linux/Documentation/DMA-API.txt
<<
>>
Prefs
   1               Dynamic DMA mapping using the generic device
   2               ============================================
   3
   4        James E.J. Bottomley <James.Bottomley@HansenPartnership.com>
   5
   6This document describes the DMA API.  For a more gentle introduction
   7phrased in terms of the pci_ equivalents (and actual examples) see
   8Documentation/PCI/PCI-DMA-mapping.txt.
   9
  10This API is split into two pieces.  Part I describes the API and the
  11corresponding pci_ API.  Part II describes the extensions to the API
  12for supporting non-consistent memory machines.  Unless you know that
  13your driver absolutely has to support non-consistent platforms (this
  14is usually only legacy platforms) you should only use the API
  15described in part I.
  16
  17Part I - pci_ and dma_ Equivalent API 
  18-------------------------------------
  19
  20To get the pci_ API, you must #include <linux/pci.h>
  21To get the dma_ API, you must #include <linux/dma-mapping.h>
  22
  23
  24Part Ia - Using large dma-coherent buffers
  25------------------------------------------
  26
  27void *
  28dma_alloc_coherent(struct device *dev, size_t size,
  29                             dma_addr_t *dma_handle, gfp_t flag)
  30void *
  31pci_alloc_consistent(struct pci_dev *dev, size_t size,
  32                             dma_addr_t *dma_handle)
  33
  34Consistent memory is memory for which a write by either the device or
  35the processor can immediately be read by the processor or device
  36without having to worry about caching effects.  (You may however need
  37to make sure to flush the processor's write buffers before telling
  38devices to read that memory.)
  39
  40This routine allocates a region of <size> bytes of consistent memory.
  41It also returns a <dma_handle> which may be cast to an unsigned
  42integer the same width as the bus and used as the physical address
  43base of the region.
  44
  45Returns: a pointer to the allocated region (in the processor's virtual
  46address space) or NULL if the allocation failed.
  47
  48Note: consistent memory can be expensive on some platforms, and the
  49minimum allocation length may be as big as a page, so you should
  50consolidate your requests for consistent memory as much as possible.
  51The simplest way to do that is to use the dma_pool calls (see below).
  52
  53The flag parameter (dma_alloc_coherent only) allows the caller to
  54specify the GFP_ flags (see kmalloc) for the allocation (the
  55implementation may choose to ignore flags that affect the location of
  56the returned memory, like GFP_DMA).  For pci_alloc_consistent, you
  57must assume GFP_ATOMIC behaviour.
  58
  59void
  60dma_free_coherent(struct device *dev, size_t size, void *cpu_addr,
  61                           dma_addr_t dma_handle)
  62void
  63pci_free_consistent(struct pci_dev *dev, size_t size, void *cpu_addr,
  64                           dma_addr_t dma_handle)
  65
  66Free the region of consistent memory you previously allocated.  dev,
  67size and dma_handle must all be the same as those passed into the
  68consistent allocate.  cpu_addr must be the virtual address returned by
  69the consistent allocate.
  70
  71Note that unlike their sibling allocation calls, these routines
  72may only be called with IRQs enabled.
  73
  74
  75Part Ib - Using small dma-coherent buffers
  76------------------------------------------
  77
  78To get this part of the dma_ API, you must #include <linux/dmapool.h>
  79
  80Many drivers need lots of small dma-coherent memory regions for DMA
  81descriptors or I/O buffers.  Rather than allocating in units of a page
  82or more using dma_alloc_coherent(), you can use DMA pools.  These work
  83much like a struct kmem_cache, except that they use the dma-coherent allocator,
  84not __get_free_pages().  Also, they understand common hardware constraints
  85for alignment, like queue heads needing to be aligned on N-byte boundaries.
