1 DMA Buffer Sharing API Guide 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3 4 Sumit Semwal 5 <sumit dot semwal at linaro dot org> 6 <sumit dot semwal at ti dot com> 7 8This document serves as a guide to device-driver writers on what is the dma-buf 9buffer sharing API, how to use it for exporting and using shared buffers. 10 11Any device driver which wishes to be a part of DMA buffer sharing, can do so as 12either the 'exporter' of buffers, or the 'user' of buffers. 13 14Say a driver A wants to use buffers created by driver B, then we call B as the 15exporter, and A as buffer-user. 16 17The exporter 18- implements and manages operations[1] for the buffer 19- allows other users to share the buffer by using dma_buf sharing APIs, 20- manages the details of buffer allocation, 21- decides about the actual backing storage where this allocation happens, 22- takes care of any migration of scatterlist - for all (shared) users of this 23 buffer, 24 25The buffer-user 26- is one of (many) sharing users of the buffer. 27- doesn't need to worry about how the buffer is allocated, or where. 28- needs a mechanism to get access to the scatterlist that makes up this buffer 29 in memory, mapped into its own address space, so it can access the same area 30 of memory. 31 32dma-buf operations for device dma only 33-------------------------------------- 34 35The dma_buf buffer sharing API usage contains the following steps: 36 371. Exporter announces that it wishes to export a buffer 382. Userspace gets the file descriptor associated with the exported buffer, and 39 passes it around to potential buffer-users based on use case 403. Each buffer-user 'connects' itself to the buffer 414. When needed, buffer-user requests access to the buffer from exporter 425. When finished with its use, the buffer-user notifies end-of-DMA to exporter 436. when buffer-user is done using this buffer completely, it 'disconnects' 44 itself from the buffer. 45 46 471. Exporter's announcement of buffer export 48 49 The buffer exporter announces its wish to export a buffer. In this, it 50 connects its own private buffer data, provides implementation for operations 51 that can be performed on the exported dma_buf, and flags for the file 52 associated with this buffer. 53 54 Interface: 55 struct dma_buf *dma_buf_export_named(void *priv, struct dma_buf_ops *ops, 56 size_t size, int flags, 57 const char *exp_name) 58 59 If this succeeds, dma_buf_export allocates a dma_buf structure, and returns a 60 pointer to the same. It also associates an anonymous file with this buffer, 61 so it can be exported. On failure to allocate the dma_buf object, it returns 62 NULL. 63 64 'exp_name' is the name of exporter - to facilitate information while 65 debugging. 66 67 Exporting modules which do not wish to provide any specific name may use the 68 helper define 'dma_buf_export()', with the same arguments as above, but 69 without the last argument; a __FILE__ pre-processor directive will be 70 inserted in place of 'exp_name' instead. 71 722. Userspace gets a handle to pass around to potential buffer-users 73 74 Userspace entity requests for a file-descriptor (fd) which is a handle to the 75 anonymous file associated with the buffer. It can then share the fd with other 76 drivers and/or processes. 77 78 Interface: 79 int dma_buf_fd(struct dma_buf *dmabuf) 80 81 This API installs an fd for the anonymous file associated with this buffer; 82 returns either 'fd', or error. 83 843. Each buffer-user 'connects' itself to the buffer 85 86 Each buffer-user now gets a reference to the buffer, using the fd passed to 87 it. 88 89 Interface: 90 struct dma_buf *dma_buf_get(int fd) 91 92 This API will return a reference to the dma_buf, and increment refcount for 93 it. 94 95 After this, the buffer-user needs to attach its device with the buffer, which 96 helps the exporter to know of device buffer constraints. 97 98 Interface: 99 struct dma_buf_attachment *dma_buf_attach(struct dma_buf *dmabuf, 100 struct device *dev) 101 102 This API returns reference to an attachment structure, which is then used 103 for scatterlist operations. It will optionally call the 'attach' dma_buf 104 operation, if provided by the exporter. 