1                     Dynamic DMA mapping Guide
   2                     =========================
   4                 David S. Miller <>
   5                 Richard Henderson <>
   6                  Jakub Jelinek <>
   8This is a guide to device driver writers on how to use the DMA API
   9with example pseudo-code.  For a concise description of the API, see
  12Most of the 64bit platforms have special hardware that translates bus
  13addresses (DMA addresses) into physical addresses.  This is similar to
  14how page tables and/or a TLB translates virtual addresses to physical
  15addresses on a CPU.  This is needed so that e.g. PCI devices can
  16access with a Single Address Cycle (32bit DMA address) any page in the
  1764bit physical address space.  Previously in Linux those 64bit
  18platforms had to set artificial limits on the maximum RAM size in the
  19system, so that the virt_to_bus() static scheme works (the DMA address
  20translation tables were simply filled on bootup to map each bus
  21address to the physical page __pa(bus_to_virt())).
  23So that Linux can use the dynamic DMA mapping, it needs some help from the
  24drivers, namely it has to take into account that DMA addresses should be
  25mapped only for the time they are actually used and unmapped after the DMA
  28The following API will work of course even on platforms where no such
  29hardware exists.
  31Note that the DMA API works with any bus independent of the underlying
  32microprocessor architecture. You should use the DMA API rather than
  33the bus specific DMA API (e.g. pci_dma_*).
  35First of all, you should make sure
  37#include <linux/dma-mapping.h>
  39is in your driver. This file will obtain for you the definition of the
  40dma_addr_t (which can hold any valid DMA address for the platform)
  41type which should be used everywhere you hold a DMA (bus) address
  42returned from the DMA mapping functions.
  44                         What memory is DMA'able?
  46The first piece of information you must know is what kernel memory can
  47be used with the DMA mapping facilities.  There has been an unwritten
  48set of rules regarding this, and this text is an attempt to finally
  49write them down.
  51If you acquired your memory via the page allocator
  52(i.e. __get_free_page*()) or the generic memory allocators
  53(i.e. kmalloc() or kmem_cache_alloc()) then you may DMA to/from
  54that memory using the addresses returned from those routines.
  56This means specifically that you may _not_ use the memory/addresses
  57returned from vmalloc() for DMA.  It is possible to DMA to the
  58_underlying_ memory mapped into a vmalloc() area, but this requires
  59walking page tables to get the physical addresses, and then
  60translating each of those pages back to a kernel address using
  61something like __va().  [ EDIT: Update this when we integrate
  62Gerd Knorr's generic code which does this. ]
  64This rule also means that you may use neither kernel image addresses
  65(items in data/text/bss segments), nor module image addresses, nor
  66stack addresses for DMA.  These could all be mapped somewhere entirely
  67different than the rest of physical memory.  Even if those classes of
  68memory could physically work with DMA, you'd need to ensure the I/O
  69buffers were cacheline-aligned.  Without that, you'd see cacheline
  70sharing problems (data corruption) on CPUs with DMA-incoherent caches.
  71(The CPU could write to one word, DMA would write to a different one
  72in the same cache line, and one of them could be overwritten.)
  74Also, this means that you cannot take the return of a kmap()
  75call and DMA to/from that.  This is similar to vmalloc().
  77What about block I/O and networking buffers?  The block I/O and
  78networking subsystems make sure that the buffers they use are valid
  79for you to DMA from/to.
  81                        DMA addressing limitations
  83Does your device have any DMA addressing limitations?  For example, is
  84your device only capable of driving the low order 24-bits of address?
  85If so, you need to inform the kernel of this fact.
  87By default, the kernel assumes that your device can address the full
  8832-bits.  For a 64-bit capable device, this needs to be increased.
  89And for a device with limitations, as discussed in the previous
  90paragraph, it needs to be decreased.
  92Special note about PCI: PCI-X specification requires PCI-X devices to
  93support 64-bit addressing (DAC) for all transactions.  And at least
  94one platform (SGI SN2) requires 64-bit consistent allocations to
  95operate correctly when the IO bus is in PCI-X mode.