  86
  87
  88        struct dma_pool *
  89        dma_pool_create(const char *name, struct device *dev,
  90                        size_t size, size_t align, size_t alloc);
  91
  92        struct pci_pool *
  93        pci_pool_create(const char *name, struct pci_device *dev,
  94                        size_t size, size_t align, size_t alloc);
  95
  96The pool create() routines initialize a pool of dma-coherent buffers
  97for use with a given device.  It must be called in a context which
  98can sleep.
  99
 100The "name" is for diagnostics (like a struct kmem_cache name); dev and size
 101are like what you'd pass to dma_alloc_coherent().  The device's hardware
 102alignment requirement for this type of data is "align" (which is expressed
 103in bytes, and must be a power of two).  If your device has no boundary
 104crossing restrictions, pass 0 for alloc; passing 4096 says memory allocated
 105from this pool must not cross 4KByte boundaries.
 106
 107
 108        void *dma_pool_alloc(struct dma_pool *pool, gfp_t gfp_flags,
 109                        dma_addr_t *dma_handle);
 110
 111        void *pci_pool_alloc(struct pci_pool *pool, gfp_t gfp_flags,
 112                        dma_addr_t *dma_handle);
 113
 114This allocates memory from the pool; the returned memory will meet the size
 115and alignment requirements specified at creation time.  Pass GFP_ATOMIC to
 116prevent blocking, or if it's permitted (not in_interrupt, not holding SMP locks),
 117pass GFP_KERNEL to allow blocking.  Like dma_alloc_coherent(), this returns
 118two values:  an address usable by the cpu, and the dma address usable by the
 119pool's device.
 120
 121
 122        void dma_pool_free(struct dma_pool *pool, void *vaddr,
 123                        dma_addr_t addr);
 124
 125        void pci_pool_free(struct pci_pool *pool, void *vaddr,
 126                        dma_addr_t addr);
 127
 128This puts memory back into the pool.  The pool is what was passed to
 129the pool allocation routine; the cpu (vaddr) and dma addresses are what
 130were returned when that routine allocated the memory being freed.
 131
 132
 133        void dma_pool_destroy(struct dma_pool *pool);
 134
 135        void pci_pool_destroy(struct pci_pool *pool);
 136
 137The pool destroy() routines free the resources of the pool.  They must be
 138called in a context which can sleep.  Make sure you've freed all allocated
 139memory back to the pool before you destroy it.
 140
 141
 142Part Ic - DMA addressing limitations
 143------------------------------------
 144
 145int
 146dma_supported(struct device *dev, u64 mask)
 147int
 148pci_dma_supported(struct pci_dev *hwdev, u64 mask)
 149
 150Checks to see if the device can support DMA to the memory described by
 151mask.
 152
 153Returns: 1 if it can and 0 if it can't.
 154
 155Notes: This routine merely tests to see if the mask is possible.  It
 156won't change the current mask settings.  It is more intended as an
 157internal API for use by the platform than an external API for use by
 158driver writers.
 159
 160int
 161dma_set_mask(struct device *dev, u64 mask)
 162int
 163pci_set_dma_mask(struct pci_device *dev, u64 mask)
 164
 165Checks to see if the mask is possible and updates the device
 166parameters if it is.
 167
 168Returns: 0 if successful and a negative error if not.
 169
 170u64
 171dma_get_required_mask(struct device *dev)
 172
 173This API returns the mask that the platform requires to
 174operate efficiently.  Usually this means the returned mask
 175is the minimum required to cover all of memory.  Examining the
 176required mask gives drivers with variable descriptor sizes the
 177opportunity to use smaller descriptors as necessary.
 178
 179Requesting the required mask does not alter the current mask.  If you
 180wish to take advantage of it, you should issue a dma_set_mask()
 181call to set the mask to the value returned.
 182
 183
 184Part Id - Streaming DMA mappings
 185--------------------------------
 186
 187dma_addr_t
 188dma_map_single(struct device *dev, void *cpu_addr, size_t size,
 189                      enum dma_data_direction direction)
 190dma_addr_t
 191pci_map_single(struct pci_dev *hwdev, void *cpu_addr, size_t size,
 192                      int direction)
 193
 194Maps a piece of processor virtual memory so it can be accessed by the
 195device and returns the physical handle of the memory.
