diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2012-12-16 20:39:14 -0500 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2012-12-16 20:39:14 -0500 |
commit | 9b690c3d56ce15dd265b6398f9d8d58c29c17032 (patch) | |
tree | 56477d1f4e596011f17d1c64e8597613330e5439 /lib | |
parent | 36cd5c19c3fe8291fac45a262c44c00bd14b531a (diff) | |
parent | af51a9f1848ff50079a10def56a2c064f326af22 (diff) |
Merge tag 'stable/for-linus-3.8-rc0-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/swiotlb
Pull swiotlb update from Konrad Rzeszutek Wilk:
"Feature:
- Use dma addresses instead of the virt_to_phys and vice versa
functions.
Remove the multitude of phys_to_virt/virt_to_phys calls and instead
operate on the physical addresses instead of virtual in many of the
internal functions. This does provide a speed up in interrupt
handlers that do DMA operations and use SWIOTLB."
* tag 'stable/for-linus-3.8-rc0-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/swiotlb:
swiotlb: Do not export swiotlb_bounce since there are no external consumers
swiotlb: Use physical addresses instead of virtual in swiotlb_tbl_sync_single
swiotlb: Use physical addresses for swiotlb_tbl_unmap_single
swiotlb: Return physical addresses when calling swiotlb_tbl_map_single
swiotlb: Make io_tlb_overflow_buffer a physical address
swiotlb: Make io_tlb_start a physical address instead of a virtual one
swiotlb: Make io_tlb_end a physical address instead of a virtual one
Diffstat (limited to 'lib')
-rw-r--r-- | lib/swiotlb.c | 269 |
1 files changed, 139 insertions, 130 deletions
diff --git a/lib/swiotlb.c b/lib/swiotlb.c index f114bf6a8e13..196b06984dec 100644 --- a/lib/swiotlb.c +++ b/lib/swiotlb.c | |||
@@ -57,7 +57,7 @@ int swiotlb_force; | |||
57 | * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this | 57 | * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this |
58 | * API. | 58 | * API. |
59 | */ | 59 | */ |
60 | static char *io_tlb_start, *io_tlb_end; | 60 | static phys_addr_t io_tlb_start, io_tlb_end; |
61 | 61 | ||
62 | /* | 62 | /* |
63 | * The number of IO TLB blocks (in groups of 64) between io_tlb_start and | 63 | * The number of IO TLB blocks (in groups of 64) between io_tlb_start and |
@@ -70,7 +70,7 @@ static unsigned long io_tlb_nslabs; | |||
70 | */ | 70 | */ |
71 | static unsigned long io_tlb_overflow = 32*1024; | 71 | static unsigned long io_tlb_overflow = 32*1024; |
72 | 72 | ||
73 | static void *io_tlb_overflow_buffer; | 73 | static phys_addr_t io_tlb_overflow_buffer; |
74 | 74 | ||
75 | /* | 75 | /* |
76 | * This is a free list describing the number of free entries available from | 76 | * This is a free list describing the number of free entries available from |
@@ -125,27 +125,38 @@ static dma_addr_t swiotlb_virt_to_bus(struct device *hwdev, | |||
125 | void swiotlb_print_info(void) | 125 | void swiotlb_print_info(void) |
126 | { | 126 | { |
127 | unsigned long bytes = io_tlb_nslabs << IO_TLB_SHIFT; | 127 | unsigned long bytes = io_tlb_nslabs << IO_TLB_SHIFT; |
128 | phys_addr_t pstart, pend; | 128 | unsigned char *vstart, *vend; |
129 | 129 | ||
130 | pstart = virt_to_phys(io_tlb_start); | 130 | vstart = phys_to_virt(io_tlb_start); |
131 | pend = virt_to_phys(io_tlb_end); | 131 | vend = phys_to_virt(io_tlb_end); |
132 | 132 | ||
133 | printk(KERN_INFO "software IO TLB [mem %#010llx-%#010llx] (%luMB) mapped at [%p-%p]\n", | 133 | printk(KERN_INFO "software IO TLB [mem %#010llx-%#010llx] (%luMB) mapped at [%p-%p]\n", |
134 | (unsigned long long)pstart, (unsigned long long)pend - 1, | 134 | (unsigned long long)io_tlb_start, |
135 | bytes >> 20, io_tlb_start, io_tlb_end - 1); | 135 | (unsigned long long)io_tlb_end, |
136 | bytes >> 20, vstart, vend - 1); | ||
136 | } | 137 | } |
137 | 138 | ||
138 | void __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose) | 139 | void __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose) |
139 | { | 140 | { |
141 | void *v_overflow_buffer; | ||
140 | unsigned long i, bytes; | 142 | unsigned long i, bytes; |
141 | 143 | ||
142 | bytes = nslabs << IO_TLB_SHIFT; | 144 | bytes = nslabs << IO_TLB_SHIFT; |
143 | 145 | ||
144 | io_tlb_nslabs = nslabs; | 146 | io_tlb_nslabs = nslabs; |
