) { |
1197 | phys_addr_t phys = __pfn_to_phys(page_to_pfn(pages[i])); | |
1198 | if (ioremap_page_range(p, p + PAGE_SIZE, phys, prot)) | |
1199 | goto err; | |
1200 | p += PAGE_SIZE; | |
4ce63fcd | 1201 | } |
e9da6e99 MS |
1202 | return area->addr; |
1203 | err: | |
1204 | unmap_kernel_range((unsigned long)area->addr, size); | |
1205 | vunmap(area->addr); | |
4ce63fcd MS |
1206 | return NULL; |
1207 | } | |
1208 | ||
1209 | /* | |
1210 | * Create a mapping in device IO address space for specified pages | |
1211 | */ | |
1212 | static dma_addr_t | |
1213 | __iommu_create_mapping(struct device *dev, struct page **pages, size_t size) | |
1214 | { | |
1215 | struct dma_iommu_mapping *mapping = dev->archdata.mapping; | |
1216 | unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT; | |
1217 | dma_addr_t dma_addr, iova; | |
1218 | int i, ret = DMA_ERROR_CODE; | |
1219 | ||
1220 | dma_addr = __alloc_iova(mapping, size); | |
1221 | if (dma_addr == DMA_ERROR_CODE) | |
1222 | return dma_addr; | |
1223 | ||
1224 | iova = dma_addr; | |
1225 | for (i = 0; i < count; ) { | |
1226 | unsigned int next_pfn = page_to_pfn(pages[i]) + 1; | |
1227 | phys_addr_t phys = page_to_phys(pages[i]); | |
1228 | unsigned int len, j; | |
1229 | ||
1230 | for (j = i + 1; j < count; j++, next_pfn++) | |
1231 | if (page_to_pfn(pages[j]) != next_pfn) | |
1232 | break; | |
1233 | ||
1234 | len = (j - i) << PAGE_SHIFT; | |
1235 | ret = iommu_map(mapping->domain, iova, phys, len, 0); | |
1236 | if (ret < 0) | |
1237 | goto fail; | |
1238 | iova += len; | |
1239 | i = j; | |
1240 | } | |
1241 | return dma_addr; | |
1242 | fail: | |
1243 | iommu_unmap(mapping->domain, dma_addr, iova-dma_addr); | |
1244 | __free_iova(mapping, dma_addr, size); | |
1245 | return DMA_ERROR_CODE; | |
1246 | } | |
1247 | ||
1248 | static int __iommu_remove_mapping(struct device *dev, dma_addr_t iova, size_t size) | |
1249 | { | |
1250 | struct dma_iommu_mapping *mapping = dev->archdata.mapping; | |
1251 | ||
1252 | /* | |
1253 | * add optional in-page offset from iova to size and align | |
1254 | * result to page size | |
1255 | */ | |
1256 | size = PAGE_ALIGN((iova & ~PAGE_MASK) + size); | |
1257 | iova &= PAGE_MASK; | |
1258 | ||
1259 | iommu_unmap(mapping->domain, iova, size); | |
1260 | __free_iova(mapping, iova, size); | |
1261 | return 0; | |
1262 | } | |
1263 | ||
665bad7b HD |
1264 | static struct page **__atomic_get_pages(void *addr) |
1265 | { | |
1266 | struct dma_pool *pool = &atomic_pool; | |
1267 | struct page **pages = pool->pages; | |
1268 | int offs = (addr - pool->vaddr) >> PAGE_SHIFT; | |
1269 | ||
1270 | return pages + offs; | |
1271 | } | |
1272 | ||
955c757e | 1273 | static struct page **__iommu_get_pages(void *cpu_addr, struct dma_attrs *attrs) |
e9da6e99 MS |
1274 | { |
1275 | struct vm_struct *area; | |
1276 | ||
665bad7b HD |
1277 | if (__in_atomic_pool(cpu_addr, PAGE_SIZE)) |
1278 | return __atomic_get_pages(cpu_addr); | |
1279 | ||
955c757e MS |
1280 | if (dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING, attrs)) |
1281 | return cpu_addr; | |
1282 | ||
e9da6e99 MS |
1283 | area = find_vm_area(cpu_addr); |
1284 | if (area && (area->flags & VM_ARM_DMA_CONSISTENT)) | |
1285 | return area->pages; | |
1286 | return NULL; | |
1287 | } | |
1288 | ||
479ed93a HD |
1289 | static void *__iommu_alloc_atomic(struct device *dev, size_t size, |
1290 | dma_addr_t *handle) | |
1291 | { | |
1292 | struct page *page; | |
1293 | void *addr; | |
1294 | ||
1295 | addr = __alloc_from_pool(size, &page); | |
1296 | if (!addr) | |
1297 | return NULL; | |
1298 | ||
1299 | *handle = __iommu_create_mapping(dev, &page, size); | |