 196
 197The direction for both api's may be converted freely by casting.
 198However the dma_ API uses a strongly typed enumerator for its
 199direction:
 200
 201DMA_NONE                = PCI_DMA_NONE          no direction (used for
 202                                                debugging)
 203DMA_TO_DEVICE           = PCI_DMA_TODEVICE      data is going from the
 204                                                memory to the device
 205DMA_FROM_DEVICE         = PCI_DMA_FROMDEVICE    data is coming from
 206                                                the device to the
 207                                                memory
 208DMA_BIDIRECTIONAL       = PCI_DMA_BIDIRECTIONAL direction isn't known
 209
 210Notes:  Not all memory regions in a machine can be mapped by this
 211API.  Further, regions that appear to be physically contiguous in
 212kernel virtual space may not be contiguous as physical memory.  Since
 213this API does not provide any scatter/gather capability, it will fail
 214if the user tries to map a non-physically contiguous piece of memory.
 215For this reason, it is recommended that memory mapped by this API be
 216obtained only from sources which guarantee it to be physically contiguous
 217(like kmalloc).
 218
 219Further, the physical address of the memory must be within the
 220dma_mask of the device (the dma_mask represents a bit mask of the
 221addressable region for the device.  I.e., if the physical address of
 222the memory anded with the dma_mask is still equal to the physical
 223address, then the device can perform DMA to the memory).  In order to
 224ensure that the memory allocated by kmalloc is within the dma_mask,
 225the driver may specify various platform-dependent flags to restrict
 226the physical memory range of the allocation (e.g. on x86, GFP_DMA
 227guarantees to be within the first 16Mb of available physical memory,
 228as required by ISA devices).
 229
 230Note also that the above constraints on physical contiguity and
 231dma_mask may not apply if the platform has an IOMMU (a device which
 232supplies a physical to virtual mapping between the I/O memory bus and
 233the device).  However, to be portable, device driver writers may *not*
 234assume that such an IOMMU exists.
 235
 236Warnings:  Memory coherency operates at a granularity called the cache
 237line width.  In order for memory mapped by this API to operate
 238correctly, the mapped region must begin exactly on a cache line
 239boundary and end exactly on one (to prevent two separately mapped
 240regions from sharing a single cache line).  Since the cache line size
 241may not be known at compile time, the API will not enforce this
 242requirement.  Therefore, it is recommended that driver writers who
 243don't take special care to determine the cache line size at run time
 244only map virtual regions that begin and end on page boundaries (which
 245are guaranteed also to be cache line boundaries).
 246
 247DMA_TO_DEVICE synchronisation must be done after the last modification
 248of the memory region by the software and before it is handed off to
 249the driver.  Once this primitive is used, memory covered by this
 250primitive should be treated as read-only by the device.  If the device
 251may write to it at any point, it should be DMA_BIDIRECTIONAL (see
 252below).
 253
 254DMA_FROM_DEVICE synchronisation must be done before the driver
 255accesses data that may be changed by the device.  This memory should
 256be treated as read-only by the driver.  If the driver needs to write
 257to it at any point, it should be DMA_BIDIRECTIONAL (see below).
 258
 259DMA_BIDIRECTIONAL requires special handling: it means that the driver
 260isn't sure if the memory was modified before being handed off to the
 261device and also isn't sure if the device will also modify it.  Thus,
 262you must always sync bidirectional memory twice: once before the
 263memory is handed off to the device (to make sure all memory changes
 264are flushed from the processor) and once before the data may be
 265accessed after being used by the device (to make sure any processor
 266cache lines are updated with data that the device may have changed).