  97For correct operation, you must interrogate the kernel in your device
  98probe routine to see if the DMA controller on the machine can properly
  99support the DMA addressing limitation your device has.  It is good
 100style to do this even if your device holds the default setting,
 101because this shows that you did think about these issues wrt. your
 104The query is performed via a call to dma_set_mask():
 106        int dma_set_mask(struct device *dev, u64 mask);
 108The query for consistent allocations is performed via a call to
 111        int dma_set_coherent_mask(struct device *dev, u64 mask);
 113Here, dev is a pointer to the device struct of your device, and mask
 114is a bit mask describing which bits of an address your device
 115supports.  It returns zero if your card can perform DMA properly on
 116the machine given the address mask you provided.  In general, the
 117device struct of your device is embedded in the bus specific device
 118struct of your device.  For example, a pointer to the device struct of
 119your PCI device is pdev->dev (pdev is a pointer to the PCI device
 120struct of your device).
 122If it returns non-zero, your device cannot perform DMA properly on
 123this platform, and attempting to do so will result in undefined
 124behavior.  You must either use a different mask, or not use DMA.
 126This means that in the failure case, you have three options:
 1281) Use another DMA mask, if possible (see below).
 1292) Use some non-DMA mode for data transfer, if possible.
 1303) Ignore this device and do not initialize it.
 132It is recommended that your driver print a kernel KERN_WARNING message
 133when you end up performing either #2 or #3.  In this manner, if a user
 134of your driver reports that performance is bad or that the device is not
 135even detected, you can ask them for the kernel messages to find out
 136exactly why.
 138The standard 32-bit addressing device would do something like this:
 140        if (dma_set_mask(dev, DMA_BIT_MASK(32))) {
 141                printk(KERN_WARNING
 142                       "mydev: No suitable DMA available.\n");
 143                goto ignore_this_device;
 144        }
 146Another common scenario is a 64-bit capable device.  The approach here
 147is to try for 64-bit addressing, but back down to a 32-bit mask that
 148should not fail.  The kernel may fail the 64-bit mask not because the
 149platform is not capable of 64-bit addressing.  Rather, it may fail in
 150this case simply because 32-bit addressing is done more efficiently
 151than 64-bit addressing.  For example, Sparc64 PCI SAC addressing is
 152more efficient than DAC addressing.
 154Here is how you would handle a 64-bit capable device which can drive
 155all 64-bits when accessing streaming DMA:
 157        int using_dac;
 159        if (!dma_set_mask(dev, DMA_BIT_MASK(64))) {
 160                using_dac = 1;
 161        } else if (!dma_set_mask(dev, DMA_BIT_MASK(32))) {
 162                using_dac = 0;
 163        } else {
 164                printk(KERN_WARNING
 165                       "mydev: No suitable DMA available.\n");
 166                goto ignore_this_device;
 167        }
 169If a card is capable of using 64-bit consistent allocations as well,
 170the case would look like this:
 172        int using_dac, consistent_using_dac;
 174        if (!dma_set_mask(dev, DMA_BIT_MASK(64))) {
 175                using_dac = 1;
 176                consistent_using_dac = 1;
 177                dma_set_coherent_mask(dev, DMA_BIT_MASK(64));
 178        } else if (!dma_set_mask(dev, DMA_BIT_MASK(32))) {
 179                using_dac = 0;
 180                consistent_using_dac = 0;
 181                dma_set_coherent_mask(dev, DMA_BIT_MASK(32));
 182        } else {
 183                printk(KERN_WARNING
 184                       "mydev: No suitable DMA available.\n");
 185                goto ignore_this_device;
 186        }
 188dma_set_coherent_mask() will always be able to set the same or a
 189smaller mask as dma_set_mask(). However for the rare case that a
 190device driver only uses consistent allocations, one would have to
 191check the return value from dma_set_coherent_mask().
 193Finally, if your device can only drive the low 24-bits of
 194address you might do something like:
 196        if (dma_set_mask(dev, DMA_BIT_MASK(24))) {
 197                printk(KERN_WARNING
 198                       "mydev: 24-bit DMA addressing not available.\n");
 199                goto ignore_this_device;
 200        }
 202When dma_set_mask() is successful, and returns zero, the kernel saves
 203away this mask you have provided.  The kernel will use this
 204information later when you make DMA mappings.