1300 | if (*handle == DMA_ERROR_CODE) | |
1301 | goto err_mapping; | |
1302 | ||
1303 | return addr; | |
1304 | ||
1305 | err_mapping: | |
1306 | __free_from_pool(addr, size); | |
1307 | return NULL; | |
1308 | } | |
1309 | ||
d5898291 | 1310 | static void __iommu_free_atomic(struct device *dev, void *cpu_addr, |
479ed93a HD |
1311 | dma_addr_t handle, size_t size) |
1312 | { | |
1313 | __iommu_remove_mapping(dev, handle, size); | |
d5898291 | 1314 | __free_from_pool(cpu_addr, size); |
479ed93a HD |
1315 | } |
1316 | ||
4ce63fcd MS |
1317 | static void *arm_iommu_alloc_attrs(struct device *dev, size_t size, |
1318 | dma_addr_t *handle, gfp_t gfp, struct dma_attrs *attrs) | |
1319 | { | |
1320 | pgprot_t prot = __get_dma_pgprot(attrs, pgprot_kernel); | |
1321 | struct page **pages; | |
1322 | void *addr = NULL; | |
1323 | ||
1324 | *handle = DMA_ERROR_CODE; | |
1325 | size = PAGE_ALIGN(size); | |
1326 | ||
479ed93a HD |
1327 | if (gfp & GFP_ATOMIC) |
1328 | return __iommu_alloc_atomic(dev, size, handle); | |
1329 | ||
5b91a98c RZ |
1330 | /* |
1331 | * Following is a work-around (a.k.a. hack) to prevent pages | |
1332 | * with __GFP_COMP being passed to split_page() which cannot | |
1333 | * handle them. The real problem is that this flag probably | |
1334 | * should be 0 on ARM as it is not supported on this | |
1335 | * platform; see CONFIG_HUGETLBFS. | |
1336 | */ | |
1337 | gfp &= ~(__GFP_COMP); | |
1338 | ||
549a17e4 | 1339 | pages = __iommu_alloc_buffer(dev, size, gfp, attrs); |
4ce63fcd MS |
1340 | if (!pages) |
1341 | return NULL; | |
1342 | ||
1343 | *handle = __iommu_create_mapping(dev, pages, size); | |
1344 | if (*handle == DMA_ERROR_CODE) | |
1345 | goto err_buffer; | |
1346 | ||
955c757e MS |
1347 | if (dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING, attrs)) |
1348 | return pages; | |
1349 | ||
e9da6e99 MS |
1350 | addr = __iommu_alloc_remap(pages, size, gfp, prot, |
1351 | __builtin_return_address(0)); | |
4ce63fcd MS |
1352 | if (!addr) |
1353 | goto err_mapping; | |
1354 | ||
1355 | return addr; | |
1356 | ||
1357 | err_mapping: | |
1358 | __iommu_remove_mapping(dev, *handle, size); | |
1359 | err_buffer: | |
549a17e4 | 1360 | __iommu_free_buffer(dev, pages, size, attrs); |
4ce63fcd MS |
1361 | return NULL; |
1362 | } | |
1363 | ||
1364 | static int arm_iommu_mmap_attrs(struct device *dev, struct vm_area_struct *vma, | |
1365 | void *cpu_addr, dma_addr_t dma_addr, size_t size, | |
1366 | struct dma_attrs *attrs) | |
1367 | { | |
e9da6e99 MS |
1368 | unsigned long uaddr = vma->vm_start; |
1369 | unsigned long usize = vma->vm_end - vma->vm_start; | |
955c757e | 1370 | struct page **pages = __iommu_get_pages(cpu_addr, attrs); |
4ce63fcd MS |
1371 | |
1372 | vma->vm_page_prot = __get_dma_pgprot(attrs, vma->vm_page_prot); | |
4ce63fcd | 1373 | |
e9da6e99 MS |
1374 | if (!pages) |
1375 | return -ENXIO; | |
4ce63fcd | 1376 | |
e9da6e99 MS |
1377 | do { |
1378 | int ret = vm_insert_page(vma, uaddr, *pages++); | |
1379 | if (ret) { | |
1380 | pr_err("Remapping memory failed: %d\n", ret); | |
1381 | return ret; | |
1382 | } | |
1383 | uaddr += PAGE_SIZE; | |
1384 | usize -= PAGE_SIZE; | |
1385 | } while (usize > 0); | |
4ce63fcd | 1386 | |
4ce63fcd MS |
1387 | return 0; |
1388 | } | |
1389 | ||
1390 | /* | |
1391 | * free a page as defined by the above mapping. | |
1392 | * Must not be called with IRQs disabled. | |
1393 | */ | |
1394 | void arm_iommu_free_attrs(struct device *dev, size_t size, void *cpu_addr, | |
1395 | dma_addr_t handle, struct dma_attrs *attrs) | |
1396 | { | |
836bfa0d | 1397 | struct page **pages; |
4ce63fcd MS |
1398 | size = PAGE_ALIGN(size); |
1399 | ||
836bfa0d YC |