 267
 268void
 269dma_unmap_single(struct device *dev, dma_addr_t dma_addr, size_t size,
 270                 enum dma_data_direction direction)
 271void
 272pci_unmap_single(struct pci_dev *hwdev, dma_addr_t dma_addr,
 273                 size_t size, int direction)
 274
 275Unmaps the region previously mapped.  All the parameters passed in
 276must be identical to those passed in (and returned) by the mapping
 277API.
 278
 279dma_addr_t
 280dma_map_page(struct device *dev, struct page *page,
 281                    unsigned long offset, size_t size,
 282                    enum dma_data_direction direction)
 283dma_addr_t
 284pci_map_page(struct pci_dev *hwdev, struct page *page,
 285                    unsigned long offset, size_t size, int direction)
 286void
 287dma_unmap_page(struct device *dev, dma_addr_t dma_address, size_t size,
 288               enum dma_data_direction direction)
 289void
 290pci_unmap_page(struct pci_dev *hwdev, dma_addr_t dma_address,
 291               size_t size, int direction)
 292
 293API for mapping and unmapping for pages.  All the notes and warnings
 294for the other mapping APIs apply here.  Also, although the <offset>
 295and <size> parameters are provided to do partial page mapping, it is
 296recommended that you never use these unless you really know what the
 297cache width is.
 298
 299int
 300dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 301
 302int
 303pci_dma_mapping_error(struct pci_dev *hwdev, dma_addr_t dma_addr)
 304
 305In some circumstances dma_map_single and dma_map_page will fail to create
 306a mapping. A driver can check for these errors by testing the returned
 307dma address with dma_mapping_error(). A non-zero return value means the mapping
 308could not be created and the driver should take appropriate action (e.g.
 309reduce current DMA mapping usage or delay and try again later).
 310
 311        int
 312        dma_map_sg(struct device *dev, struct scatterlist *sg,
 313                int nents, enum dma_data_direction direction)
 314        int
 315        pci_map_sg(struct pci_dev *hwdev, struct scatterlist *sg,
 316                int nents, int direction)
 317
 318Returns: the number of physical segments mapped (this may be shorter
 319than <nents> passed in if some elements of the scatter/gather list are
 320physically or virtually adjacent and an IOMMU maps them with a single
 321entry).
 322
 323Please note that the sg cannot be mapped again if it has been mapped once.
 324The mapping process is allowed to destroy information in the sg.
 325
 326As with the other mapping interfaces, dma_map_sg can fail. When it
 327does, 0 is returned and a driver must take appropriate action. It is
 328critical that the driver do something, in the case of a block driver
 329aborting the request or even oopsing is better than doing nothing and
 330corrupting the filesystem.
 331
 332With scatterlists, you use the resulting mapping like this:
 333
 334        int i, count = dma_map_sg(dev, sglist, nents, direction);
 335        struct scatterlist *sg;
 336
 337        for_each_sg(sglist, sg, count, i) {
 338                hw_address[i] = sg_dma_address(sg);
 339                hw_len[i] = sg_dma_len(sg);
 340        }
 341
 342where nents is the number of entries in the sglist.
 343
 344The implementation is free to merge several consecutive sglist entries
 345into one (e.g. with an IOMMU, or if several pages just happen to be
 346physically contiguous) and returns the actual number of sg entries it
 347mapped them to. On failure 0, is returned.
 348
 349Then you should loop count times (note: this can be less than nents times)
 350and use sg_dma_address() and sg_dma_len() macros where you previously
 351accessed sg->address and sg->length as shown above.
 352
 353        void
 354        dma_unmap_sg(struct device *dev, struct scatterlist *sg,
 355                int nhwentries, enum dma_data_direction direction)
 356        void
 357        pci_unmap_sg(struct pci_dev *hwdev, struct scatterlist *sg,
 358                int nents, int direction)
 359
 360Unmap the previously mapped scatter/gather list.  All the parameters
 361must be the same as those and passed in to the scatter/gather mapping
 362API.
