Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | Dynamic DMA mapping using the generic device |
2 | ============================================ | |
3 | ||
4 | James E.J. Bottomley <James.Bottomley@HansenPartnership.com> | |
5 | ||
6 | This document describes the DMA API. For a more gentle introduction | |
f5a69f4c | 7 | of the API (and actual examples) see |
51e7364e | 8 | Documentation/DMA-API-HOWTO.txt. |
1da177e4 | 9 | |
f5a69f4c FT |
10 | This API is split into two pieces. Part I describes the API. Part II |
11 | describes the extensions to the API for supporting non-consistent | |
12 | memory machines. Unless you know that your driver absolutely has to | |
13 | support non-consistent platforms (this is usually only legacy | |
14 | platforms) you should only use the API described in part I. | |
1da177e4 | 15 | |
f5a69f4c | 16 | Part I - dma_ API |
1da177e4 LT |
17 | ------------------------------------- |
18 | ||
1da177e4 LT |
19 | To get the dma_ API, you must #include <linux/dma-mapping.h> |
20 | ||
21 | ||
22 | Part Ia - Using large dma-coherent buffers | |
23 | ------------------------------------------ | |
24 | ||
25 | void * | |
26 | dma_alloc_coherent(struct device *dev, size_t size, | |
a12e2c6c | 27 | dma_addr_t *dma_handle, gfp_t flag) |
1da177e4 LT |
28 | |
29 | Consistent memory is memory for which a write by either the device or | |
30 | the processor can immediately be read by the processor or device | |
21440d31 DB |
31 | without having to worry about caching effects. (You may however need |
32 | to make sure to flush the processor's write buffers before telling | |
33 | devices to read that memory.) | |
1da177e4 LT |
34 | |
35 | This routine allocates a region of <size> bytes of consistent memory. | |
a12e2c6c | 36 | It also returns a <dma_handle> which may be cast to an unsigned |
1da177e4 LT |
37 | integer the same width as the bus and used as the physical address |
38 | base of the region. | |
39 | ||
40 | Returns: a pointer to the allocated region (in the processor's virtual | |
41 | address space) or NULL if the allocation failed. | |
42 | ||
43 | Note: consistent memory can be expensive on some platforms, and the | |
44 | minimum allocation length may be as big as a page, so you should | |
45 | consolidate your requests for consistent memory as much as possible. | |
46 | The simplest way to do that is to use the dma_pool calls (see below). | |
47 | ||
48 | The flag parameter (dma_alloc_coherent only) allows the caller to | |
49 | specify the GFP_ flags (see kmalloc) for the allocation (the | |
a12e2c6c | 50 | implementation may choose to ignore flags that affect the location of |
f5a69f4c | 51 | the returned memory, like GFP_DMA). |
1da177e4 | 52 | |
842fa69f AM |
53 | void * |
54 | dma_zalloc_coherent(struct device *dev, size_t size, | |
55 | dma_addr_t *dma_handle, gfp_t flag) | |
56 | ||
57 | Wraps dma_alloc_coherent() and also zeroes the returned memory if the | |
58 | allocation attempt succeeded. | |
59 | ||
1da177e4 | 60 | void |
a12e2c6c | 61 | dma_free_coherent(struct device *dev, size_t size, void *cpu_addr, |
1da177e4 | 62 | dma_addr_t dma_handle) |
1da177e4 LT |
63 | |
64 | Free the region of consistent memory you previously allocated. dev, | |
65 | size and dma_handle must all be the same as those passed into the | |
66 | consistent allocate. cpu_addr must be the virtual address returned by | |
a12e2c6c | 67 | the consistent allocate. |
1da177e4 | 68 | |
aa24886e DB |
69 | Note that unlike their sibling allocation calls, these routines |
70 | may only be called with IRQs enabled. | |
71 | ||
1da177e4 LT |
72 | |
73 | Part Ib - Using small dma-coherent buffers | |
74 | ------------------------------------------ | |
75 | ||
76 | To get this part of the dma_ API, you must #include <linux/dmapool.h> | |
77 | ||
78 | Many drivers need lots of small dma-coherent memory regions for DMA | |
79 | descriptors or I/O buffers. Rather than allocating in units of a page | |
80 | or more using dma_alloc_coherent(), you can use DMA pools. These work | |
a12e2c6c | 81 | much like a struct kmem_cache, except that they use the dma-coherent allocator, |
1da177e4 | 82 | not __get_free_pages(). Also, they understand common hardware constraints |
a12e2c6c | 83 | for alignment, like queue heads needing to be aligned on N-byte boundaries. |
1da177e4 LT |
84 | |
85 | ||
86 | struct dma_pool * | |
87 | dma_pool_create(const char *name, struct device *dev, | |