1400 | if (__in_atomic_pool(cpu_addr, size)) { |
1401 | __iommu_free_atomic(dev, cpu_addr, handle, size); | |
e9da6e99 | 1402 | return; |
4ce63fcd | 1403 | } |
e9da6e99 | 1404 | |
836bfa0d YC |
1405 | pages = __iommu_get_pages(cpu_addr, attrs); |
1406 | if (!pages) { | |
1407 | WARN(1, "trying to free invalid coherent area: %p\n", cpu_addr); | |
479ed93a HD |
1408 | return; |
1409 | } | |
1410 | ||
955c757e MS |
1411 | if (!dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING, attrs)) { |
1412 | unmap_kernel_range((unsigned long)cpu_addr, size); | |
1413 | vunmap(cpu_addr); | |
1414 | } | |
e9da6e99 MS |
1415 | |
1416 | __iommu_remove_mapping(dev, handle, size); | |
549a17e4 | 1417 | __iommu_free_buffer(dev, pages, size, attrs); |
4ce63fcd MS |
1418 | } |
1419 | ||
dc2832e1 MS |
1420 | static int arm_iommu_get_sgtable(struct device *dev, struct sg_table *sgt, |
1421 | void *cpu_addr, dma_addr_t dma_addr, | |
1422 | size_t size, struct dma_attrs *attrs) | |
1423 | { | |
1424 | unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT; | |
1425 | struct page **pages = __iommu_get_pages(cpu_addr, attrs); | |
1426 | ||
1427 | if (!pages) | |
1428 | return -ENXIO; | |
1429 | ||
1430 | return sg_alloc_table_from_pages(sgt, pages, count, 0, size, | |
1431 | GFP_KERNEL); | |
4ce63fcd MS |
1432 | } |
1433 | ||
1434 | /* | |
1435 | * Map a part of the scatter-gather list into contiguous io address space | |
1436 | */ | |
1437 | static int __map_sg_chunk(struct device *dev, struct scatterlist *sg, | |
1438 | size_t size, dma_addr_t *handle, | |
0fa478df RH |
1439 | enum dma_data_direction dir, struct dma_attrs *attrs, |
1440 | bool is_coherent) | |
4ce63fcd MS |
1441 | { |
1442 | struct dma_iommu_mapping *mapping = dev->archdata.mapping; | |
1443 | dma_addr_t iova, iova_base; | |
1444 | int ret = 0; | |
1445 | unsigned int count; | |
1446 | struct scatterlist *s; | |
1447 | ||
1448 | size = PAGE_ALIGN(size); | |
1449 | *handle = DMA_ERROR_CODE; | |
1450 | ||
1451 | iova_base = iova = __alloc_iova(mapping, size); | |
1452 | if (iova == DMA_ERROR_CODE) | |
1453 | return -ENOMEM; | |
1454 | ||
1455 | for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { | |
1456 | phys_addr_t phys = page_to_phys(sg_page(s)); | |
1457 | unsigned int len = PAGE_ALIGN(s->offset + s->length); | |
1458 | ||
0fa478df RH |
1459 | if (!is_coherent && |
1460 | !dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) | |
4ce63fcd MS |
1461 | __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, dir); |
1462 | ||
1463 | ret = iommu_map(mapping->domain, iova, phys, len, 0); | |
1464 | if (ret < 0) | |
1465 | goto fail; | |
1466 | count += len >> PAGE_SHIFT; | |
1467 | iova += len; | |
1468 | } | |
1469 | *handle = iova_base; | |
1470 | ||
1471 | return 0; | |
1472 | fail: | |
1473 | iommu_unmap(mapping->domain, iova_base, count * PAGE_SIZE); | |
1474 | __free_iova(mapping, iova_base, size); | |
1475 | return ret; | |
1476 | } | |
1477 | ||
0fa478df RH |
1478 | static int __iommu_map_sg(struct device *dev, struct scatterlist *sg, int nents, |
1479 | enum dma_data_direction dir, struct dma_attrs *attrs, | |
1480 | bool is_coherent) | |
4ce63fcd MS |
1481 | { |
1482 | struct scatterlist *s = sg, *dma = sg, *start = sg; | |
1483 | int i, count = 0; | |
1484 | unsigned int offset = s->offset; | |
1485 | unsigned int size = s->offset + s->length; | |
1486 | unsigned int max = dma_get_max_seg_size(dev); | |
1487 | ||
1488 | for (i = 1; i < nents; i++) { | |
1489 | s = sg_next(s); | |
1490 | ||
1491 | s->dma_address = DMA_ERROR_CODE; | |
1492 | s->dma_length = 0; | |
1493 | ||
1494 | if (s->offset || (size & ~PAGE_MASK) || size + s->length > max) { | |