145 | io_tlb_start = tlb; | 147 | io_tlb_start = __pa(tlb); |
146 | io_tlb_end = io_tlb_start + bytes; | 148 | io_tlb_end = io_tlb_start + bytes; |
147 | 149 | ||
148 | /* | 150 | /* |
151 | * Get the overflow emergency buffer | ||
152 | */ | ||
153 | v_overflow_buffer = alloc_bootmem_low_pages(PAGE_ALIGN(io_tlb_overflow)); | ||
154 | if (!v_overflow_buffer) | ||
155 | panic("Cannot allocate SWIOTLB overflow buffer!\n"); | ||
156 | |||
157 | io_tlb_overflow_buffer = __pa(v_overflow_buffer); | ||
158 | |||
159 | /* | ||
149 | * Allocate and initialize the free list array. This array is used | 160 | * Allocate and initialize the free list array. This array is used |
150 | * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE | 161 | * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE |
151 | * between io_tlb_start and io_tlb_end. | 162 | * between io_tlb_start and io_tlb_end. |
@@ -156,12 +167,6 @@ void __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose) | |||
156 | io_tlb_index = 0; | 167 | io_tlb_index = 0; |
157 | io_tlb_orig_addr = alloc_bootmem_pages(PAGE_ALIGN(io_tlb_nslabs * sizeof(phys_addr_t))); | 168 | io_tlb_orig_addr = alloc_bootmem_pages(PAGE_ALIGN(io_tlb_nslabs * sizeof(phys_addr_t))); |
158 | 169 | ||
159 | /* | ||
160 | * Get the overflow emergency buffer | ||
161 | */ | ||
162 | io_tlb_overflow_buffer = alloc_bootmem_low_pages(PAGE_ALIGN(io_tlb_overflow)); | ||
163 | if (!io_tlb_overflow_buffer) | ||
164 | panic("Cannot allocate SWIOTLB overflow buffer!\n"); | ||
165 | if (verbose) | 170 | if (verbose) |
166 | swiotlb_print_info(); | 171 | swiotlb_print_info(); |
167 | } | 172 | } |
@@ -173,6 +178,7 @@ void __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose) | |||
173 | static void __init | 178 | static void __init |
174 | swiotlb_init_with_default_size(size_t default_size, int verbose) | 179 | swiotlb_init_with_default_size(size_t default_size, int verbose) |
175 | { | 180 | { |
181 | unsigned char *vstart; | ||
176 | unsigned long bytes; | 182 | unsigned long bytes; |
177 | 183 | ||
178 | if (!io_tlb_nslabs) { | 184 | if (!io_tlb_nslabs) { |
@@ -185,11 +191,11 @@ swiotlb_init_with_default_size(size_t default_size, int verbose) | |||
185 | /* | 191 | /* |
186 | * Get IO TLB memory from the low pages | 192 | * Get IO TLB memory from the low pages |
187 | */ | 193 | */ |
188 | io_tlb_start = alloc_bootmem_low_pages(PAGE_ALIGN(bytes)); | 194 | vstart = alloc_bootmem_low_pages(PAGE_ALIGN(bytes)); |
189 | if (!io_tlb_start) | 195 | if (!vstart) |
190 | panic("Cannot allocate SWIOTLB buffer"); | 196 | panic("Cannot allocate SWIOTLB buffer"); |
191 | 197 | ||
192 | swiotlb_init_with_tbl(io_tlb_start, io_tlb_nslabs, verbose); | 198 | swiotlb_init_with_tbl(vstart, io_tlb_nslabs, verbose); |
193 | } | 199 | } |
194 | 200 | ||
195 | void __init | 201 | void __init |
@@ -207,6 +213,7 @@ int | |||
207 | swiotlb_late_init_with_default_size(size_t default_size) | 213 | swiotlb_late_init_with_default_size(size_t default_size) |
208 | { | 214 | { |
209 | unsigned long bytes, req_nslabs = io_tlb_nslabs; | 215 | unsigned long bytes, req_nslabs = io_tlb_nslabs; |
216 | unsigned char *vstart = NULL; | ||
210 | unsigned int order; | 217 | unsigned int order; |
211 | int rc = 0; | 218 | int rc = 0; |
212 | 219 | ||
@@ -223,14 +230,14 @@ swiotlb_late_init_with_default_size(size_t default_size) | |||
223 | bytes = io_tlb_nslabs << IO_TLB_SHIFT; | 230 | bytes = io_tlb_nslabs << IO_TLB_SHIFT; |
224 | 231 | ||
225 | while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { | 232 | while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { |
226 | io_tlb_start = (void *)__get_free_pages(GFP_DMA | __GFP_NOWARN, | 233 | vstart = (void *)__get_free_pages(GFP_DMA | __GFP_NOWARN, |
227 | order); | 234 | order); |
228 | if (io_tlb_start) | 235 | if (vstart) |
229 | break; | 236 | break; |
230 | order--; | 237 | order--; |
231 | } | 238 | } |
232 | 239 | ||
233 | if (!io_tlb_start) { | 240 | if (!vstart) { |
234 | io_tlb_nslabs = req_nslabs; | 241 | io_tlb_nslabs = req_nslabs; |
235 | return -ENOMEM; | 242 | return -ENOMEM; |
236 | } | 243 | } |