88 | size_t size, size_t align, size_t alloc); | |
89 | ||
1da177e4 LT |
90 | The pool create() routines initialize a pool of dma-coherent buffers |
91 | for use with a given device. It must be called in a context which | |
92 | can sleep. | |
93 | ||
e18b890b | 94 | The "name" is for diagnostics (like a struct kmem_cache name); dev and size |
1da177e4 LT |
95 | are like what you'd pass to dma_alloc_coherent(). The device's hardware |
96 | alignment requirement for this type of data is "align" (which is expressed | |
97 | in bytes, and must be a power of two). If your device has no boundary | |
98 | crossing restrictions, pass 0 for alloc; passing 4096 says memory allocated | |
99 | from this pool must not cross 4KByte boundaries. | |
100 | ||
101 | ||
a12e2c6c | 102 | void *dma_pool_alloc(struct dma_pool *pool, gfp_t gfp_flags, |
1da177e4 LT |
103 | dma_addr_t *dma_handle); |
104 | ||
1da177e4 LT |
105 | This allocates memory from the pool; the returned memory will meet the size |
106 | and alignment requirements specified at creation time. Pass GFP_ATOMIC to | |
a12e2c6c | 107 | prevent blocking, or if it's permitted (not in_interrupt, not holding SMP locks), |
1da177e4 LT |
108 | pass GFP_KERNEL to allow blocking. Like dma_alloc_coherent(), this returns |
109 | two values: an address usable by the cpu, and the dma address usable by the | |
110 | pool's device. | |
111 | ||
112 | ||
113 | void dma_pool_free(struct dma_pool *pool, void *vaddr, | |
114 | dma_addr_t addr); | |
115 | ||
1da177e4 | 116 | This puts memory back into the pool. The pool is what was passed to |
a12e2c6c | 117 | the pool allocation routine; the cpu (vaddr) and dma addresses are what |
1da177e4 LT |
118 | were returned when that routine allocated the memory being freed. |
119 | ||
120 | ||
121 | void dma_pool_destroy(struct dma_pool *pool); | |
122 | ||
1da177e4 LT |
123 | The pool destroy() routines free the resources of the pool. They must be |
124 | called in a context which can sleep. Make sure you've freed all allocated | |
125 | memory back to the pool before you destroy it. | |
126 | ||
127 | ||
128 | Part Ic - DMA addressing limitations | |
129 | ------------------------------------ | |
130 | ||
131 | int | |
132 | dma_supported(struct device *dev, u64 mask) | |
1da177e4 LT |
133 | |
134 | Checks to see if the device can support DMA to the memory described by | |
135 | mask. | |
136 | ||
137 | Returns: 1 if it can and 0 if it can't. | |
138 | ||
139 | Notes: This routine merely tests to see if the mask is possible. It | |
140 | won't change the current mask settings. It is more intended as an | |
141 | internal API for use by the platform than an external API for use by | |
142 | driver writers. | |
143 | ||
4aa806b7 RK |
144 | int |
145 | dma_set_mask_and_coherent(struct device *dev, u64 mask) | |
146 | ||
147 | Checks to see if the mask is possible and updates the device | |
148 | streaming and coherent DMA mask parameters if it is. | |
149 | ||
150 | Returns: 0 if successful and a negative error if not. | |
151 | ||
1da177e4 LT |
152 | int |
153 | dma_set_mask(struct device *dev, u64 mask) | |
1da177e4 LT |
154 | |
155 | Checks to see if the mask is possible and updates the device | |
156 | parameters if it is. | |
157 | ||
158 | Returns: 0 if successful and a negative error if not. | |
159 | ||
6a1961f4 FT |
160 | int |
161 | dma_set_coherent_mask(struct device *dev, u64 mask) | |
6a1961f4 FT |
162 | |
163 | Checks to see if the mask is possible and updates the device | |
164 | parameters if it is. | |
165 | ||
166 | Returns: 0 if successful and a negative error if not. | |
167 | ||
1da177e4 LT |
168 | u64 |
169 | dma_get_required_mask(struct device *dev) | |
170 | ||
175add19 JK |
171 | This API returns the mask that the platform requires to |
172 | operate efficiently. Usually this means the returned mask | |
1da177e4 LT |
173 | is the minimum required to cover all of memory. Examining the |
174 | required mask gives drivers with variable descriptor sizes the | |
175 | opportunity to use smaller descriptors as necessary. | |
176 | ||
177 | Requesting the required mask does not alter the current mask. If you | |
175add19 JK |
178 | wish to take advantage of it, you should issue a dma_set_mask() |
179 | call to set the mask to the value returned. | |
1da177e4 LT |
180 | |
181 | ||
182 | Part Id - Streaming DMA mappings | |
183 | -------------------------------- | |
184 | ||
185 | dma_addr_t | |