1495 | if (__map_sg_chunk(dev, start, size, &dma->dma_address, | |
0fa478df | 1496 | dir, attrs, is_coherent) < 0) |
4ce63fcd MS |
1497 | goto bad_mapping; |
1498 | ||
1499 | dma->dma_address += offset; | |
1500 | dma->dma_length = size - offset; | |
1501 | ||
1502 | size = offset = s->offset; | |
1503 | start = s; | |
1504 | dma = sg_next(dma); | |
1505 | count += 1; | |
1506 | } | |
1507 | size += s->length; | |
1508 | } | |
0fa478df RH |
1509 | if (__map_sg_chunk(dev, start, size, &dma->dma_address, dir, attrs, |
1510 | is_coherent) < 0) | |
4ce63fcd MS |
1511 | goto bad_mapping; |
1512 | ||
1513 | dma->dma_address += offset; | |
1514 | dma->dma_length = size - offset; | |
1515 | ||
1516 | return count+1; | |
1517 | ||
1518 | bad_mapping: | |
1519 | for_each_sg(sg, s, count, i) | |
1520 | __iommu_remove_mapping(dev, sg_dma_address(s), sg_dma_len(s)); | |
1521 | return 0; | |
1522 | } | |
1523 | ||
1524 | /** | |
0fa478df | 1525 | * arm_coherent_iommu_map_sg - map a set of SG buffers for streaming mode DMA |
4ce63fcd MS |
1526 | * @dev: valid struct device pointer |
1527 | * @sg: list of buffers | |
0fa478df RH |
1528 | * @nents: number of buffers to map |
1529 | * @dir: DMA transfer direction | |
4ce63fcd | 1530 | * |
0fa478df RH |
1531 | * Map a set of i/o coherent buffers described by scatterlist in streaming |
1532 | * mode for DMA. The scatter gather list elements are merged together (if | |
1533 | * possible) and tagged with the appropriate dma address and length. They are | |
1534 | * obtained via sg_dma_{address,length}. | |
4ce63fcd | 1535 | */ |
0fa478df RH |
1536 | int arm_coherent_iommu_map_sg(struct device *dev, struct scatterlist *sg, |
1537 | int nents, enum dma_data_direction dir, struct dma_attrs *attrs) | |
1538 | { | |
1539 | return __iommu_map_sg(dev, sg, nents, dir, attrs, true); | |
1540 | } | |
1541 | ||
1542 | /** | |
1543 | * arm_iommu_map_sg - map a set of SG buffers for streaming mode DMA | |
1544 | * @dev: valid struct device pointer | |
1545 | * @sg: list of buffers | |
1546 | * @nents: number of buffers to map | |
1547 | * @dir: DMA transfer direction | |
1548 | * | |
1549 | * Map a set of buffers described by scatterlist in streaming mode for DMA. | |
1550 | * The scatter gather list elements are merged together (if possible) and | |
1551 | * tagged with the appropriate dma address and length. They are obtained via | |
1552 | * sg_dma_{address,length}. | |
1553 | */ | |
1554 | int arm_iommu_map_sg(struct device *dev, struct scatterlist *sg, | |
1555 | int nents, enum dma_data_direction dir, struct dma_attrs *attrs) | |
1556 | { | |
1557 | return __iommu_map_sg(dev, sg, nents, dir, attrs, false); | |
1558 | } | |
1559 | ||
1560 | static void __iommu_unmap_sg(struct device *dev, struct scatterlist *sg, | |
1561 | int nents, enum dma_data_direction dir, struct dma_attrs *attrs, | |
1562 | bool is_coherent) | |
4ce63fcd MS |
1563 | { |
1564 | struct scatterlist *s; | |
1565 | int i; | |
1566 | ||
1567 | for_each_sg(sg, s, nents, i) { | |
1568 | if (sg_dma_len(s)) | |
1569 | __iommu_remove_mapping(dev, sg_dma_address(s), | |
1570 | sg_dma_len(s)); | |
0fa478df | 1571 | if (!is_coherent && |
97ef952a | 1572 | !dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) |
4ce63fcd MS |
1573 | __dma_page_dev_to_cpu(sg_page(s), s->offset, |
1574 | s->length, dir); | |
1575 | } | |
1576 | } | |
1577 | ||
0fa478df RH |
1578 | /** |
1579 | * arm_coherent_iommu_unmap_sg - unmap a set of SG buffers mapped by dma_map_sg | |
1580 | * @dev: valid struct device pointer | |
1581 | * @sg: list of buffers | |
1582 | * @nents: number of buffers to unmap (same as was passed to dma_map_sg) | |