105 106 The dma-buf sharing framework does the bookkeeping bits related to managing 107 the list of all attachments to a buffer. 108 109Until this stage, the buffer-exporter has the option to choose not to actually 110allocate the backing storage for this buffer, but wait for the first buffer-user 111to request use of buffer for allocation. 112 113 1144. When needed, buffer-user requests access to the buffer 115 116 Whenever a buffer-user wants to use the buffer for any DMA, it asks for 117 access to the buffer using dma_buf_map_attachment API. At least one attach to 118 the buffer must have happened before map_dma_buf can be called. 119 120 Interface: 121 struct sg_table * dma_buf_map_attachment(struct dma_buf_attachment *, 122 enum dma_data_direction); 123 124 This is a wrapper to dma_buf->ops->map_dma_buf operation, which hides the 125 "dma_buf->ops->" indirection from the users of this interface. 126 127 In struct dma_buf_ops, map_dma_buf is defined as 128 struct sg_table * (*map_dma_buf)(struct dma_buf_attachment *, 129 enum dma_data_direction); 130 131 It is one of the buffer operations that must be implemented by the exporter. 132 It should return the sg_table containing scatterlist for this buffer, mapped 133 into caller's address space. 134 135 If this is being called for the first time, the exporter can now choose to 136 scan through the list of attachments for this buffer, collate the requirements 137 of the attached devices, and choose an appropriate backing storage for the 138 buffer. 139 140 Based on enum dma_data_direction, it might be possible to have multiple users 141 accessing at the same time (for reading, maybe), or any other kind of sharing 142 that the exporter might wish to make available to buffer-users. 143 144 map_dma_buf() operation can return -EINTR if it is interrupted by a signal. 145 146 1475. When finished, the buffer-user notifies end-of-DMA to exporter 148 149 Once the DMA for the current buffer-user is over, it signals 'end-of-DMA' to 150 the exporter using the dma_buf_unmap_attachment API. 151 152 Interface: 153 void dma_buf_unmap_attachment(struct dma_buf_attachment *, 154 struct sg_table *); 155 156 This is a wrapper to dma_buf->ops->unmap_dma_buf() operation, which hides the 157 "dma_buf->ops->" indirection from the users of this interface. 158 159 In struct dma_buf_ops, unmap_dma_buf is defined as 160 void (*unmap_dma_buf)(struct dma_buf_attachment *, struct sg_table *); 161 162 unmap_dma_buf signifies the end-of-DMA for the attachment provided. Like 163 map_dma_buf, this API also must be implemented by the exporter. 164 165 1666. when buffer-user is done using this buffer, it 'disconnects' itself from the 167 buffer. 168 169 After the buffer-user has no more interest in using this buffer, it should 170 disconnect itself from the buffer: 171 172 - it first detaches itself from the buffer. 173 174 Interface: 175 void dma_buf_detach(struct dma_buf *dmabuf, 176 struct dma_buf_attachment *dmabuf_attach); 177 178 This API removes the attachment from the list in dmabuf, and optionally calls 179 dma_buf->ops->detach(), if provided by exporter, for any housekeeping bits. 180 181 - Then, the buffer-user returns the buffer reference to exporter. 182 183 Interface: 184 void dma_buf_put(struct dma_buf *dmabuf); 185 186 This API then reduces the refcount for this buffer. 187 188 If, as a result of this call, the refcount becomes 0, the 'release' file 189 operation related to this fd is called. It calls the dmabuf->ops->release() 190 operation in turn, and frees the memory allocated for dmabuf when exported. 191 192NOTES: 193- Importance of attach-detach and {map,unmap}_dma_buf operation pairs 194 The attach-detach calls allow the exporter to figure out backing-storage 195 constraints for the currently-interested devices. This allows preferential 196 allocation, and/or migration of pages across different types of storage 197 available, if possible. 198 199 Bracketing of DMA access with {map,unmap}_dma_buf operations is essential 200 to allow just-in-time backing of storage, and migration mid-way through a 201 use-case. 202 203- Migration of backing storage if needed 204 If after 205 - at least one map_dma_buf has happened, 206 - and the backing storage has been allocated for this buffer, 207 another new buffer-user intends to attach itself to this buffer, it might 208 be allowed, if possible for the exporter. 