186 | dma_map_single(struct device *dev, void *cpu_addr, size_t size, | |
187 | enum dma_data_direction direction) | |
1da177e4 LT |
188 | |
189 | Maps a piece of processor virtual memory so it can be accessed by the | |
190 | device and returns the physical handle of the memory. | |
191 | ||
192 | The direction for both api's may be converted freely by casting. | |
193 | However the dma_ API uses a strongly typed enumerator for its | |
194 | direction: | |
195 | ||
f5a69f4c FT |
196 | DMA_NONE no direction (used for debugging) |
197 | DMA_TO_DEVICE data is going from the memory to the device | |
198 | DMA_FROM_DEVICE data is coming from the device to the memory | |
199 | DMA_BIDIRECTIONAL direction isn't known | |
1da177e4 LT |
200 | |
201 | Notes: Not all memory regions in a machine can be mapped by this | |
202 | API. Further, regions that appear to be physically contiguous in | |
203 | kernel virtual space may not be contiguous as physical memory. Since | |
204 | this API does not provide any scatter/gather capability, it will fail | |
a12e2c6c | 205 | if the user tries to map a non-physically contiguous piece of memory. |
1da177e4 | 206 | For this reason, it is recommended that memory mapped by this API be |
a12e2c6c | 207 | obtained only from sources which guarantee it to be physically contiguous |
1da177e4 LT |
208 | (like kmalloc). |
209 | ||
210 | Further, the physical address of the memory must be within the | |
211 | dma_mask of the device (the dma_mask represents a bit mask of the | |
a12e2c6c | 212 | addressable region for the device. I.e., if the physical address of |
1da177e4 LT |
213 | the memory anded with the dma_mask is still equal to the physical |
214 | address, then the device can perform DMA to the memory). In order to | |
215 | ensure that the memory allocated by kmalloc is within the dma_mask, | |
a12e2c6c | 216 | the driver may specify various platform-dependent flags to restrict |
1da177e4 LT |
217 | the physical memory range of the allocation (e.g. on x86, GFP_DMA |
218 | guarantees to be within the first 16Mb of available physical memory, | |
219 | as required by ISA devices). | |
220 | ||
221 | Note also that the above constraints on physical contiguity and | |
222 | dma_mask may not apply if the platform has an IOMMU (a device which | |
223 | supplies a physical to virtual mapping between the I/O memory bus and | |
224 | the device). However, to be portable, device driver writers may *not* | |
225 | assume that such an IOMMU exists. | |
226 | ||
227 | Warnings: Memory coherency operates at a granularity called the cache | |
228 | line width. In order for memory mapped by this API to operate | |
229 | correctly, the mapped region must begin exactly on a cache line | |
230 | boundary and end exactly on one (to prevent two separately mapped | |
231 | regions from sharing a single cache line). Since the cache line size | |
232 | may not be known at compile time, the API will not enforce this | |
233 | requirement. Therefore, it is recommended that driver writers who | |
234 | don't take special care to determine the cache line size at run time | |
235 | only map virtual regions that begin and end on page boundaries (which | |
236 | are guaranteed also to be cache line boundaries). | |
237 | ||
238 | DMA_TO_DEVICE synchronisation must be done after the last modification | |
239 | of the memory region by the software and before it is handed off to | |
a12e2c6c RD |
240 | the driver. Once this primitive is used, memory covered by this |
241 | primitive should be treated as read-only by the device. If the device | |
1da177e4 LT |
242 | may write to it at any point, it should be DMA_BIDIRECTIONAL (see |
243 | below). | |
244 | ||
245 | DMA_FROM_DEVICE synchronisation must be done before the driver | |
246 | accesses data that may be changed by the device. This memory should | |
a12e2c6c | 247 | be treated as read-only by the driver. If the driver needs to write |
1da177e4 LT |
248 | to it at any point, it should be DMA_BIDIRECTIONAL (see below). |
249 | ||
250 | DMA_BIDIRECTIONAL requires special handling: it means that the driver | |
251 | isn't sure if the memory was modified before being handed off to the | |
252 | device and also isn't sure if the device will also modify it. Thus, | |
253 | you must always sync bidirectional memory twice: once before the | |
254 | memory is handed off to the device (to make sure all memory changes | |
255 | are flushed from the processor) and once before the data may be | |