@@ -239,9 +246,9 @@ swiotlb_late_init_with_default_size(size_t default_size) | |||
239 | "for software IO TLB\n", (PAGE_SIZE << order) >> 20); | 246 | "for software IO TLB\n", (PAGE_SIZE << order) >> 20); |
240 | io_tlb_nslabs = SLABS_PER_PAGE << order; | 247 | io_tlb_nslabs = SLABS_PER_PAGE << order; |
241 | } | 248 | } |
242 | rc = swiotlb_late_init_with_tbl(io_tlb_start, io_tlb_nslabs); | 249 | rc = swiotlb_late_init_with_tbl(vstart, io_tlb_nslabs); |
243 | if (rc) | 250 | if (rc) |
244 | free_pages((unsigned long)io_tlb_start, order); | 251 | free_pages((unsigned long)vstart, order); |
245 | return rc; | 252 | return rc; |
246 | } | 253 | } |
247 | 254 | ||
@@ -249,14 +256,25 @@ int | |||
249 | swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs) | 256 | swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs) |
250 | { | 257 | { |
251 | unsigned long i, bytes; | 258 | unsigned long i, bytes; |
259 | unsigned char *v_overflow_buffer; | ||
252 | 260 | ||
253 | bytes = nslabs << IO_TLB_SHIFT; | 261 | bytes = nslabs << IO_TLB_SHIFT; |
254 | 262 | ||
255 | io_tlb_nslabs = nslabs; | 263 | io_tlb_nslabs = nslabs; |
256 | io_tlb_start = tlb; | 264 | io_tlb_start = virt_to_phys(tlb); |
257 | io_tlb_end = io_tlb_start + bytes; | 265 | io_tlb_end = io_tlb_start + bytes; |
258 | 266 | ||
259 | memset(io_tlb_start, 0, bytes); | 267 | memset(tlb, 0, bytes); |
268 | |||
269 | /* | ||
270 | * Get the overflow emergency buffer | ||
271 | */ | ||
272 | v_overflow_buffer = (void *)__get_free_pages(GFP_DMA, | ||
273 | get_order(io_tlb_overflow)); | ||
274 | if (!v_overflow_buffer) | ||
275 | goto cleanup2; | ||
276 | |||
277 | io_tlb_overflow_buffer = virt_to_phys(v_overflow_buffer); | ||
260 | 278 | ||
261 | /* | 279 | /* |
262 | * Allocate and initialize the free list array. This array is used | 280 | * Allocate and initialize the free list array. This array is used |
@@ -266,7 +284,7 @@ swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs) | |||
266 | io_tlb_list = (unsigned int *)__get_free_pages(GFP_KERNEL, | 284 | io_tlb_list = (unsigned int *)__get_free_pages(GFP_KERNEL, |
267 | get_order(io_tlb_nslabs * sizeof(int))); | 285 | get_order(io_tlb_nslabs * sizeof(int))); |
268 | if (!io_tlb_list) | 286 | if (!io_tlb_list) |
269 | goto cleanup2; | 287 | goto cleanup3; |
270 | 288 | ||
271 | for (i = 0; i < io_tlb_nslabs; i++) | 289 | for (i = 0; i < io_tlb_nslabs; i++) |
272 | io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE); | 290 | io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE); |
@@ -277,18 +295,10 @@ swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs) | |||
277 | get_order(io_tlb_nslabs * | 295 | get_order(io_tlb_nslabs * |
278 | sizeof(phys_addr_t))); | 296 | sizeof(phys_addr_t))); |
279 | if (!io_tlb_orig_addr) | 297 | if (!io_tlb_orig_addr) |
280 | goto cleanup3; | 298 | goto cleanup4; |
281 | 299 | ||
282 | memset(io_tlb_orig_addr, 0, io_tlb_nslabs * sizeof(phys_addr_t)); | 300 | memset(io_tlb_orig_addr, 0, io_tlb_nslabs * sizeof(phys_addr_t)); |
283 | 301 | ||
284 | /* | ||
285 | * Get the overflow emergency buffer | ||
286 | */ | ||
287 | io_tlb_overflow_buffer = (void *)__get_free_pages(GFP_DMA, | ||
288 | get_order(io_tlb_overflow)); | ||
289 | if (!io_tlb_overflow_buffer) | ||
290 | goto cleanup4; | ||
291 | |||
292 | swiotlb_print_info(); | 302 | swiotlb_print_info(); |
293 | 303 | ||
294 | late_alloc = 1; | 304 | late_alloc = 1; |
@@ -296,42 +306,42 @@ swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs) | |||
296 | return 0; | 306 | return 0; |
297 | 307 | ||
298 | cleanup4: | 308 | cleanup4: |
299 | free_pages((unsigned long)io_tlb_orig_addr, | ||
300 | get_order(io_tlb_nslabs * sizeof(phys_addr_t))); | ||
301 | io_tlb_orig_addr = NULL; | ||
302 | cleanup3: | ||
303 | free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs * | 309 | free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs * |
304 | sizeof(int))); | 310 | sizeof(int))); |
305 | io_tlb_list = NULL; | 311 | io_tlb_list = NULL; |
312 | cleanup3: | ||
313 | free_pages((unsigned long)v_overflow_buffer, | ||
314 | get_order(io_tlb_overflow)); | ||
315 | io_tlb_overflow_buffer = 0; | ||
306 | cleanup2: | 316 | cleanup2: |