209 210 In case it is allowed by the exporter: 211 if the new buffer-user has stricter 'backing-storage constraints', and the 212 exporter can handle these constraints, the exporter can just stall on the 213 map_dma_buf until all outstanding access is completed (as signalled by 214 unmap_dma_buf). 215 Once all users have finished accessing and have unmapped this buffer, the 216 exporter could potentially move the buffer to the stricter backing-storage, 217 and then allow further {map,unmap}_dma_buf operations from any buffer-user 218 from the migrated backing-storage. 219 220 If the exporter cannot fulfil the backing-storage constraints of the new 221 buffer-user device as requested, dma_buf_attach() would return an error to 222 denote non-compatibility of the new buffer-sharing request with the current 223 buffer. 224 225 If the exporter chooses not to allow an attach() operation once a 226 map_dma_buf() API has been called, it simply returns an error. 227 228Kernel cpu access to a dma-buf buffer object 229-------------------------------------------- 230 231The motivation to allow cpu access from the kernel to a dma-buf object from the 232importers side are: 233- fallback operations, e.g. if the devices is connected to a usb bus and the 234 kernel needs to shuffle the data around first before sending it away. 235- full transparency for existing users on the importer side, i.e. userspace 236 should not notice the difference between a normal object from that subsystem 237 and an imported one backed by a dma-buf. This is really important for drm 238 opengl drivers that expect to still use all the existing upload/download 239 paths. 240 241Access to a dma_buf from the kernel context involves three steps: 242 2431. Prepare access, which invalidate any necessary caches and make the object 244 available for cpu access. 2452. Access the object page-by-page with the dma_buf map apis 2463. Finish access, which will flush any necessary cpu caches and free reserved 247 resources. 248 2491. Prepare access 250 251 Before an importer can access a dma_buf object with the cpu from the kernel 252 context, it needs to notify the exporter of the access that is about to 253 happen. 254 255 Interface: 256 int dma_buf_begin_cpu_access(struct dma_buf *dmabuf, 257 size_t start, size_t len, 258 enum dma_data_direction direction) 259 260 This allows the exporter to ensure that the memory is actually available for 261 cpu access - the exporter might need to allocate or swap-in and pin the 262 backing storage. The exporter also needs to ensure that cpu access is 263 coherent for the given range and access direction. The range and access 264 direction can be used by the exporter to optimize the cache flushing, i.e. 265 access outside of the range or with a different direction (read instead of 266 write) might return stale or even bogus data (e.g. when the exporter needs to 267 copy the data to temporary storage). 268 269 This step might fail, e.g. in oom conditions. 270 2712. Accessing the buffer 272 273 To support dma_buf objects residing in highmem cpu access is page-based using 274 an api similar to kmap. Accessing a dma_buf is done in aligned chunks of 275 PAGE_SIZE size. Before accessing a chunk it needs to be mapped, which returns 276 a pointer in kernel virtual address space. Afterwards the chunk needs to be 277 unmapped again. There is no limit on how often a given chunk can be mapped 278 and unmapped, i.e. the importer does not need to call begin_cpu_access again 279 before mapping the same chunk again. 280 281 Interfaces: 282 void *dma_buf_kmap(struct dma_buf *, unsigned long); 283 void dma_buf_kunmap(struct dma_buf *, unsigned long, void *); 284 285 There are also atomic variants of these interfaces. Like for kmap they 286 facilitate non-blocking fast-paths. Neither the importer nor the exporter (in 287 the callback) is allowed to block when using these. 