 206There is a case which we are aware of at this time, which is worth
 207mentioning in this documentation.  If your device supports multiple
 208functions (for example a sound card provides playback and record
 209functions) and the various different functions have _different_
 210DMA addressing limitations, you may wish to probe each mask and
 211only provide the functionality which the machine can handle.  It
 212is important that the last call to dma_set_mask() be for the
 213most specific mask.
 215Here is pseudo-code showing how this might be done:
 217        #define PLAYBACK_ADDRESS_BITS   DMA_BIT_MASK(32)
 218        #define RECORD_ADDRESS_BITS     DMA_BIT_MASK(24)
 220        struct my_sound_card *card;
 221        struct device *dev;
 223        ...
 224        if (!dma_set_mask(dev, PLAYBACK_ADDRESS_BITS)) {
 225                card->playback_enabled = 1;
 226        } else {
 227                card->playback_enabled = 0;
 228                printk(KERN_WARNING "%s: Playback disabled due to DMA limitations.\n",
 229                       card->name);
 230        }
 231        if (!dma_set_mask(dev, RECORD_ADDRESS_BITS)) {
 232                card->record_enabled = 1;
 233        } else {
 234                card->record_enabled = 0;
 235                printk(KERN_WARNING "%s: Record disabled due to DMA limitations.\n",
 236                       card->name);
 237        }
 239A sound card was used as an example here because this genre of PCI
 240devices seems to be littered with ISA chips given a PCI front end,
 241and thus retaining the 16MB DMA addressing limitations of ISA.
 243                        Types of DMA mappings
 245There are two types of DMA mappings:
 247- Consistent DMA mappings which are usually mapped at driver
 248  initialization, unmapped at the end and for which the hardware should
 249  guarantee that the device and the CPU can access the data
 250  in parallel and will see updates made by each other without any
 251  explicit software flushing.
 253  Think of "consistent" as "synchronous" or "coherent".
 255  The current default is to return consistent memory in the low 32
 256  bits of the bus space.  However, for future compatibility you should
 257  set the consistent mask even if this default is fine for your
 258  driver.
 260  Good examples of what to use consistent mappings for are:
 262        - Network card DMA ring descriptors.
 263        - SCSI adapter mailbox command data structures.
 264        - Device firmware microcode executed out of
 265          main memory.
 267  The invariant these examples all require is that any CPU store
 268  to memory is immediately visible to the device, and vice
 269  versa.  Consistent mappings guarantee this.
 271  IMPORTANT: Consistent DMA memory does not preclude the usage of
 272             proper memory barriers.  The CPU may reorder stores to
 273             consistent memory just as it may normal memory.  Example:
 274             if it is important for the device to see the first word
 275             of a descriptor updated before the second, you must do
 276             something like:
 278                desc->word0 = address;
 279                wmb();
 280                desc->word1 = DESC_VALID;
 282             in order to get correct behavior on all platforms.
 284             Also, on some platforms your driver may need to flush CPU write
 285             buffers in much the same way as it needs to flush write buffers
 286             found in PCI bridges (such as by reading a register's value
 287             after writing it).
 289- Streaming DMA mappings which are usually mapped for one DMA
 290  transfer, unmapped right after it (unless you use dma_sync_* below)
 291  and for which hardware can optimize for sequential accesses.
 293  This of "streaming" as "asynchronous" or "outside the coherency
 294  domain".
 296  Good examples of what to use streaming mappings for are:
 298        - Networking buffers transmitted/received by a device.
 299        - Filesystem buffers written/read by a SCSI device.
 301  The interfaces for using this type of mapping were designed in
 302  such a way that an implementation can make whatever performance
 303  optimizations the hardware allows.  To this end, when using
 304  such mappings you must be explicit about what you want to happen.
 306Neither type of DMA mapping has alignment restrictions that come from
 307the underlying bus, although some devices may have such restrictions.
 308Also, systems with caches that aren't DMA-coherent will work better
 309when the underlying buffers don't share cache lines with other data.
 312                 Using Consistent DMA mappings.
 314To allocate and map large (PAGE_SIZE or so) consistent DMA regions,
 315you should do:
 317        dma_addr_t dma_handle;
 319        cpu_addr = dma_alloc_coherent(dev, size, &dma_handle, gfp);
 321where device is a struct device *. This may be called in interrupt
 322context with the GFP_ATOMIC flag.