 363
 364Note: <nents> must be the number you passed in, *not* the number of
 365physical entries returned.
 366
 367void
 368dma_sync_single(struct device *dev, dma_addr_t dma_handle, size_t size,
 369                enum dma_data_direction direction)
 370void
 371pci_dma_sync_single(struct pci_dev *hwdev, dma_addr_t dma_handle,
 372                           size_t size, int direction)
 373void
 374dma_sync_sg(struct device *dev, struct scatterlist *sg, int nelems,
 375                          enum dma_data_direction direction)
 376void
 377pci_dma_sync_sg(struct pci_dev *hwdev, struct scatterlist *sg,
 378                       int nelems, int direction)
 379
 380Synchronise a single contiguous or scatter/gather mapping.  All the
 381parameters must be the same as those passed into the single mapping
 382API.
 383
 384Notes:  You must do this:
 385
 386- Before reading values that have been written by DMA from the device
 387  (use the DMA_FROM_DEVICE direction)
 388- After writing values that will be written to the device using DMA
 389  (use the DMA_TO_DEVICE) direction
 390- before *and* after handing memory to the device if the memory is
 391  DMA_BIDIRECTIONAL
 392
 393See also dma_map_single().
 394
 395dma_addr_t
 396dma_map_single_attrs(struct device *dev, void *cpu_addr, size_t size,
 397                     enum dma_data_direction dir,
 398                     struct dma_attrs *attrs)
 399
 400void
 401dma_unmap_single_attrs(struct device *dev, dma_addr_t dma_addr,
 402                       size_t size, enum dma_data_direction dir,
 403                       struct dma_attrs *attrs)
 404
 405int
 406dma_map_sg_attrs(struct device *dev, struct scatterlist *sgl,
 407                 int nents, enum dma_data_direction dir,
 408                 struct dma_attrs *attrs)
 409
 410void
 411dma_unmap_sg_attrs(struct device *dev, struct scatterlist *sgl,
 412                   int nents, enum dma_data_direction dir,
 413                   struct dma_attrs *attrs)
 414
 415The four functions above are just like the counterpart functions
 416without the _attrs suffixes, except that they pass an optional
 417struct dma_attrs*.
 418
 419struct dma_attrs encapsulates a set of "dma attributes". For the
 420definition of struct dma_attrs see linux/dma-attrs.h.
 421
 422The interpretation of dma attributes is architecture-specific, and
 423each attribute should be documented in Documentation/DMA-attributes.txt.
 424
 425If struct dma_attrs* is NULL, the semantics of each of these
 426functions is identical to those of the corresponding function
 427without the _attrs suffix. As a result dma_map_single_attrs()
 428can generally replace dma_map_single(), etc.
 429
 430As an example of the use of the *_attrs functions, here's how
 431you could pass an attribute DMA_ATTR_FOO when mapping memory
 432for DMA:
 433
 434#include <linux/dma-attrs.h>
 435/* DMA_ATTR_FOO should be defined in linux/dma-attrs.h and
 436 * documented in Documentation/DMA-attributes.txt */
 437...
 438
 439        DEFINE_DMA_ATTRS(attrs);
 440        dma_set_attr(DMA_ATTR_FOO, &attrs);
 441        ....
 442        n = dma_map_sg_attrs(dev, sg, nents, DMA_TO_DEVICE, &attr);
 443        ....
 444
 445Architectures that care about DMA_ATTR_FOO would check for its
 446presence in their implementations of the mapping and unmapping
 447routines, e.g.:
 448
 449void whizco_dma_map_sg_attrs(struct device *dev, dma_addr_t dma_addr,
 450                             size_t size, enum dma_data_direction dir,
 451                             struct dma_attrs *attrs)
 452{
 453        ....
 454        int foo =  dma_get_attr(DMA_ATTR_FOO, attrs);
 455        ....