288 289 Interfaces: 290 void *dma_buf_kmap_atomic(struct dma_buf *, unsigned long); 291 void dma_buf_kunmap_atomic(struct dma_buf *, unsigned long, void *); 292 293 For importers all the restrictions of using kmap apply, like the limited 294 supply of kmap_atomic slots. Hence an importer shall only hold onto at most 2 295 atomic dma_buf kmaps at the same time (in any given process context). 296 297 dma_buf kmap calls outside of the range specified in begin_cpu_access are 298 undefined. If the range is not PAGE_SIZE aligned, kmap needs to succeed on 299 the partial chunks at the beginning and end but may return stale or bogus 300 data outside of the range (in these partial chunks). 301 302 Note that these calls need to always succeed. The exporter needs to complete 303 any preparations that might fail in begin_cpu_access. 304 305 For some cases the overhead of kmap can be too high, a vmap interface 306 is introduced. This interface should be used very carefully, as vmalloc 307 space is a limited resources on many architectures. 308 309 Interfaces: 310 void *dma_buf_vmap(struct dma_buf *dmabuf) 311 void dma_buf_vunmap(struct dma_buf *dmabuf, void *vaddr) 312 313 The vmap call can fail if there is no vmap support in the exporter, or if it 314 runs out of vmalloc space. Fallback to kmap should be implemented. Note that 315 the dma-buf layer keeps a reference count for all vmap access and calls down 316 into the exporter's vmap function only when no vmapping exists, and only 317 unmaps it once. Protection against concurrent vmap/vunmap calls is provided 318 by taking the dma_buf->lock mutex. 319 3203. Finish access 321 322 When the importer is done accessing the range specified in begin_cpu_access, 323 it needs to announce this to the exporter (to facilitate cache flushing and 324 unpinning of any pinned resources). The result of of any dma_buf kmap calls 325 after end_cpu_access is undefined. 326 327 Interface: 328 void dma_buf_end_cpu_access(struct dma_buf *dma_buf, 329 size_t start, size_t len, 330 enum dma_data_direction dir); 331 332 333Direct Userspace Access/mmap Support 334------------------------------------ 335 336Being able to mmap an export dma-buf buffer object has 2 main use-cases: 337- CPU fallback processing in a pipeline and 338- supporting existing mmap interfaces in importers. 339 3401. CPU fallback processing in a pipeline 341 342 In many processing pipelines it is sometimes required that the cpu can access 343 the data in a dma-buf (e.g. for thumbnail creation, snapshots, ...). To avoid 344 the need to handle this specially in userspace frameworks for buffer sharing 345 it's ideal if the dma_buf fd itself can be used to access the backing storage 346 from userspace using mmap. 347 348 Furthermore Android's ION framework already supports this (and is otherwise 349 rather similar to dma-buf from a userspace consumer side with using fds as 350 handles, too). So it's beneficial to support this in a similar fashion on 351 dma-buf to have a good transition path for existing Android userspace. 352 353 No special interfaces, userspace simply calls mmap on the dma-buf fd. 354 3552. Supporting existing mmap interfaces in exporters 356 357 Similar to the motivation for kernel cpu access it is again important that 358 the userspace code of a given importing subsystem can use the same interfaces 359 with a imported dma-buf buffer object as with a native buffer object. This is 360 especially important for drm where the userspace part of contemporary OpenGL, 361 X, and other drivers is huge, and reworking them to use a different way to 362 mmap a buffer rather invasive. 363 364 The assumption in the current dma-buf interfaces is that redirecting the 365 initial mmap is all that's needed. A survey of some of the existing 366 subsystems shows that no driver seems to do any nefarious thing like syncing 367 up with outstanding asynchronous processing on the device or allocating 368 special resources at fault time. So hopefully this is good enough, since 369 adding interfaces to intercept pagefaults and allow pte shootdowns would 370 increase the complexity quite a bit. 371 372 Interface: 373 int dma_buf_mmap(struct dma_buf *, struct vm_area_struct *, 374 unsigned long); 375 376 If the importing subsystem simply provides a special-purpose mmap call to set 377 up a mapping in userspace, calling do_mmap with dma_buf->file will equally 378 achieve that for a dma-buf object. 