 324Size is the length of the region you want to allocate, in bytes.
 326This routine will allocate RAM for that region, so it acts similarly to
 327__get_free_pages (but takes size instead of a page order).  If your
 328driver needs regions sized smaller than a page, you may prefer using
 329the dma_pool interface, described below.
 331The consistent DMA mapping interfaces, for non-NULL dev, will by
 332default return a DMA address which is 32-bit addressable.  Even if the
 333device indicates (via DMA mask) that it may address the upper 32-bits,
 334consistent allocation will only return > 32-bit addresses for DMA if
 335the consistent DMA mask has been explicitly changed via
 336dma_set_coherent_mask().  This is true of the dma_pool interface as
 339dma_alloc_coherent returns two values: the virtual address which you
 340can use to access it from the CPU and dma_handle which you pass to the
 343The cpu return address and the DMA bus master address are both
 344guaranteed to be aligned to the smallest PAGE_SIZE order which
 345is greater than or equal to the requested size.  This invariant
 346exists (for example) to guarantee that if you allocate a chunk
 347which is smaller than or equal to 64 kilobytes, the extent of the
 348buffer you receive will not cross a 64K boundary.
 350To unmap and free such a DMA region, you call:
 352        dma_free_coherent(dev, size, cpu_addr, dma_handle);
 354where dev, size are the same as in the above call and cpu_addr and
 355dma_handle are the values dma_alloc_coherent returned to you.
 356This function may not be called in interrupt context.
 358If your driver needs lots of smaller memory regions, you can write
 359custom code to subdivide pages returned by dma_alloc_coherent,
 360or you can use the dma_pool API to do that.  A dma_pool is like
 361a kmem_cache, but it uses dma_alloc_coherent not __get_free_pages.
 362Also, it understands common hardware constraints for alignment,
 363like queue heads needing to be aligned on N byte boundaries.
 365Create a dma_pool like this:
 367        struct dma_pool *pool;
 369        pool = dma_pool_create(name, dev, size, align, alloc);
 371The "name" is for diagnostics (like a kmem_cache name); dev and size
 372are as above.  The device's hardware alignment requirement for this
 373type of data is "align" (which is expressed in bytes, and must be a
 374power of two).  If your device has no boundary crossing restrictions,
 375pass 0 for alloc; passing 4096 says memory allocated from this pool
 376must not cross 4KByte boundaries (but at that time it may be better to
 377go for dma_alloc_coherent directly instead).
 379Allocate memory from a dma pool like this:
 381        cpu_addr = dma_pool_alloc(pool, flags, &dma_handle);
 383flags are SLAB_KERNEL if blocking is permitted (not in_interrupt nor
 384holding SMP locks), SLAB_ATOMIC otherwise.  Like dma_alloc_coherent,
 385this returns two values, cpu_addr and dma_handle.
 387Free memory that was allocated from a dma_pool like this:
 389        dma_pool_free(pool, cpu_addr, dma_handle);
 391where pool is what you passed to dma_pool_alloc, and cpu_addr and
 392dma_handle are the values dma_pool_alloc returned. This function
 393may be called in interrupt context.
 395Destroy a dma_pool by calling:
 397        dma_pool_destroy(pool);
 399Make sure you've called dma_pool_free for all memory allocated
 400from a pool before you destroy the pool. This function may not
 401be called in interrupt context.
 403                        DMA Direction
 405The interfaces described in subsequent portions of this document
 406take a DMA direction argument, which is an integer and takes on
 407one of the following values:
 414One should provide the exact DMA direction if you know it.
 416DMA_TO_DEVICE means "from main memory to the device"
 417DMA_FROM_DEVICE means "from the device to main memory"
 418It is the direction in which the data moves during the DMA
 421You are _strongly_ encouraged to specify this as precisely
 422as you possibly can.
 424If you absolutely cannot know the direction of the DMA transfer,
 425specify DMA_BIDIRECTIONAL.  It means that the DMA can go in
 426either direction.  The platform guarantees that you may legally
 427specify this, and that it will work, but this may be at the
 428cost of performance for example.