307 | io_tlb_end = NULL; | 317 | io_tlb_end = 0; |
308 | io_tlb_start = NULL; | 318 | io_tlb_start = 0; |
309 | io_tlb_nslabs = 0; | 319 | io_tlb_nslabs = 0; |
310 | return -ENOMEM; | 320 | return -ENOMEM; |
311 | } | 321 | } |
312 | 322 | ||
313 | void __init swiotlb_free(void) | 323 | void __init swiotlb_free(void) |
314 | { | 324 | { |
315 | if (!io_tlb_overflow_buffer) | 325 | if (!io_tlb_orig_addr) |
316 | return; | 326 | return; |
317 | 327 | ||
318 | if (late_alloc) { | 328 | if (late_alloc) { |
319 | free_pages((unsigned long)io_tlb_overflow_buffer, | 329 | free_pages((unsigned long)phys_to_virt(io_tlb_overflow_buffer), |
320 | get_order(io_tlb_overflow)); | 330 | get_order(io_tlb_overflow)); |
321 | free_pages((unsigned long)io_tlb_orig_addr, | 331 | free_pages((unsigned long)io_tlb_orig_addr, |
322 | get_order(io_tlb_nslabs * sizeof(phys_addr_t))); | 332 | get_order(io_tlb_nslabs * sizeof(phys_addr_t))); |
323 | free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs * | 333 | free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs * |
324 | sizeof(int))); | 334 | sizeof(int))); |
325 | free_pages((unsigned long)io_tlb_start, | 335 | free_pages((unsigned long)phys_to_virt(io_tlb_start), |
326 | get_order(io_tlb_nslabs << IO_TLB_SHIFT)); | 336 | get_order(io_tlb_nslabs << IO_TLB_SHIFT)); |
327 | } else { | 337 | } else { |
328 | free_bootmem_late(__pa(io_tlb_overflow_buffer), | 338 | free_bootmem_late(io_tlb_overflow_buffer, |
329 | PAGE_ALIGN(io_tlb_overflow)); | 339 | PAGE_ALIGN(io_tlb_overflow)); |
330 | free_bootmem_late(__pa(io_tlb_orig_addr), | 340 | free_bootmem_late(__pa(io_tlb_orig_addr), |
331 | PAGE_ALIGN(io_tlb_nslabs * sizeof(phys_addr_t))); | 341 | PAGE_ALIGN(io_tlb_nslabs * sizeof(phys_addr_t))); |
332 | free_bootmem_late(__pa(io_tlb_list), | 342 | free_bootmem_late(__pa(io_tlb_list), |
333 | PAGE_ALIGN(io_tlb_nslabs * sizeof(int))); | 343 | PAGE_ALIGN(io_tlb_nslabs * sizeof(int))); |
334 | free_bootmem_late(__pa(io_tlb_start), | 344 | free_bootmem_late(io_tlb_start, |
335 | PAGE_ALIGN(io_tlb_nslabs << IO_TLB_SHIFT)); | 345 | PAGE_ALIGN(io_tlb_nslabs << IO_TLB_SHIFT)); |
336 | } | 346 | } |
337 | io_tlb_nslabs = 0; | 347 | io_tlb_nslabs = 0; |
@@ -339,21 +349,21 @@ void __init swiotlb_free(void) | |||
339 | 349 | ||
340 | static int is_swiotlb_buffer(phys_addr_t paddr) | 350 | static int is_swiotlb_buffer(phys_addr_t paddr) |
341 | { | 351 | { |
342 | return paddr >= virt_to_phys(io_tlb_start) && | 352 | return paddr >= io_tlb_start && paddr < io_tlb_end; |
343 | paddr < virt_to_phys(io_tlb_end); | ||
344 | } | 353 | } |
345 | 354 | ||
346 | /* | 355 | /* |
347 | * Bounce: copy the swiotlb buffer back to the original dma location | 356 | * Bounce: copy the swiotlb buffer back to the original dma location |
348 | */ | 357 | */ |
349 | void swiotlb_bounce(phys_addr_t phys, char *dma_addr, size_t size, | 358 | static void swiotlb_bounce(phys_addr_t orig_addr, phys_addr_t tlb_addr, |
350 | enum dma_data_direction dir) | 359 | size_t size, enum dma_data_direction dir) |
351 | { | 360 | { |
352 | unsigned long pfn = PFN_DOWN(phys); | 361 | unsigned long pfn = PFN_DOWN(orig_addr); |
362 | unsigned char *vaddr = phys_to_virt(tlb_addr); | ||
353 | 363 | ||
354 | if (PageHighMem(pfn_to_page(pfn))) { | 364 | if (PageHighMem(pfn_to_page(pfn))) { |
355 | /* The buffer does not have a mapping. Map it in and copy */ | 365 | /* The buffer does not have a mapping. Map it in and copy */ |
356 | unsigned int offset = phys & ~PAGE_MASK; | 366 | unsigned int offset = orig_addr & ~PAGE_MASK; |
357 | char *buffer; | 367 | char *buffer; |
358 | unsigned int sz = 0; | 368 | unsigned int sz = 0; |
359 | unsigned long flags; | 369 | unsigned long flags; |
@@ -364,32 +374,31 @@ void swiotlb_bounce(phys_addr_t phys, char *dma_addr, size_t size, | |||
364 | local_irq_save(flags); | 374 | local_irq_save(flags); |
365 | buffer = kmap_atomic(pfn_to_page(pfn)); | 375 | buffer = kmap_atomic(pfn_to_page(pfn)); |
366 | if (dir == DMA_TO_DEVICE) | 376 | if (dir == DMA_TO_DEVICE) |
367 | memcpy(dma_addr, buffer + offset, sz); | 377 | memcpy(vaddr, buffer + offset, sz); |
368 | else | 378 | else |