1583 | * @dir: DMA transfer direction (same as was passed to dma_map_sg) | |
1584 | * | |
1585 | * Unmap a set of streaming mode DMA translations. Again, CPU access | |
1586 | * rules concerning calls here are the same as for dma_unmap_single(). | |
1587 | */ | |
1588 | void arm_coherent_iommu_unmap_sg(struct device *dev, struct scatterlist *sg, | |
1589 | int nents, enum dma_data_direction dir, struct dma_attrs *attrs) | |
1590 | { | |
1591 | __iommu_unmap_sg(dev, sg, nents, dir, attrs, true); | |
1592 | } | |
1593 | ||
1594 | /** | |
1595 | * arm_iommu_unmap_sg - unmap a set of SG buffers mapped by dma_map_sg | |
1596 | * @dev: valid struct device pointer | |
1597 | * @sg: list of buffers | |
1598 | * @nents: number of buffers to unmap (same as was passed to dma_map_sg) | |
1599 | * @dir: DMA transfer direction (same as was passed to dma_map_sg) | |
1600 | * | |
1601 | * Unmap a set of streaming mode DMA translations. Again, CPU access | |
1602 | * rules concerning calls here are the same as for dma_unmap_single(). | |
1603 | */ | |
1604 | void arm_iommu_unmap_sg(struct device *dev, struct scatterlist *sg, int nents, | |
1605 | enum dma_data_direction dir, struct dma_attrs *attrs) | |
1606 | { | |
1607 | __iommu_unmap_sg(dev, sg, nents, dir, attrs, false); | |
1608 | } | |
1609 | ||
4ce63fcd MS |
1610 | /** |
1611 | * arm_iommu_sync_sg_for_cpu | |
1612 | * @dev: valid struct device pointer | |
1613 | * @sg: list of buffers | |
1614 | * @nents: number of buffers to map (returned from dma_map_sg) | |
1615 | * @dir: DMA transfer direction (same as was passed to dma_map_sg) | |
1616 | */ | |
1617 | void arm_iommu_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg, | |
1618 | int nents, enum dma_data_direction dir) | |
1619 | { | |
1620 | struct scatterlist *s; | |
1621 | int i; | |
1622 | ||
1623 | for_each_sg(sg, s, nents, i) | |
0fa478df | 1624 | __dma_page_dev_to_cpu(sg_page(s), s->offset, s->length, dir); |
4ce63fcd MS |
1625 | |
1626 | } | |
1627 | ||
1628 | /** | |
1629 | * arm_iommu_sync_sg_for_device | |
1630 | * @dev: valid struct device pointer | |
1631 | * @sg: list of buffers | |
1632 | * @nents: number of buffers to map (returned from dma_map_sg) | |
1633 | * @dir: DMA transfer direction (same as was passed to dma_map_sg) | |
1634 | */ | |
1635 | void arm_iommu_sync_sg_for_device(struct device *dev, struct scatterlist *sg, | |
1636 | int nents, enum dma_data_direction dir) | |
1637 | { | |
1638 | struct scatterlist *s; | |
1639 | int i; | |
1640 | ||
1641 | for_each_sg(sg, s, nents, i) | |
0fa478df | 1642 | __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, dir); |
4ce63fcd MS |
1643 | } |
1644 | ||
1645 | ||
1646 | /** | |
0fa478df | 1647 | * arm_coherent_iommu_map_page |
4ce63fcd MS |
1648 | * @dev: valid struct device pointer |
1649 | * @page: page that buffer resides in | |
1650 | * @offset: offset into page for start of buffer | |
1651 | * @size: size of buffer to map | |
1652 | * @dir: DMA transfer direction | |
1653 | * | |
0fa478df | 1654 | * Coherent IOMMU aware version of arm_dma_map_page() |
4ce63fcd | 1655 | */ |
0fa478df | 1656 | static dma_addr_t arm_coherent_iommu_map_page(struct device *dev, struct page *page, |
4ce63fcd MS |
1657 | unsigned long offset, size_t size, enum dma_data_direction dir, |
1658 | struct dma_attrs *attrs) | |
1659 | { | |
1660 | struct dma_iommu_mapping *mapping = dev->archdata.mapping; | |
1661 | dma_addr_t dma_addr; | |
13987d68 | 1662 | int ret, prot, len = PAGE_ALIGN(size + offset); |
4ce63fcd | 1663 | |
4ce63fcd MS |
1664 | dma_addr = __alloc_iova(mapping, len); |
1665 | if (dma_addr == DMA_ERROR_CODE) | |