 430The value DMA_NONE is to be used for debugging.  One can
 431hold this in a data structure before you come to know the
 432precise direction, and this will help catch cases where your
 433direction tracking logic has failed to set things up properly.
 435Another advantage of specifying this value precisely (outside of
 436potential platform-specific optimizations of such) is for debugging.
 437Some platforms actually have a write permission boolean which DMA
 438mappings can be marked with, much like page protections in the user
 439program address space.  Such platforms can and do report errors in the
 440kernel logs when the DMA controller hardware detects violation of the
 441permission setting.
 443Only streaming mappings specify a direction, consistent mappings
 444implicitly have a direction attribute setting of
 447The SCSI subsystem tells you the direction to use in the
 448'sc_data_direction' member of the SCSI command your driver is
 449working on.
 451For Networking drivers, it's a rather simple affair.  For transmit
 452packets, map/unmap them with the DMA_TO_DEVICE direction
 453specifier.  For receive packets, just the opposite, map/unmap them
 454with the DMA_FROM_DEVICE direction specifier.
 456                  Using Streaming DMA mappings
 458The streaming DMA mapping routines can be called from interrupt
 459context.  There are two versions of each map/unmap, one which will
 460map/unmap a single memory region, and one which will map/unmap a
 463To map a single region, you do:
 465        struct device *dev = &my_dev->dev;
 466        dma_addr_t dma_handle;
 467        void *addr = buffer->ptr;
 468        size_t size = buffer->len;
 470        dma_handle = dma_map_single(dev, addr, size, direction);
 471        if (dma_mapping_error(dma_handle)) {
 472                /*
 473                 * reduce current DMA mapping usage,
 474                 * delay and try again later or
 475                 * reset driver.
 476                 */
 477                goto map_error_handling;
 478        }
 480and to unmap it:
 482        dma_unmap_single(dev, dma_handle, size, direction);
 484You should call dma_mapping_error() as dma_map_single() could fail and return
 485error. Not all dma implementations support dma_mapping_error() interface.
 486However, it is a good practice to call dma_mapping_error() interface, which
 487will invoke the generic mapping error check interface. Doing so will ensure
 488that the mapping code will work correctly on all dma implementations without
 489any dependency on the specifics of the underlying implementation. Using the
 490returned address without checking for errors could result in failures ranging
 491from panics to silent data corruption. Couple of example of incorrect ways to
 492check for errors that make assumptions about the underlying dma implementation
 493are as follows and these are applicable to dma_map_page() as well.
 495Incorrect example 1:
 496        dma_addr_t dma_handle;
 498        dma_handle = dma_map_single(dev, addr, size, direction);
 499        if ((dma_handle & 0xffff != 0) || (dma_handle >= 0x1000000)) {
 500                goto map_error;
 501        }
 503Incorrect example 2:
 504        dma_addr_t dma_handle;
 506        dma_handle = dma_map_single(dev, addr, size, direction);
 507        if (dma_handle == DMA_ERROR_CODE) {
 508                goto map_error;
 509        }
 511You should call dma_unmap_single when the DMA activity is finished, e.g.
 512from the interrupt which told you that the DMA transfer is done.
 514Using cpu pointers like this for single mappings has a disadvantage,
 515you cannot reference HIGHMEM memory in this way.  Thus, there is a
 516map/unmap interface pair akin to dma_{map,unmap}_single.  These
 517interfaces deal with page/offset pairs instead of cpu pointers.
 520        struct device *dev = &my_dev->dev;
 521        dma_addr_t dma_handle;
 522        struct page *page = buffer->page;
 523        unsigned long offset = buffer->offset;
 524        size_t size = buffer->len;
 526        dma_handle = dma_map_page(dev, page, offset, size, direction);
 527        if (dma_mapping_error(dma_handle)) {
 528                /*
 529                 * reduce current DMA mapping usage,
 530                 * delay and try again later or
 531                 * reset driver.
 532                 */
 533                goto map_error_handling;
 534        }
 536        ...
 538        dma_unmap_page(dev, dma_handle, size, direction);
 540Here, "offset" means byte offset within the given page.
 542You should call dma_mapping_error() as dma_map_page() could fail and return
 543error as outlined under the dma_map_single() discussion.
 545You should call dma_unmap_page when the DMA activity is finished, e.g.