369 | memcpy(buffer + offset, dma_addr, sz); | 379 | memcpy(buffer + offset, vaddr, sz); |
370 | kunmap_atomic(buffer); | 380 | kunmap_atomic(buffer); |
371 | local_irq_restore(flags); | 381 | local_irq_restore(flags); |
372 | 382 | ||
373 | size -= sz; | 383 | size -= sz; |
374 | pfn++; | 384 | pfn++; |
375 | dma_addr += sz; | 385 | vaddr += sz; |
376 | offset = 0; | 386 | offset = 0; |
377 | } | 387 | } |
388 | } else if (dir == DMA_TO_DEVICE) { | ||
389 | memcpy(vaddr, phys_to_virt(orig_addr), size); | ||
378 | } else { | 390 | } else { |
379 | if (dir == DMA_TO_DEVICE) | 391 | memcpy(phys_to_virt(orig_addr), vaddr, size); |
380 | memcpy(dma_addr, phys_to_virt(phys), size); | ||
381 | else | ||
382 | memcpy(phys_to_virt(phys), dma_addr, size); | ||
383 | } | 392 | } |
384 | } | 393 | } |
385 | EXPORT_SYMBOL_GPL(swiotlb_bounce); | ||
386 | 394 | ||
387 | void *swiotlb_tbl_map_single(struct device *hwdev, dma_addr_t tbl_dma_addr, | 395 | phys_addr_t swiotlb_tbl_map_single(struct device *hwdev, |
388 | phys_addr_t phys, size_t size, | 396 | dma_addr_t tbl_dma_addr, |
389 | enum dma_data_direction dir) | 397 | phys_addr_t orig_addr, size_t size, |
398 | enum dma_data_direction dir) | ||
390 | { | 399 | { |
391 | unsigned long flags; | 400 | unsigned long flags; |
392 | char *dma_addr; | 401 | phys_addr_t tlb_addr; |
393 | unsigned int nslots, stride, index, wrap; | 402 | unsigned int nslots, stride, index, wrap; |
394 | int i; | 403 | int i; |
395 | unsigned long mask; | 404 | unsigned long mask; |
@@ -453,7 +462,7 @@ void *swiotlb_tbl_map_single(struct device *hwdev, dma_addr_t tbl_dma_addr, | |||
453 | io_tlb_list[i] = 0; | 462 | io_tlb_list[i] = 0; |
454 | for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE - 1) && io_tlb_list[i]; i--) | 463 | for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE - 1) && io_tlb_list[i]; i--) |
455 | io_tlb_list[i] = ++count; | 464 | io_tlb_list[i] = ++count; |
456 | dma_addr = io_tlb_start + (index << IO_TLB_SHIFT); | 465 | tlb_addr = io_tlb_start + (index << IO_TLB_SHIFT); |
457 | 466 | ||
458 | /* | 467 | /* |
459 | * Update the indices to avoid searching in the next | 468 | * Update the indices to avoid searching in the next |
@@ -471,7 +480,7 @@ void *swiotlb_tbl_map_single(struct device *hwdev, dma_addr_t tbl_dma_addr, | |||
471 | 480 | ||
472 | not_found: | 481 | not_found: |
473 | spin_unlock_irqrestore(&io_tlb_lock, flags); | 482 | spin_unlock_irqrestore(&io_tlb_lock, flags); |
474 | return NULL; | 483 | return SWIOTLB_MAP_ERROR; |
475 | found: | 484 | found: |
476 | spin_unlock_irqrestore(&io_tlb_lock, flags); | 485 | spin_unlock_irqrestore(&io_tlb_lock, flags); |
477 | 486 | ||
@@ -481,11 +490,11 @@ found: | |||
481 | * needed. | 490 | * needed. |
482 | */ | 491 | */ |
483 | for (i = 0; i < nslots; i++) | 492 | for (i = 0; i < nslots; i++) |
484 | io_tlb_orig_addr[index+i] = phys + (i << IO_TLB_SHIFT); | 493 | io_tlb_orig_addr[index+i] = orig_addr + (i << IO_TLB_SHIFT); |
485 | if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) | 494 | if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) |
486 | swiotlb_bounce(phys, dma_addr, size, DMA_TO_DEVICE); | 495 | swiotlb_bounce(orig_addr, tlb_addr, size, DMA_TO_DEVICE); |
487 | 496 | ||
488 | return dma_addr; | 497 | return tlb_addr; |
489 | } | 498 | } |
490 | EXPORT_SYMBOL_GPL(swiotlb_tbl_map_single); | 499 | EXPORT_SYMBOL_GPL(swiotlb_tbl_map_single); |
491 | 500 | ||
@@ -493,11 +502,10 @@ EXPORT_SYMBOL_GPL(swiotlb_tbl_map_single); | |||
493 | * Allocates bounce buffer and returns its kernel virtual address. | 502 | * Allocates bounce buffer and returns its kernel virtual address. |
494 | */ | 503 | */ |
495 | 504 | ||
496 | static void * | 505 | phys_addr_t map_single(struct device *hwdev, phys_addr_t phys, size_t size, |
497 | map_single(struct device *hwdev, phys_addr_t phys, size_t size, | 506 | enum dma_data_direction dir) |
498 | enum dma_data_direction dir) | ||
499 | { | 507 | { |
500 | dma_addr_t start_dma_addr = swiotlb_virt_to_bus(hwdev, io_tlb_start); | 508 | dma_addr_t start_dma_addr = phys_to_dma(hwdev, io_tlb_start); |
501 | 509 | ||