256 | accessed after being used by the device (to make sure any processor | |
a12e2c6c | 257 | cache lines are updated with data that the device may have changed). |
1da177e4 LT |
258 | |
259 | void | |
260 | dma_unmap_single(struct device *dev, dma_addr_t dma_addr, size_t size, | |
261 | enum dma_data_direction direction) | |
1da177e4 LT |
262 | |
263 | Unmaps the region previously mapped. All the parameters passed in | |
264 | must be identical to those passed in (and returned) by the mapping | |
265 | API. | |
266 | ||
267 | dma_addr_t | |
268 | dma_map_page(struct device *dev, struct page *page, | |
269 | unsigned long offset, size_t size, | |
270 | enum dma_data_direction direction) | |
1da177e4 LT |
271 | void |
272 | dma_unmap_page(struct device *dev, dma_addr_t dma_address, size_t size, | |
273 | enum dma_data_direction direction) | |
1da177e4 LT |
274 | |
275 | API for mapping and unmapping for pages. All the notes and warnings | |
276 | for the other mapping APIs apply here. Also, although the <offset> | |
277 | and <size> parameters are provided to do partial page mapping, it is | |
278 | recommended that you never use these unless you really know what the | |
279 | cache width is. | |
280 | ||
281 | int | |
8d8bb39b | 282 | dma_mapping_error(struct device *dev, dma_addr_t dma_addr) |
1da177e4 | 283 | |
1da177e4 LT |
284 | In some circumstances dma_map_single and dma_map_page will fail to create |
285 | a mapping. A driver can check for these errors by testing the returned | |
a12e2c6c RD |
286 | dma address with dma_mapping_error(). A non-zero return value means the mapping |
287 | could not be created and the driver should take appropriate action (e.g. | |
1da177e4 LT |
288 | reduce current DMA mapping usage or delay and try again later). |
289 | ||
21440d31 DB |
290 | int |
291 | dma_map_sg(struct device *dev, struct scatterlist *sg, | |
292 | int nents, enum dma_data_direction direction) | |
1da177e4 | 293 | |
a12e2c6c | 294 | Returns: the number of physical segments mapped (this may be shorter |
1d678f36 FT |
295 | than <nents> passed in if some elements of the scatter/gather list are |
296 | physically or virtually adjacent and an IOMMU maps them with a single | |
297 | entry). | |
1da177e4 LT |
298 | |
299 | Please note that the sg cannot be mapped again if it has been mapped once. | |
300 | The mapping process is allowed to destroy information in the sg. | |
301 | ||
302 | As with the other mapping interfaces, dma_map_sg can fail. When it | |
303 | does, 0 is returned and a driver must take appropriate action. It is | |
304 | critical that the driver do something, in the case of a block driver | |
305 | aborting the request or even oopsing is better than doing nothing and | |
306 | corrupting the filesystem. | |
307 | ||
21440d31 DB |
308 | With scatterlists, you use the resulting mapping like this: |
309 | ||
310 | int i, count = dma_map_sg(dev, sglist, nents, direction); | |
311 | struct scatterlist *sg; | |
312 | ||
79eb0145 | 313 | for_each_sg(sglist, sg, count, i) { |
21440d31 DB |
314 | hw_address[i] = sg_dma_address(sg); |
315 | hw_len[i] = sg_dma_len(sg); | |
316 | } | |
317 | ||
318 | where nents is the number of entries in the sglist. | |
319 | ||
320 | The implementation is free to merge several consecutive sglist entries | |
321 | into one (e.g. with an IOMMU, or if several pages just happen to be | |
322 | physically contiguous) and returns the actual number of sg entries it | |
323 | mapped them to. On failure 0, is returned. | |
324 | ||
325 | Then you should loop count times (note: this can be less than nents times) | |
326 | and use sg_dma_address() and sg_dma_len() macros where you previously | |
327 | accessed sg->address and sg->length as shown above. | |
328 | ||
329 | void | |
330 | dma_unmap_sg(struct device *dev, struct scatterlist *sg, | |
331 | int nhwentries, enum dma_data_direction direction) | |
1da177e4 | 332 | |
a12e2c6c | 333 | Unmap the previously mapped scatter/gather list. All the parameters |
1da177e4 LT |
334 | must be the same as those and passed in to the scatter/gather mapping |
335 | API. | |
336 | ||
337 | Note: <nents> must be the number you passed in, *not* the number of | |
338 | physical entries returned. | |
339 | ||
9705ef7e FT |
340 | void |
341 | dma_sync_single_for_cpu(struct device *dev, dma_addr_t dma_handle, size_t size, | |
342 | enum dma_data_direction direction) | |
343 | void | |
9705ef7e FT |