1666 | return dma_addr; | |
1667 | ||
13987d68 WD |
1668 | switch (dir) { |
1669 | case DMA_BIDIRECTIONAL: | |
1670 | prot = IOMMU_READ | IOMMU_WRITE; | |
1671 | break; | |
1672 | case DMA_TO_DEVICE: | |
1673 | prot = IOMMU_READ; | |
1674 | break; | |
1675 | case DMA_FROM_DEVICE: | |
1676 | prot = IOMMU_WRITE; | |
1677 | break; | |
1678 | default: | |
1679 | prot = 0; | |
1680 | } | |
1681 | ||
1682 | ret = iommu_map(mapping->domain, dma_addr, page_to_phys(page), len, prot); | |
4ce63fcd MS |
1683 | if (ret < 0) |
1684 | goto fail; | |
1685 | ||
1686 | return dma_addr + offset; | |
1687 | fail: | |
1688 | __free_iova(mapping, dma_addr, len); | |
1689 | return DMA_ERROR_CODE; | |
1690 | } | |
1691 | ||
0fa478df RH |
1692 | /** |
1693 | * arm_iommu_map_page | |
1694 | * @dev: valid struct device pointer | |
1695 | * @page: page that buffer resides in | |
1696 | * @offset: offset into page for start of buffer | |
1697 | * @size: size of buffer to map | |
1698 | * @dir: DMA transfer direction | |
1699 | * | |
1700 | * IOMMU aware version of arm_dma_map_page() | |
1701 | */ | |
1702 | static dma_addr_t arm_iommu_map_page(struct device *dev, struct page *page, | |
1703 | unsigned long offset, size_t size, enum dma_data_direction dir, | |
1704 | struct dma_attrs *attrs) | |
1705 | { | |
1706 | if (!dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) | |
1707 | __dma_page_cpu_to_dev(page, offset, size, dir); | |
1708 | ||
1709 | return arm_coherent_iommu_map_page(dev, page, offset, size, dir, attrs); | |
1710 | } | |
1711 | ||
1712 | /** | |
1713 | * arm_coherent_iommu_unmap_page | |
1714 | * @dev: valid struct device pointer | |
1715 | * @handle: DMA address of buffer | |
1716 | * @size: size of buffer (same as passed to dma_map_page) | |
1717 | * @dir: DMA transfer direction (same as passed to dma_map_page) | |
1718 | * | |
1719 | * Coherent IOMMU aware version of arm_dma_unmap_page() | |
1720 | */ | |
1721 | static void arm_coherent_iommu_unmap_page(struct device *dev, dma_addr_t handle, | |
1722 | size_t size, enum dma_data_direction dir, | |
1723 | struct dma_attrs *attrs) | |
1724 | { | |
1725 | struct dma_iommu_mapping *mapping = dev->archdata.mapping; | |
1726 | dma_addr_t iova = handle & PAGE_MASK; | |
0fa478df RH |
1727 | int offset = handle & ~PAGE_MASK; |
1728 | int len = PAGE_ALIGN(size + offset); | |
1729 | ||
1730 | if (!iova) | |
1731 | return; | |
1732 | ||
1733 | iommu_unmap(mapping->domain, iova, len); | |
1734 | __free_iova(mapping, iova, len); | |
1735 | } | |
1736 | ||
4ce63fcd MS |
1737 | /** |
1738 | * arm_iommu_unmap_page | |
1739 | * @dev: valid struct device pointer | |
1740 | * @handle: DMA address of buffer | |
1741 | * @size: size of buffer (same as passed to dma_map_page) | |
1742 | * @dir: DMA transfer direction (same as passed to dma_map_page) | |
1743 | * | |
1744 | * IOMMU aware version of arm_dma_unmap_page() | |
1745 | */ | |
1746 | static void arm_iommu_unmap_page(struct device *dev, dma_addr_t handle, | |
1747 | size_t size, enum dma_data_direction dir, | |
1748 | struct dma_attrs *attrs) | |
1749 | { | |
1750 | struct dma_iommu_mapping *mapping = dev->archdata.mapping; | |
1751 | dma_addr_t iova = handle & PAGE_MASK; | |
1752 | struct page *page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova)); | |
1753 | int offset = handle & ~PAGE_MASK; | |
1754 | int len = PAGE_ALIGN(size + offset); | |
1755 | ||
1756 | if (!iova) | |
1757 | return; | |
1758 | ||
0fa478df | 1759 | if (!dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) |
4ce63fcd MS |
1760 | __dma_page_dev_to_cpu(page, offset, size, dir); |
1761 | ||
1762 | iommu_unmap(mapping->domain, iova, len); | |