502 | return swiotlb_tbl_map_single(hwdev, start_dma_addr, phys, size, dir); | 510 | return swiotlb_tbl_map_single(hwdev, start_dma_addr, phys, size, dir); |
503 | } | 511 | } |
@@ -505,20 +513,19 @@ map_single(struct device *hwdev, phys_addr_t phys, size_t size, | |||
505 | /* | 513 | /* |
506 | * dma_addr is the kernel virtual address of the bounce buffer to unmap. | 514 | * dma_addr is the kernel virtual address of the bounce buffer to unmap. |
507 | */ | 515 | */ |
508 | void | 516 | void swiotlb_tbl_unmap_single(struct device *hwdev, phys_addr_t tlb_addr, |
509 | swiotlb_tbl_unmap_single(struct device *hwdev, char *dma_addr, size_t size, | 517 | size_t size, enum dma_data_direction dir) |
510 | enum dma_data_direction dir) | ||
511 | { | 518 | { |
512 | unsigned long flags; | 519 | unsigned long flags; |
513 | int i, count, nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; | 520 | int i, count, nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; |
514 | int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; | 521 | int index = (tlb_addr - io_tlb_start) >> IO_TLB_SHIFT; |
515 | phys_addr_t phys = io_tlb_orig_addr[index]; | 522 | phys_addr_t orig_addr = io_tlb_orig_addr[index]; |
516 | 523 | ||
517 | /* | 524 | /* |
518 | * First, sync the memory before unmapping the entry | 525 | * First, sync the memory before unmapping the entry |
519 | */ | 526 | */ |
520 | if (phys && ((dir == DMA_FROM_DEVICE) || (dir == DMA_BIDIRECTIONAL))) | 527 | if (orig_addr && ((dir == DMA_FROM_DEVICE) || (dir == DMA_BIDIRECTIONAL))) |
521 | swiotlb_bounce(phys, dma_addr, size, DMA_FROM_DEVICE); | 528 | swiotlb_bounce(orig_addr, tlb_addr, size, DMA_FROM_DEVICE); |
522 | 529 | ||
523 | /* | 530 | /* |
524 | * Return the buffer to the free list by setting the corresponding | 531 | * Return the buffer to the free list by setting the corresponding |
@@ -547,26 +554,27 @@ swiotlb_tbl_unmap_single(struct device *hwdev, char *dma_addr, size_t size, | |||
547 | } | 554 | } |
548 | EXPORT_SYMBOL_GPL(swiotlb_tbl_unmap_single); | 555 | EXPORT_SYMBOL_GPL(swiotlb_tbl_unmap_single); |
549 | 556 | ||
550 | void | 557 | void swiotlb_tbl_sync_single(struct device *hwdev, phys_addr_t tlb_addr, |
551 | swiotlb_tbl_sync_single(struct device *hwdev, char *dma_addr, size_t size, | 558 | size_t size, enum dma_data_direction dir, |
552 | enum dma_data_direction dir, | 559 | enum dma_sync_target target) |
553 | enum dma_sync_target target) | ||
554 | { | 560 | { |
555 | int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; | 561 | int index = (tlb_addr - io_tlb_start) >> IO_TLB_SHIFT; |
556 | phys_addr_t phys = io_tlb_orig_addr[index]; | 562 | phys_addr_t orig_addr = io_tlb_orig_addr[index]; |
557 | 563 | ||
558 | phys += ((unsigned long)dma_addr & ((1 << IO_TLB_SHIFT) - 1)); | 564 | orig_addr += (unsigned long)tlb_addr & ((1 << IO_TLB_SHIFT) - 1); |
559 | 565 | ||
560 | switch (target) { | 566 | switch (target) { |
561 | case SYNC_FOR_CPU: | 567 | case SYNC_FOR_CPU: |
562 | if (likely(dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)) | 568 | if (likely(dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)) |
563 | swiotlb_bounce(phys, dma_addr, size, DMA_FROM_DEVICE); | 569 | swiotlb_bounce(orig_addr, tlb_addr, |
570 | size, DMA_FROM_DEVICE); | ||
564 | else | 571 | else |
565 | BUG_ON(dir != DMA_TO_DEVICE); | 572 | BUG_ON(dir != DMA_TO_DEVICE); |
566 | break; | 573 | break; |
567 | case SYNC_FOR_DEVICE: | 574 | case SYNC_FOR_DEVICE: |
568 | if (likely(dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)) | 575 | if (likely(dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)) |
569 | swiotlb_bounce(phys, dma_addr, size, DMA_TO_DEVICE); | 576 | swiotlb_bounce(orig_addr, tlb_addr, |
577 | size, DMA_TO_DEVICE); | ||
570 | else | 578 | else |
571 | BUG_ON(dir != DMA_FROM_DEVICE); | 579 | BUG_ON(dir != DMA_FROM_DEVICE); |
572 | break; | 580 | break; |
@@ -589,12 +597,15 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size, | |||
589 | dma_mask = hwdev->coherent_dma_mask; | 597 | dma_mask = hwdev->coherent_dma_mask; |
590 | 598 | ||
591 | ret = (void *)__get_free_pages(flags, order); | 599 | ret = (void *)__get_free_pages(flags, order); |
592 | if (ret && swiotlb_virt_to_bus(hwdev, ret) + size - 1 > dma_mask) { | 600 | if (ret) { |