344 | dma_sync_single_for_device(struct device *dev, dma_addr_t dma_handle, size_t size, |
345 | enum dma_data_direction direction) | |
346 | void | |
9705ef7e FT |
347 | dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg, int nelems, |
348 | enum dma_data_direction direction) | |
349 | void | |
9705ef7e FT |
350 | dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg, int nelems, |
351 | enum dma_data_direction direction) | |
9705ef7e FT |
352 | |
353 | Synchronise a single contiguous or scatter/gather mapping for the cpu | |
354 | and device. With the sync_sg API, all the parameters must be the same | |
355 | as those passed into the single mapping API. With the sync_single API, | |
356 | you can use dma_handle and size parameters that aren't identical to | |
357 | those passed into the single mapping API to do a partial sync. | |
358 | ||
359 | Notes: You must do this: | |
360 | ||
361 | - Before reading values that have been written by DMA from the device | |
362 | (use the DMA_FROM_DEVICE direction) | |
363 | - After writing values that will be written to the device using DMA | |
364 | (use the DMA_TO_DEVICE) direction | |
365 | - before *and* after handing memory to the device if the memory is | |
366 | DMA_BIDIRECTIONAL | |
367 | ||
368 | See also dma_map_single(). | |
369 | ||
a75b0a2f AK |
370 | dma_addr_t |
371 | dma_map_single_attrs(struct device *dev, void *cpu_addr, size_t size, | |
372 | enum dma_data_direction dir, | |
373 | struct dma_attrs *attrs) | |
374 | ||
375 | void | |
376 | dma_unmap_single_attrs(struct device *dev, dma_addr_t dma_addr, | |
377 | size_t size, enum dma_data_direction dir, | |
378 | struct dma_attrs *attrs) | |
379 | ||
380 | int | |
381 | dma_map_sg_attrs(struct device *dev, struct scatterlist *sgl, | |
382 | int nents, enum dma_data_direction dir, | |
383 | struct dma_attrs *attrs) | |
384 | ||
385 | void | |
386 | dma_unmap_sg_attrs(struct device *dev, struct scatterlist *sgl, | |
387 | int nents, enum dma_data_direction dir, | |
388 | struct dma_attrs *attrs) | |
389 | ||
390 | The four functions above are just like the counterpart functions | |
391 | without the _attrs suffixes, except that they pass an optional | |
392 | struct dma_attrs*. | |
393 | ||
394 | struct dma_attrs encapsulates a set of "dma attributes". For the | |
395 | definition of struct dma_attrs see linux/dma-attrs.h. | |
396 | ||
397 | The interpretation of dma attributes is architecture-specific, and | |
398 | each attribute should be documented in Documentation/DMA-attributes.txt. | |
399 | ||
400 | If struct dma_attrs* is NULL, the semantics of each of these | |
401 | functions is identical to those of the corresponding function | |
402 | without the _attrs suffix. As a result dma_map_single_attrs() | |
403 | can generally replace dma_map_single(), etc. | |
404 | ||
405 | As an example of the use of the *_attrs functions, here's how | |
406 | you could pass an attribute DMA_ATTR_FOO when mapping memory | |
407 | for DMA: | |
408 | ||
409 | #include <linux/dma-attrs.h> | |
410 | /* DMA_ATTR_FOO should be defined in linux/dma-attrs.h and | |
411 | * documented in Documentation/DMA-attributes.txt */ | |
412 | ... | |
413 | ||
414 | DEFINE_DMA_ATTRS(attrs); | |
415 | dma_set_attr(DMA_ATTR_FOO, &attrs); | |
416 | .... | |
417 | n = dma_map_sg_attrs(dev, sg, nents, DMA_TO_DEVICE, &attr); | |
418 | .... | |
419 | ||
420 | Architectures that care about DMA_ATTR_FOO would check for its | |
421 | presence in their implementations of the mapping and unmapping | |
422 | routines, e.g.: | |
423 | ||
424 | void whizco_dma_map_sg_attrs(struct device *dev, dma_addr_t dma_addr, | |
425 | size_t size, enum dma_data_direction dir, | |
426 | struct dma_attrs *attrs) | |
427 | { | |
428 | .... | |
429 | int foo = dma_get_attr(DMA_ATTR_FOO, attrs); | |
430 | .... | |
431 | if (foo) | |
432 | /* twizzle the frobnozzle */ | |
433 | .... | |
434 | ||
1da177e4 LT |
435 | |
436 | Part II - Advanced dma_ usage | |
437 | ----------------------------- | |
438 | ||
f5a69f4c FT |
439 | Warning: These pieces of the DMA API should not be used in the |
440 | majority of cases, since they cater for unlikely corner cases that | |
441 | don't belong in usual drivers. | |
1da177e4 LT |
442 | |
443 | If you don't understand how cache line coherency works between a | |
444 | processor and an I/O device, you should not be using this part of the | |
445 | API at all. | |
446 | ||
447 | void * | |