 456        if (foo)
 457                /* twizzle the frobnozzle */
 458        ....
 459
 460
 461Part II - Advanced dma_ usage
 462-----------------------------
 463
 464Warning: These pieces of the DMA API have no PCI equivalent.  They
 465should also not be used in the majority of cases, since they cater for
 466unlikely corner cases that don't belong in usual drivers.
 467
 468If you don't understand how cache line coherency works between a
 469processor and an I/O device, you should not be using this part of the
 470API at all.
 471
 472void *
 473dma_alloc_noncoherent(struct device *dev, size_t size,
 474                               dma_addr_t *dma_handle, gfp_t flag)
 475
 476Identical to dma_alloc_coherent() except that the platform will
 477choose to return either consistent or non-consistent memory as it sees
 478fit.  By using this API, you are guaranteeing to the platform that you
 479have all the correct and necessary sync points for this memory in the
 480driver should it choose to return non-consistent memory.
 481
 482Note: where the platform can return consistent memory, it will
 483guarantee that the sync points become nops.
 484
 485Warning:  Handling non-consistent memory is a real pain.  You should
 486only ever use this API if you positively know your driver will be
 487required to work on one of the rare (usually non-PCI) architectures
 488that simply cannot make consistent memory.
 489
 490void
 491dma_free_noncoherent(struct device *dev, size_t size, void *cpu_addr,
 492                              dma_addr_t dma_handle)
 493
 494Free memory allocated by the nonconsistent API.  All parameters must
 495be identical to those passed in (and returned by
 496dma_alloc_noncoherent()).
 497
 498int
 499dma_is_consistent(struct device *dev, dma_addr_t dma_handle)
 500
 501Returns true if the device dev is performing consistent DMA on the memory
 502area pointed to by the dma_handle.
 503
 504int
 505dma_get_cache_alignment(void)
 506
 507Returns the processor cache alignment.  This is the absolute minimum
 508alignment *and* width that you must observe when either mapping
 509memory or doing partial flushes.
 510
 511Notes: This API may return a number *larger* than the actual cache
 512line, but it will guarantee that one or more cache lines fit exactly
 513into the width returned by this call.  It will also always be a power
 514of two for easy alignment.
 515
 516void
 517dma_sync_single_range(struct device *dev, dma_addr_t dma_handle,
 518                      unsigned long offset, size_t size,
 519                      enum dma_data_direction direction)
 520
 521Does a partial sync, starting at offset and continuing for size.  You
 522must be careful to observe the cache alignment and width when doing
 523anything like this.  You must also be extra careful about accessing
 524memory you intend to sync partially.
 525
 526void
 527dma_cache_sync(struct device *dev, void *vaddr, size_t size,
 528               enum dma_data_direction direction)
 529
 530Do a partial sync of memory that was allocated by
 531dma_alloc_noncoherent(), starting at virtual address vaddr and
 532continuing on for size.  Again, you *must* observe the cache line
 533boundaries when doing this.
 534
 535int
 536dma_declare_coherent_memory(struct device *dev, dma_addr_t bus_addr,
 537                            dma_addr_t device_addr, size_t size, int
 538                            flags)
 539
 540Declare region of memory to be handed out by dma_alloc_coherent when
 541it's asked for coherent memory for this device.
 542
 543bus_addr is the physical address to which the memory is currently
 544assigned in the bus responding region (this will be used by the
 545platform to perform the mapping).
 546
 547device_addr is the physical address the device needs to be programmed
 548with actually to address this memory (this will be handed out as the
 549dma_addr_t in dma_alloc_coherent()).
 550
 551size is the size of the area (must be multiples of PAGE_SIZE).
 552
 553flags can be or'd together and are:
 554
 555DMA_MEMORY_MAP - request that the memory returned from
 556dma_alloc_coherent() be directly writable.
 557
 558DMA_MEMORY_IO - request that the memory returned from
 559dma_alloc_coherent() be addressable using read/write/memcpy_toio etc.