379 3803. Implementation notes for exporters 381 382 Because dma-buf buffers have invariant size over their lifetime, the dma-buf 383 core checks whether a vma is too large and rejects such mappings. The 384 exporter hence does not need to duplicate this check. 385 386 Because existing importing subsystems might presume coherent mappings for 387 userspace, the exporter needs to set up a coherent mapping. If that's not 388 possible, it needs to fake coherency by manually shooting down ptes when 389 leaving the cpu domain and flushing caches at fault time. Note that all the 390 dma_buf files share the same anon inode, hence the exporter needs to replace 391 the dma_buf file stored in vma->vm_file with it's own if pte shootdown is 392 required. This is because the kernel uses the underlying inode's address_space 393 for vma tracking (and hence pte tracking at shootdown time with 394 unmap_mapping_range). 395 396 If the above shootdown dance turns out to be too expensive in certain 397 scenarios, we can extend dma-buf with a more explicit cache tracking scheme 398 for userspace mappings. But the current assumption is that using mmap is 399 always a slower path, so some inefficiencies should be acceptable. 400 401 Exporters that shoot down mappings (for any reasons) shall not do any 402 synchronization at fault time with outstanding device operations. 403 Synchronization is an orthogonal issue to sharing the backing storage of a 404 buffer and hence should not be handled by dma-buf itself. This is explicitly 405 mentioned here because many people seem to want something like this, but if 406 different exporters handle this differently, buffer sharing can fail in 407 interesting ways depending upong the exporter (if userspace starts depending 408 upon this implicit synchronization). 409 410Miscellaneous notes 411------------------- 412 413- Any exporters or users of the dma-buf buffer sharing framework must have 414 a 'select DMA_SHARED_BUFFER' in their respective Kconfigs. 415 416- In order to avoid fd leaks on exec, the FD_CLOEXEC flag must be set 417 on the file descriptor. This is not just a resource leak, but a 418 potential security hole. It could give the newly exec'd application 419 access to buffers, via the leaked fd, to which it should otherwise 420 not be permitted access. 421 422 The problem with doing this via a separate fcntl() call, versus doing it 423 atomically when the fd is created, is that this is inherently racy in a 424 multi-threaded app[3]. The issue is made worse when it is library code 425 opening/creating the file descriptor, as the application may not even be 426 aware of the fd's. 427 428 To avoid this problem, userspace must have a way to request O_CLOEXEC 429 flag be set when the dma-buf fd is created. So any API provided by 430 the exporting driver to create a dmabuf fd must provide a way to let 431 userspace control setting of O_CLOEXEC flag passed in to dma_buf_fd(). 432 433- If an exporter needs to manually flush caches and hence needs to fake 434 coherency for mmap support, it needs to be able to zap all the ptes pointing 435 at the backing storage. Now linux mm needs a struct address_space associated 436 with the struct file stored in vma->vm_file to do that with the function 437 unmap_mapping_range. But the dma_buf framework only backs every dma_buf fd 438 with the anon_file struct file, i.e. all dma_bufs share the same file. 439 440 Hence exporters need to setup their own file (and address_space) association 441 by setting vma->vm_file and adjusting vma->vm_pgoff in the dma_buf mmap 442 callback. In the specific case of a gem driver the exporter could use the 443 shmem file already provided by gem (and set vm_pgoff = 0). Exporters can then 444 zap ptes by unmapping the corresponding range of the struct address_space 445 associated with their own file. 446 447References: 448[1] struct dma_buf_ops in include/linux/dma-buf.h 449[2] All interfaces mentioned above defined in include/linux/dma-buf.h 450[3] https://lwn.net/Articles/236486/ 451