448 | dma_alloc_noncoherent(struct device *dev, size_t size, | |
a12e2c6c | 449 | dma_addr_t *dma_handle, gfp_t flag) |
1da177e4 LT |
450 | |
451 | Identical to dma_alloc_coherent() except that the platform will | |
452 | choose to return either consistent or non-consistent memory as it sees | |
453 | fit. By using this API, you are guaranteeing to the platform that you | |
454 | have all the correct and necessary sync points for this memory in the | |
455 | driver should it choose to return non-consistent memory. | |
456 | ||
457 | Note: where the platform can return consistent memory, it will | |
458 | guarantee that the sync points become nops. | |
459 | ||
460 | Warning: Handling non-consistent memory is a real pain. You should | |
461 | only ever use this API if you positively know your driver will be | |
462 | required to work on one of the rare (usually non-PCI) architectures | |
463 | that simply cannot make consistent memory. | |
464 | ||
465 | void | |
466 | dma_free_noncoherent(struct device *dev, size_t size, void *cpu_addr, | |
467 | dma_addr_t dma_handle) | |
468 | ||
a12e2c6c | 469 | Free memory allocated by the nonconsistent API. All parameters must |
1da177e4 LT |
470 | be identical to those passed in (and returned by |
471 | dma_alloc_noncoherent()). | |
472 | ||
1da177e4 LT |
473 | int |
474 | dma_get_cache_alignment(void) | |
475 | ||
a12e2c6c | 476 | Returns the processor cache alignment. This is the absolute minimum |
1da177e4 LT |
477 | alignment *and* width that you must observe when either mapping |
478 | memory or doing partial flushes. | |
479 | ||
480 | Notes: This API may return a number *larger* than the actual cache | |
481 | line, but it will guarantee that one or more cache lines fit exactly | |
482 | into the width returned by this call. It will also always be a power | |
a12e2c6c | 483 | of two for easy alignment. |
1da177e4 | 484 | |
1da177e4 | 485 | void |
d3fa72e4 | 486 | dma_cache_sync(struct device *dev, void *vaddr, size_t size, |
1da177e4 LT |
487 | enum dma_data_direction direction) |
488 | ||
489 | Do a partial sync of memory that was allocated by | |
490 | dma_alloc_noncoherent(), starting at virtual address vaddr and | |
491 | continuing on for size. Again, you *must* observe the cache line | |
492 | boundaries when doing this. | |
493 | ||
494 | int | |
495 | dma_declare_coherent_memory(struct device *dev, dma_addr_t bus_addr, | |
496 | dma_addr_t device_addr, size_t size, int | |
497 | flags) | |
498 | ||
1da177e4 LT |
499 | Declare region of memory to be handed out by dma_alloc_coherent when |
500 | it's asked for coherent memory for this device. | |
501 | ||
502 | bus_addr is the physical address to which the memory is currently | |
503 | assigned in the bus responding region (this will be used by the | |
a12e2c6c | 504 | platform to perform the mapping). |
1da177e4 LT |
505 | |
506 | device_addr is the physical address the device needs to be programmed | |
507 | with actually to address this memory (this will be handed out as the | |
a12e2c6c | 508 | dma_addr_t in dma_alloc_coherent()). |
1da177e4 LT |
509 | |
510 | size is the size of the area (must be multiples of PAGE_SIZE). | |
511 | ||
a12e2c6c | 512 | flags can be or'd together and are: |
1da177e4 LT |
513 | |
514 | DMA_MEMORY_MAP - request that the memory returned from | |
4ae0edc2 | 515 | dma_alloc_coherent() be directly writable. |
1da177e4 LT |
516 | |
517 | DMA_MEMORY_IO - request that the memory returned from | |
518 | dma_alloc_coherent() be addressable using read/write/memcpy_toio etc. | |
519 | ||
a12e2c6c | 520 | One or both of these flags must be present. |
1da177e4 LT |
521 | |
522 | DMA_MEMORY_INCLUDES_CHILDREN - make the declared memory be allocated by | |
523 | dma_alloc_coherent of any child devices of this one (for memory residing | |
524 | on a bridge). | |
525 | ||
526 | DMA_MEMORY_EXCLUSIVE - only allocate memory from the declared regions. | |
527 | Do not allow dma_alloc_coherent() to fall back to system memory when | |
528 | it's out of memory in the declared region. | |
529 | ||
530 | The return value will be either DMA_MEMORY_MAP or DMA_MEMORY_IO and | |
531 | must correspond to a passed in flag (i.e. no returning DMA_MEMORY_IO | |
532 | if only DMA_MEMORY_MAP were passed in) for success or zero for | |
533 | failure. | |
534 | ||
535 | Note, for DMA_MEMORY_IO returns, all subsequent memory returned by | |
536 | dma_alloc_coherent() may no longer be accessed directly, but instead | |