593 | /* | 601 | dev_addr = swiotlb_virt_to_bus(hwdev, ret); |
594 | * The allocated memory isn't reachable by the device. | 602 | if (dev_addr + size - 1 > dma_mask) { |
595 | */ | 603 | /* |
596 | free_pages((unsigned long) ret, order); | 604 | * The allocated memory isn't reachable by the device. |
597 | ret = NULL; | 605 | */ |
606 | free_pages((unsigned long) ret, order); | ||
607 | ret = NULL; | ||
608 | } | ||
598 | } | 609 | } |
599 | if (!ret) { | 610 | if (!ret) { |
600 | /* | 611 | /* |
@@ -602,25 +613,29 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size, | |||
602 | * GFP_DMA memory; fall back on map_single(), which | 613 | * GFP_DMA memory; fall back on map_single(), which |
603 | * will grab memory from the lowest available address range. | 614 | * will grab memory from the lowest available address range. |
604 | */ | 615 | */ |
605 | ret = map_single(hwdev, 0, size, DMA_FROM_DEVICE); | 616 | phys_addr_t paddr = map_single(hwdev, 0, size, DMA_FROM_DEVICE); |
606 | if (!ret) | 617 | if (paddr == SWIOTLB_MAP_ERROR) |
607 | return NULL; | 618 | return NULL; |
608 | } | ||
609 | 619 | ||
610 | memset(ret, 0, size); | 620 | ret = phys_to_virt(paddr); |
611 | dev_addr = swiotlb_virt_to_bus(hwdev, ret); | 621 | dev_addr = phys_to_dma(hwdev, paddr); |
612 | 622 | ||
613 | /* Confirm address can be DMA'd by device */ | 623 | /* Confirm address can be DMA'd by device */ |
614 | if (dev_addr + size - 1 > dma_mask) { | 624 | if (dev_addr + size - 1 > dma_mask) { |
615 | printk("hwdev DMA mask = 0x%016Lx, dev_addr = 0x%016Lx\n", | 625 | printk("hwdev DMA mask = 0x%016Lx, dev_addr = 0x%016Lx\n", |
616 | (unsigned long long)dma_mask, | 626 | (unsigned long long)dma_mask, |
617 | (unsigned long long)dev_addr); | 627 | (unsigned long long)dev_addr); |
618 | 628 | ||
619 | /* DMA_TO_DEVICE to avoid memcpy in unmap_single */ | 629 | /* DMA_TO_DEVICE to avoid memcpy in unmap_single */ |
620 | swiotlb_tbl_unmap_single(hwdev, ret, size, DMA_TO_DEVICE); | 630 | swiotlb_tbl_unmap_single(hwdev, paddr, |
621 | return NULL; | 631 | size, DMA_TO_DEVICE); |
632 | return NULL; | ||
633 | } | ||
622 | } | 634 | } |
635 | |||
623 | *dma_handle = dev_addr; | 636 | *dma_handle = dev_addr; |
637 | memset(ret, 0, size); | ||
638 | |||
624 | return ret; | 639 | return ret; |
625 | } | 640 | } |
626 | EXPORT_SYMBOL(swiotlb_alloc_coherent); | 641 | EXPORT_SYMBOL(swiotlb_alloc_coherent); |
@@ -636,7 +651,7 @@ swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, | |||
636 | free_pages((unsigned long)vaddr, get_order(size)); | 651 | free_pages((unsigned long)vaddr, get_order(size)); |
637 | else | 652 | else |
638 | /* DMA_TO_DEVICE to avoid memcpy in swiotlb_tbl_unmap_single */ | 653 | /* DMA_TO_DEVICE to avoid memcpy in swiotlb_tbl_unmap_single */ |
639 | swiotlb_tbl_unmap_single(hwdev, vaddr, size, DMA_TO_DEVICE); | 654 | swiotlb_tbl_unmap_single(hwdev, paddr, size, DMA_TO_DEVICE); |
640 | } | 655 | } |
641 | EXPORT_SYMBOL(swiotlb_free_coherent); | 656 | EXPORT_SYMBOL(swiotlb_free_coherent); |
642 | 657 | ||
@@ -677,9 +692,8 @@ dma_addr_t swiotlb_map_page(struct device *dev, struct page *page, | |||
677 | enum dma_data_direction dir, | 692 | enum dma_data_direction dir, |
678 | struct dma_attrs *attrs) | 693 | struct dma_attrs *attrs) |
679 | { | 694 | { |
680 | phys_addr_t phys = page_to_phys(page) + offset; | 695 | phys_addr_t map, phys = page_to_phys(page) + offset; |
681 | dma_addr_t dev_addr = phys_to_dma(dev, phys); | 696 | dma_addr_t dev_addr = phys_to_dma(dev, phys); |
682 | void *map; | ||
683 | 697 | ||
684 | BUG_ON(dir == DMA_NONE); | 698 | BUG_ON(dir == DMA_NONE); |
685 | /* | 699 | /* |
@@ -690,23 +704,19 @@ dma_addr_t swiotlb_map_page(struct device *dev, struct page *page, | |||
690 | if (dma_capable(dev, dev_addr, size) && !swiotlb_force) | 704 | if (dma_capable(dev, dev_addr, size) && !swiotlb_force) |
691 | return dev_addr; | 705 | return dev_addr; |
692 | 706 | ||
693 | /* | 707 | /* Oh well, have to allocate and map a bounce buffer. */ |
694 | * Oh well, have to allocate and map a bounce buffer. | ||
695 | */ | ||