1763 | __free_iova(mapping, iova, len); | |
1764 | } | |
1765 | ||
1766 | static void arm_iommu_sync_single_for_cpu(struct device *dev, | |
1767 | dma_addr_t handle, size_t size, enum dma_data_direction dir) | |
1768 | { | |
1769 | struct dma_iommu_mapping *mapping = dev->archdata.mapping; | |
1770 | dma_addr_t iova = handle & PAGE_MASK; | |
1771 | struct page *page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova)); | |
1772 | unsigned int offset = handle & ~PAGE_MASK; | |
1773 | ||
1774 | if (!iova) | |
1775 | return; | |
1776 | ||
0fa478df | 1777 | __dma_page_dev_to_cpu(page, offset, size, dir); |
4ce63fcd MS |
1778 | } |
1779 | ||
1780 | static void arm_iommu_sync_single_for_device(struct device *dev, | |
1781 | dma_addr_t handle, size_t size, enum dma_data_direction dir) | |
1782 | { | |
1783 | struct dma_iommu_mapping *mapping = dev->archdata.mapping; | |
1784 | dma_addr_t iova = handle & PAGE_MASK; | |
1785 | struct page *page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova)); | |
1786 | unsigned int offset = handle & ~PAGE_MASK; | |
1787 | ||
1788 | if (!iova) | |
1789 | return; | |
1790 | ||
1791 | __dma_page_cpu_to_dev(page, offset, size, dir); | |
1792 | } | |
1793 | ||
1794 | struct dma_map_ops iommu_ops = { | |
1795 | .alloc = arm_iommu_alloc_attrs, | |
1796 | .free = arm_iommu_free_attrs, | |
1797 | .mmap = arm_iommu_mmap_attrs, | |
dc2832e1 | 1798 | .get_sgtable = arm_iommu_get_sgtable, |
4ce63fcd MS |
1799 | |
1800 | .map_page = arm_iommu_map_page, | |
1801 | .unmap_page = arm_iommu_unmap_page, | |
1802 | .sync_single_for_cpu = arm_iommu_sync_single_for_cpu, | |
1803 | .sync_single_for_device = arm_iommu_sync_single_for_device, | |
1804 | ||
1805 | .map_sg = arm_iommu_map_sg, | |
1806 | .unmap_sg = arm_iommu_unmap_sg, | |
1807 | .sync_sg_for_cpu = arm_iommu_sync_sg_for_cpu, | |
1808 | .sync_sg_for_device = arm_iommu_sync_sg_for_device, | |
d09e1333 HD |
1809 | |
1810 | .set_dma_mask = arm_dma_set_mask, | |
4ce63fcd MS |
1811 | }; |
1812 | ||
0fa478df RH |
1813 | struct dma_map_ops iommu_coherent_ops = { |
1814 | .alloc = arm_iommu_alloc_attrs, | |
1815 | .free = arm_iommu_free_attrs, | |
1816 | .mmap = arm_iommu_mmap_attrs, | |
1817 | .get_sgtable = arm_iommu_get_sgtable, | |
1818 | ||
1819 | .map_page = arm_coherent_iommu_map_page, | |
1820 | .unmap_page = arm_coherent_iommu_unmap_page, | |
1821 | ||
1822 | .map_sg = arm_coherent_iommu_map_sg, | |
1823 | .unmap_sg = arm_coherent_iommu_unmap_sg, | |
d09e1333 HD |
1824 | |
1825 | .set_dma_mask = arm_dma_set_mask, | |
0fa478df RH |
1826 | }; |
1827 | ||
4ce63fcd MS |
1828 | /** |
1829 | * arm_iommu_create_mapping | |
1830 | * @bus: pointer to the bus holding the client device (for IOMMU calls) | |
1831 | * @base: start address of the valid IO address space | |
1832 | * @size: size of the valid IO address space | |
1833 | * @order: accuracy of the IO addresses allocations | |
1834 | * | |
1835 | * Creates a mapping structure which holds information about used/unused | |
1836 | * IO address ranges, which is required to perform memory allocation and | |
1837 | * mapping with IOMMU aware functions. | |
1838 | * | |
1839 | * The client device need to be attached to the mapping with | |
1840 | * arm_iommu_attach_device function. | |
1841 | */ | |
1842 | struct dma_iommu_mapping * | |
1843 | arm_iommu_create_mapping(struct bus_type *bus, dma_addr_t base, size_t size, | |
1844 | int order) | |
1845 | { | |
1846 | unsigned int count = size >> (PAGE_SHIFT + order); | |
1847 | unsigned int bitmap_size = BITS_TO_LONGS(count) * sizeof(long); | |
1848 | struct dma_iommu_mapping *mapping; | |
1849 | int err = -ENOMEM; | |
1850 | ||