537 | must be accessed using the correct bus functions. If your driver | |
538 | isn't prepared to handle this contingency, it should not specify | |
539 | DMA_MEMORY_IO in the input flags. | |
540 | ||
541 | As a simplification for the platforms, only *one* such region of | |
542 | memory may be declared per device. | |
543 | ||
544 | For reasons of efficiency, most platforms choose to track the declared | |
545 | region only at the granularity of a page. For smaller allocations, | |
546 | you should use the dma_pool() API. | |
547 | ||
548 | void | |
549 | dma_release_declared_memory(struct device *dev) | |
550 | ||
551 | Remove the memory region previously declared from the system. This | |
552 | API performs *no* in-use checking for this region and will return | |
553 | unconditionally having removed all the required structures. It is the | |
a12e2c6c | 554 | driver's job to ensure that no parts of this memory region are |
1da177e4 LT |
555 | currently in use. |
556 | ||
557 | void * | |
558 | dma_mark_declared_memory_occupied(struct device *dev, | |
559 | dma_addr_t device_addr, size_t size) | |
560 | ||
561 | This is used to occupy specific regions of the declared space | |
562 | (dma_alloc_coherent() will hand out the first free region it finds). | |
563 | ||
a12e2c6c | 564 | device_addr is the *device* address of the region requested. |
1da177e4 | 565 | |
a12e2c6c | 566 | size is the size (and should be a page-sized multiple). |
1da177e4 LT |
567 | |
568 | The return value will be either a pointer to the processor virtual | |
569 | address of the memory, or an error (via PTR_ERR()) if any part of the | |
570 | region is occupied. | |
187f9c3f JR |
571 | |
572 | Part III - Debug drivers use of the DMA-API | |
573 | ------------------------------------------- | |
574 | ||
575 | The DMA-API as described above as some constraints. DMA addresses must be | |
576 | released with the corresponding function with the same size for example. With | |
577 | the advent of hardware IOMMUs it becomes more and more important that drivers | |
578 | do not violate those constraints. In the worst case such a violation can | |
579 | result in data corruption up to destroyed filesystems. | |
580 | ||
581 | To debug drivers and find bugs in the usage of the DMA-API checking code can | |
582 | be compiled into the kernel which will tell the developer about those | |
583 | violations. If your architecture supports it you can select the "Enable | |
584 | debugging of DMA-API usage" option in your kernel configuration. Enabling this | |
585 | option has a performance impact. Do not enable it in production kernels. | |
586 | ||
587 | If you boot the resulting kernel will contain code which does some bookkeeping | |
588 | about what DMA memory was allocated for which device. If this code detects an | |
589 | error it prints a warning message with some details into your kernel log. An | |
590 | example warning message may look like this: | |
591 | ||
592 | ------------[ cut here ]------------ | |
593 | WARNING: at /data2/repos/linux-2.6-iommu/lib/dma-debug.c:448 | |
594 | check_unmap+0x203/0x490() | |
595 | Hardware name: | |
596 | forcedeth 0000:00:08.0: DMA-API: device driver frees DMA memory with wrong | |
597 | function [device address=0x00000000640444be] [size=66 bytes] [mapped as | |
598 | single] [unmapped as page] | |
599 | Modules linked in: nfsd exportfs bridge stp llc r8169 | |
600 | Pid: 0, comm: swapper Tainted: G W 2.6.28-dmatest-09289-g8bb99c0 #1 | |
601 | Call Trace: | |
602 | <IRQ> [<ffffffff80240b22>] warn_slowpath+0xf2/0x130 | |
603 | [<ffffffff80647b70>] _spin_unlock+0x10/0x30 | |
604 | [<ffffffff80537e75>] usb_hcd_link_urb_to_ep+0x75/0xc0 | |
605 | [<ffffffff80647c22>] _spin_unlock_irqrestore+0x12/0x40 | |
606 | [<ffffffff8055347f>] ohci_urb_enqueue+0x19f/0x7c0 | |
607 | [<ffffffff80252f96>] queue_work+0x56/0x60 | |
608 | [<ffffffff80237e10>] enqueue_task_fair+0x20/0x50 | |
609 | [<ffffffff80539279>] usb_hcd_submit_urb+0x379/0xbc0 | |
610 | [<ffffffff803b78c3>] cpumask_next_and+0x23/0x40 | |
611 | [<ffffffff80235177>] find_busiest_group+0x207/0x8a0 | |
612 | [<ffffffff8064784f>] _spin_lock_irqsave+0x1f/0x50 | |
613 | [<ffffffff803c7ea3>] check_unmap+0x203/0x490 | |
614 | [<ffffffff803c8259>] debug_dma_unmap_page+0x49/0x50 | |
615 | [<ffffffff80485f26>] nv_tx_done_optimized+0xc6/0x2c0 | |
616 | [<ffffffff80486c13>] nv_nic_irq_optimized+0x73/0x2b0 | |
617 | [<ffffffff8026df84>] handle_IRQ_event+0x34/0x70 | |