 560
 561One or both of these flags must be present.
 562
 563DMA_MEMORY_INCLUDES_CHILDREN - make the declared memory be allocated by
 564dma_alloc_coherent of any child devices of this one (for memory residing
 565on a bridge).
 566
 567DMA_MEMORY_EXCLUSIVE - only allocate memory from the declared regions. 
 568Do not allow dma_alloc_coherent() to fall back to system memory when
 569it's out of memory in the declared region.
 570
 571The return value will be either DMA_MEMORY_MAP or DMA_MEMORY_IO and
 572must correspond to a passed in flag (i.e. no returning DMA_MEMORY_IO
 573if only DMA_MEMORY_MAP were passed in) for success or zero for
 574failure.
 575
 576Note, for DMA_MEMORY_IO returns, all subsequent memory returned by
 577dma_alloc_coherent() may no longer be accessed directly, but instead
 578must be accessed using the correct bus functions.  If your driver
 579isn't prepared to handle this contingency, it should not specify
 580DMA_MEMORY_IO in the input flags.
 581
 582As a simplification for the platforms, only *one* such region of
 583memory may be declared per device.
 584
 585For reasons of efficiency, most platforms choose to track the declared
 586region only at the granularity of a page.  For smaller allocations,
 587you should use the dma_pool() API.
 588
 589void
 590dma_release_declared_memory(struct device *dev)
 591
 592Remove the memory region previously declared from the system.  This
 593API performs *no* in-use checking for this region and will return
 594unconditionally having removed all the required structures.  It is the
 595driver's job to ensure that no parts of this memory region are
 596currently in use.
 597
 598void *
 599dma_mark_declared_memory_occupied(struct device *dev,
 600                                  dma_addr_t device_addr, size_t size)
 601
 602This is used to occupy specific regions of the declared space
 603(dma_alloc_coherent() will hand out the first free region it finds).
 604
 605device_addr is the *device* address of the region requested.
 606
 607size is the size (and should be a page-sized multiple).
 608
 609The return value will be either a pointer to the processor virtual
 610address of the memory, or an error (via PTR_ERR()) if any part of the
 611region is occupied.
 612
 613Part III - Debug drivers use of the DMA-API
 614-------------------------------------------
 615
 616The DMA-API as described above as some constraints. DMA addresses must be
 617released with the corresponding function with the same size for example. With
 618the advent of hardware IOMMUs it becomes more and more important that drivers
 619do not violate those constraints. In the worst case such a violation can
 620result in data corruption up to destroyed filesystems.
 621
 622To debug drivers and find bugs in the usage of the DMA-API checking code can
 623be compiled into the kernel which will tell the developer about those
 624violations. If your architecture supports it you can select the "Enable
 625debugging of DMA-API usage" option in your kernel configuration. Enabling this
 626option has a performance impact. Do not enable it in production kernels.