1851 | if (!count) | |
1852 | return ERR_PTR(-EINVAL); | |
1853 | ||
1854 | mapping = kzalloc(sizeof(struct dma_iommu_mapping), GFP_KERNEL); | |
1855 | if (!mapping) | |
1856 | goto err; | |
1857 | ||
1858 | mapping->bitmap = kzalloc(bitmap_size, GFP_KERNEL); | |
1859 | if (!mapping->bitmap) | |
1860 | goto err2; | |
1861 | ||
1862 | mapping->base = base; | |
1863 | mapping->bits = BITS_PER_BYTE * bitmap_size; | |
1864 | mapping->order = order; | |
1865 | spin_lock_init(&mapping->lock); | |
1866 | ||
1867 | mapping->domain = iommu_domain_alloc(bus); | |
1868 | if (!mapping->domain) | |
1869 | goto err3; | |
1870 | ||
1871 | kref_init(&mapping->kref); | |
1872 | return mapping; | |
1873 | err3: | |
1874 | kfree(mapping->bitmap); | |
1875 | err2: | |
1876 | kfree(mapping); | |
1877 | err: | |
1878 | return ERR_PTR(err); | |
1879 | } | |
18177d12 | 1880 | EXPORT_SYMBOL_GPL(arm_iommu_create_mapping); |
4ce63fcd MS |
1881 | |
1882 | static void release_iommu_mapping(struct kref *kref) | |
1883 | { | |
1884 | struct dma_iommu_mapping *mapping = | |
1885 | container_of(kref, struct dma_iommu_mapping, kref); | |
1886 | ||
1887 | iommu_domain_free(mapping->domain); | |
1888 | kfree(mapping->bitmap); | |
1889 | kfree(mapping); | |
1890 | } | |
1891 | ||
1892 | void arm_iommu_release_mapping(struct dma_iommu_mapping *mapping) | |
1893 | { | |
1894 | if (mapping) | |
1895 | kref_put(&mapping->kref, release_iommu_mapping); | |
1896 | } | |
18177d12 | 1897 | EXPORT_SYMBOL_GPL(arm_iommu_release_mapping); |
4ce63fcd MS |
1898 | |
1899 | /** | |
1900 | * arm_iommu_attach_device | |
1901 | * @dev: valid struct device pointer | |
1902 | * @mapping: io address space mapping structure (returned from | |
1903 | * arm_iommu_create_mapping) | |
1904 | * | |
1905 | * Attaches specified io address space mapping to the provided device, | |
1906 | * this replaces the dma operations (dma_map_ops pointer) with the | |
1907 | * IOMMU aware version. More than one client might be attached to | |
1908 | * the same io address space mapping. | |
1909 | */ | |
1910 | int arm_iommu_attach_device(struct device *dev, | |
1911 | struct dma_iommu_mapping *mapping) | |
1912 | { | |
1913 | int err; | |
1914 | ||
1915 | err = iommu_attach_device(mapping->domain, dev); | |
1916 | if (err) | |
1917 | return err; | |
1918 | ||
1919 | kref_get(&mapping->kref); | |
1920 | dev->archdata.mapping = mapping; | |
1921 | set_dma_ops(dev, &iommu_ops); | |
1922 | ||
75c59716 | 1923 | pr_debug("Attached IOMMU controller to %s device.\n", dev_name(dev)); |
4ce63fcd MS |
1924 | return 0; |
1925 | } | |
18177d12 | 1926 | EXPORT_SYMBOL_GPL(arm_iommu_attach_device); |
4ce63fcd | 1927 | |
6fe36758 HD |
1928 | /** |
1929 | * arm_iommu_detach_device | |
1930 | * @dev: valid struct device pointer | |
1931 | * | |
1932 | * Detaches the provided device from a previously attached map. | |
1933 | * This voids the dma operations (dma_map_ops pointer) | |
1934 | */ | |
1935 | void arm_iommu_detach_device(struct device *dev) | |
1936 | { | |
1937 | struct dma_iommu_mapping *mapping; | |
1938 | ||
1939 | mapping = to_dma_iommu_mapping(dev); | |
1940 | if (!mapping) { | |
1941 | dev_warn(dev, "Not attached\n"); | |
1942 | return; | |
1943 | } | |
1944 | ||
1945 | iommu_detach_device(mapping->domain, dev); | |
1946 | kref_put(&mapping->kref, release_iommu_mapping); | |
9e4b259d | 1947 | dev->archdata.mapping = NULL; |
6fe36758 HD |
1948 | set_dma_ops(dev, NULL); |
1949 | ||
1950 | pr_debug("Detached IOMMU controller from %s device.\n", dev_name(dev)); | |
1951 | } | |
18177d12 | 1952 | EXPORT_SYMBOL_GPL(arm_iommu_detach_device); |
6fe36758 | 1953 | |
4ce63fcd | 1954 | #endif |