618 | [<ffffffff8026ffe9>] handle_edge_irq+0xc9/0x150 | |
619 | [<ffffffff8020e3ab>] do_IRQ+0xcb/0x1c0 | |
620 | [<ffffffff8020c093>] ret_from_intr+0x0/0xa | |
621 | <EOI> <4>---[ end trace f6435a98e2a38c0e ]--- | |
622 | ||
623 | The driver developer can find the driver and the device including a stacktrace | |
624 | of the DMA-API call which caused this warning. | |
625 | ||
626 | Per default only the first error will result in a warning message. All other | |
627 | errors will only silently counted. This limitation exist to prevent the code | |
628 | from flooding your kernel log. To support debugging a device driver this can | |
629 | be disabled via debugfs. See the debugfs interface documentation below for | |
630 | details. | |
631 | ||
632 | The debugfs directory for the DMA-API debugging code is called dma-api/. In | |
633 | this directory the following files can currently be found: | |
634 | ||
635 | dma-api/all_errors This file contains a numeric value. If this | |
636 | value is not equal to zero the debugging code | |
637 | will print a warning for every error it finds | |
19f59460 ML |
638 | into the kernel log. Be careful with this |
639 | option, as it can easily flood your logs. | |
187f9c3f JR |
640 | |
641 | dma-api/disabled This read-only file contains the character 'Y' | |
642 | if the debugging code is disabled. This can | |
643 | happen when it runs out of memory or if it was | |
644 | disabled at boot time | |
645 | ||
646 | dma-api/error_count This file is read-only and shows the total | |
647 | numbers of errors found. | |
648 | ||
649 | dma-api/num_errors The number in this file shows how many | |
650 | warnings will be printed to the kernel log | |
651 | before it stops. This number is initialized to | |
652 | one at system boot and be set by writing into | |
653 | this file | |
654 | ||
655 | dma-api/min_free_entries | |
656 | This read-only file can be read to get the | |
657 | minimum number of free dma_debug_entries the | |
658 | allocator has ever seen. If this value goes | |
659 | down to zero the code will disable itself | |
660 | because it is not longer reliable. | |
661 | ||
662 | dma-api/num_free_entries | |
663 | The current number of free dma_debug_entries | |
664 | in the allocator. | |
665 | ||
016ea687 JR |
666 | dma-api/driver-filter |
667 | You can write a name of a driver into this file | |
668 | to limit the debug output to requests from that | |
669 | particular driver. Write an empty string to | |
670 | that file to disable the filter and see | |
671 | all errors again. | |
672 | ||
187f9c3f JR |
673 | If you have this code compiled into your kernel it will be enabled by default. |
674 | If you want to boot without the bookkeeping anyway you can provide | |
675 | 'dma_debug=off' as a boot parameter. This will disable DMA-API debugging. | |
676 | Notice that you can not enable it again at runtime. You have to reboot to do | |
677 | so. | |
678 | ||
016ea687 JR |
679 | If you want to see debug messages only for a special device driver you can |
680 | specify the dma_debug_driver=<drivername> parameter. This will enable the | |
681 | driver filter at boot time. The debug code will only print errors for that | |
682 | driver afterwards. This filter can be disabled or changed later using debugfs. | |
683 | ||
187f9c3f JR |
684 | When the code disables itself at runtime this is most likely because it ran |
685 | out of dma_debug_entries. These entries are preallocated at boot. The number | |
686 | of preallocated entries is defined per architecture. If it is too low for you | |
687 | boot with 'dma_debug_entries=<your_desired_number>' to overwrite the | |
688 | architectural default. | |
6c9c6d63 SK |
689 | |
690 | void debug_dmap_mapping_error(struct device *dev, dma_addr_t dma_addr); | |
691 | ||
692 | dma-debug interface debug_dma_mapping_error() to debug drivers that fail | |
693 | to check dma mapping errors on addresses returned by dma_map_single() and | |
694 | dma_map_page() interfaces. This interface clears a flag set by | |
695 | debug_dma_map_page() to indicate that dma_mapping_error() has been called by | |
696 | the driver. When driver does unmap, debug_dma_unmap() checks the flag and if | |
697 | this flag is still set, prints warning message that includes call trace that | |
698 | leads up to the unmap. This interface can be called from dma_mapping_error() | |
699 | routines to enable dma mapping error check debugging. | |
700 |