 627
 628If you boot the resulting kernel will contain code which does some bookkeeping
 629about what DMA memory was allocated for which device. If this code detects an
 630error it prints a warning message with some details into your kernel log. An
 631example warning message may look like this:
 632
 633------------[ cut here ]------------
 634WARNING: at /data2/repos/linux-2.6-iommu/lib/dma-debug.c:448
 635        check_unmap+0x203/0x490()
 636Hardware name:
 637forcedeth 0000:00:08.0: DMA-API: device driver frees DMA memory with wrong
 638        function [device address=0x00000000640444be] [size=66 bytes] [mapped as
 639single] [unmapped as page]
 640Modules linked in: nfsd exportfs bridge stp llc r8169
 641Pid: 0, comm: swapper Tainted: G        W  2.6.28-dmatest-09289-g8bb99c0 #1
 642Call Trace:
 643 <IRQ>  [<ffffffff80240b22>] warn_slowpath+0xf2/0x130
 644 [<ffffffff80647b70>] _spin_unlock+0x10/0x30
 645 [<ffffffff80537e75>] usb_hcd_link_urb_to_ep+0x75/0xc0
 646 [<ffffffff80647c22>] _spin_unlock_irqrestore+0x12/0x40
 647 [<ffffffff8055347f>] ohci_urb_enqueue+0x19f/0x7c0
 648 [<ffffffff80252f96>] queue_work+0x56/0x60
 649 [<ffffffff80237e10>] enqueue_task_fair+0x20/0x50
 650 [<ffffffff80539279>] usb_hcd_submit_urb+0x379/0xbc0
 651 [<ffffffff803b78c3>] cpumask_next_and+0x23/0x40
 652 [<ffffffff80235177>] find_busiest_group+0x207/0x8a0
 653 [<ffffffff8064784f>] _spin_lock_irqsave+0x1f/0x50
 654 [<ffffffff803c7ea3>] check_unmap+0x203/0x490
 655 [<ffffffff803c8259>] debug_dma_unmap_page+0x49/0x50
 656 [<ffffffff80485f26>] nv_tx_done_optimized+0xc6/0x2c0
 657 [<ffffffff80486c13>] nv_nic_irq_optimized+0x73/0x2b0
 658 [<ffffffff8026df84>] handle_IRQ_event+0x34/0x70
 659 [<ffffffff8026ffe9>] handle_edge_irq+0xc9/0x150
 660 [<ffffffff8020e3ab>] do_IRQ+0xcb/0x1c0
 661 [<ffffffff8020c093>] ret_from_intr+0x0/0xa
 662 <EOI> <4>---[ end trace f6435a98e2a38c0e ]---
 663
 664The driver developer can find the driver and the device including a stacktrace
 665of the DMA-API call which caused this warning.
 666
 667Per default only the first error will result in a warning message. All other
 668errors will only silently counted. This limitation exist to prevent the code
 669from flooding your kernel log. To support debugging a device driver this can
 670be disabled via debugfs. See the debugfs interface documentation below for
 671details.
 672
 673The debugfs directory for the DMA-API debugging code is called dma-api/. In
 674this directory the following files can currently be found:
 675
 676        dma-api/all_errors      This file contains a numeric value. If this
 677                                value is not equal to zero the debugging code
 678                                will print a warning for every error it finds
 679                                into the kernel log. Be carefull with this
 680                                option. It can easily flood your logs.
 681
 682        dma-api/disabled        This read-only file contains the character 'Y'
 683                                if the debugging code is disabled. This can
 684                                happen when it runs out of memory or if it was
 685                                disabled at boot time
 686
 687        dma-api/error_count     This file is read-only and shows the total
 688                                numbers of errors found.
 689
 690        dma-api/num_errors      The number in this file shows how many
 691                                warnings will be printed to the kernel log
 692                                before it stops. This number is initialized to
 693                                one at system boot and be set by writing into
 694                                this file
 695
 696        dma-api/min_free_entries
 697                                This read-only file can be read to get the
 698                                minimum number of free dma_debug_entries the
 699                                allocator has ever seen. If this value goes
 700                                down to zero the code will disable itself
 701                                because it is not longer reliable.
 702
 703        dma-api/num_free_entries
 704                                The current number of free dma_debug_entries
 705                                in the allocator.
 706
 707If you have this code compiled into your kernel it will be enabled by default.
 708If you want to boot without the bookkeeping anyway you can provide
 709'dma_debug=off' as a boot parameter. This will disable DMA-API debugging.
 710Notice that you can not enable it again at runtime. You have to reboot to do
 711so.
 712
 713When the code disables itself at runtime this is most likely because it ran
 714out of dma_debug_entries. These entries are preallocated at boot. The number
 715of preallocated entries is defined per architecture. If it is too low for you
 716boot with 'dma_debug_entries=<your_desired_number>' to overwrite the
 717architectural default.
 718
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.