696 | map = map_single(dev, phys, size, dir); | 708 | map = map_single(dev, phys, size, dir); |
697 | if (!map) { | 709 | if (map == SWIOTLB_MAP_ERROR) { |
698 | swiotlb_full(dev, size, dir, 1); | 710 | swiotlb_full(dev, size, dir, 1); |
699 | map = io_tlb_overflow_buffer; | 711 | return phys_to_dma(dev, io_tlb_overflow_buffer); |
700 | } | 712 | } |
701 | 713 | ||
702 | dev_addr = swiotlb_virt_to_bus(dev, map); | 714 | dev_addr = phys_to_dma(dev, map); |
703 | 715 | ||
704 | /* | 716 | /* Ensure that the address returned is DMA'ble */ |
705 | * Ensure that the address returned is DMA'ble | ||
706 | */ | ||
707 | if (!dma_capable(dev, dev_addr, size)) { | 717 | if (!dma_capable(dev, dev_addr, size)) { |
708 | swiotlb_tbl_unmap_single(dev, map, size, dir); | 718 | swiotlb_tbl_unmap_single(dev, map, size, dir); |
709 | dev_addr = swiotlb_virt_to_bus(dev, io_tlb_overflow_buffer); | 719 | return phys_to_dma(dev, io_tlb_overflow_buffer); |
710 | } | 720 | } |
711 | 721 | ||
712 | return dev_addr; | 722 | return dev_addr; |
@@ -729,7 +739,7 @@ static void unmap_single(struct device *hwdev, dma_addr_t dev_addr, | |||
729 | BUG_ON(dir == DMA_NONE); | 739 | BUG_ON(dir == DMA_NONE); |
730 | 740 | ||
731 | if (is_swiotlb_buffer(paddr)) { | 741 | if (is_swiotlb_buffer(paddr)) { |
732 | swiotlb_tbl_unmap_single(hwdev, phys_to_virt(paddr), size, dir); | 742 | swiotlb_tbl_unmap_single(hwdev, paddr, size, dir); |
733 | return; | 743 | return; |
734 | } | 744 | } |
735 | 745 | ||
@@ -773,8 +783,7 @@ swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr, | |||
773 | BUG_ON(dir == DMA_NONE); | 783 | BUG_ON(dir == DMA_NONE); |
774 | 784 | ||
775 | if (is_swiotlb_buffer(paddr)) { | 785 | if (is_swiotlb_buffer(paddr)) { |
776 | swiotlb_tbl_sync_single(hwdev, phys_to_virt(paddr), size, dir, | 786 | swiotlb_tbl_sync_single(hwdev, paddr, size, dir, target); |
777 | target); | ||
778 | return; | 787 | return; |
779 | } | 788 | } |
780 | 789 | ||
@@ -831,9 +840,9 @@ swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems, | |||
831 | 840 | ||
832 | if (swiotlb_force || | 841 | if (swiotlb_force || |
833 | !dma_capable(hwdev, dev_addr, sg->length)) { | 842 | !dma_capable(hwdev, dev_addr, sg->length)) { |
834 | void *map = map_single(hwdev, sg_phys(sg), | 843 | phys_addr_t map = map_single(hwdev, sg_phys(sg), |
835 | sg->length, dir); | 844 | sg->length, dir); |
836 | if (!map) { | 845 | if (map == SWIOTLB_MAP_ERROR) { |
837 | /* Don't panic here, we expect map_sg users | 846 | /* Don't panic here, we expect map_sg users |
838 | to do proper error handling. */ | 847 | to do proper error handling. */ |
839 | swiotlb_full(hwdev, sg->length, dir, 0); | 848 | swiotlb_full(hwdev, sg->length, dir, 0); |
@@ -842,7 +851,7 @@ swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems, | |||
842 | sgl[0].dma_length = 0; | 851 | sgl[0].dma_length = 0; |
843 | return 0; | 852 | return 0; |
844 | } | 853 | } |
845 | sg->dma_address = swiotlb_virt_to_bus(hwdev, map); | 854 | sg->dma_address = phys_to_dma(hwdev, map); |
846 | } else | 855 | } else |
847 | sg->dma_address = dev_addr; | 856 | sg->dma_address = dev_addr; |
848 | sg->dma_length = sg->length; | 857 | sg->dma_length = sg->length; |
@@ -925,7 +934,7 @@ EXPORT_SYMBOL(swiotlb_sync_sg_for_device); | |||
925 | int | 934 | int |
926 | swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr) | 935 | swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr) |
927 | { | 936 | { |
928 | return (dma_addr == swiotlb_virt_to_bus(hwdev, io_tlb_overflow_buffer)); | 937 | return (dma_addr == phys_to_dma(hwdev, io_tlb_overflow_buffer)); |
929 | } | 938 | } |
930 | EXPORT_SYMBOL(swiotlb_dma_mapping_error); | 939 | EXPORT_SYMBOL(swiotlb_dma_mapping_error); |
931 | 940 | ||
@@ -938,6 +947,6 @@ EXPORT_SYMBOL(swiotlb_dma_mapping_error); | |||
938 | int | 947 | int |
939 | swiotlb_dma_supported(struct device *hwdev, u64 mask) | 948 | swiotlb_dma_supported(struct device *hwdev, u64 mask) |
940 | { | 949 | { |
941 | return swiotlb_virt_to_bus(hwdev, io_tlb_end - 1) <= mask; | 950 | return phys_to_dma(hwdev, io_tlb_end - 1) <= mask; |
942 | } | 951 | } |
943 | EXPORT_SYMBOL(swiotlb_dma_supported); | 952 | EXPORT_SYMBOL(swiotlb_dma_supported); |