 546from the interrupt which told you that the DMA transfer is done.
 548With scatterlists, you map a region gathered from several regions by:
 550        int i, count = dma_map_sg(dev, sglist, nents, direction);
 551        struct scatterlist *sg;
 553        for_each_sg(sglist, sg, count, i) {
 554                hw_address[i] = sg_dma_address(sg);
 555                hw_len[i] = sg_dma_len(sg);
 556        }
 558where nents is the number of entries in the sglist.
 560The implementation is free to merge several consecutive sglist entries
 561into one (e.g. if DMA mapping is done with PAGE_SIZE granularity, any
 562consecutive sglist entries can be merged into one provided the first one
 563ends and the second one starts on a page boundary - in fact this is a huge
 564advantage for cards which either cannot do scatter-gather or have very
 565limited number of scatter-gather entries) and returns the actual number
 566of sg entries it mapped them to. On failure 0 is returned.
 568Then you should loop count times (note: this can be less than nents times)
 569and use sg_dma_address() and sg_dma_len() macros where you previously
 570accessed sg->address and sg->length as shown above.
 572To unmap a scatterlist, just call:
 574        dma_unmap_sg(dev, sglist, nents, direction);
 576Again, make sure DMA activity has already finished.
 578PLEASE NOTE:  The 'nents' argument to the dma_unmap_sg call must be
 579              the _same_ one you passed into the dma_map_sg call,
 580              it should _NOT_ be the 'count' value _returned_ from the
 581              dma_map_sg call.
 583Every dma_map_{single,sg} call should have its dma_unmap_{single,sg}
 584counterpart, because the bus address space is a shared resource (although
 585in some ports the mapping is per each BUS so less devices contend for the
 586same bus address space) and you could render the machine unusable by eating
 587all bus addresses.
 589If you need to use the same streaming DMA region multiple times and touch
 590the data in between the DMA transfers, the buffer needs to be synced
 591properly in order for the cpu and device to see the most uptodate and
 592correct copy of the DMA buffer.
 594So, firstly, just map it with dma_map_{single,sg}, and after each DMA
 595transfer call either:
 597        dma_sync_single_for_cpu(dev, dma_handle, size, direction);
 601        dma_sync_sg_for_cpu(dev, sglist, nents, direction);
 603as appropriate.
 605Then, if you wish to let the device get at the DMA area again,
 606finish accessing the data with the cpu, and then before actually
 607giving the buffer to the hardware call either:
 609        dma_sync_single_for_device(dev, dma_handle, size, direction);
 613        dma_sync_sg_for_device(dev, sglist, nents, direction);
 615as appropriate.
 617After the last DMA transfer call one of the DMA unmap routines
 618dma_unmap_{single,sg}. If you don't touch the data from the first dma_map_*
 619call till dma_unmap_*, then you don't have to call the dma_sync_*
 620routines at all.
 622Here is pseudo code which shows a situation in which you would need
 623to use the dma_sync_*() interfaces.
 625        my_card_setup_receive_buffer(struct my_card *cp, char *buffer, int len)
 626        {
 627                dma_addr_t mapping;
 629                mapping = dma_map_single(cp->dev, buffer, len, DMA_FROM_DEVICE);
 630                if (dma_mapping_error(dma_handle)) {
 631                        /*
 632                         * reduce current DMA mapping usage,
 633                         * delay and try again later or
 634                         * reset driver.
 635                         */
 636                        goto map_error_handling;
 637                }
 639                cp->rx_buf = buffer;
 640                cp->rx_len = len;
 641                cp->rx_dma = mapping;
 643                give_rx_buf_to_card(cp);
 644        }
 646        ...
 648        my_card_interrupt_handler(int irq, void *devid, struct pt_regs *regs)
 649        {
 650                struct my_card *cp = devid;
 652                ...
 653                if (read_card_status(cp) == RX_BUF_TRANSFERRED) {
 654                        struct my_card_header *hp;
 656                        /* Examine the header to see if we wish
 657                         * to accept the data.  But synchronize
 658                         * the DMA transfer with the CPU first
 659                         * so that we see updated contents.
 660                         */
 661                        dma_sync_single_for_cpu(&cp->dev, cp->rx_dma,
 662                                                cp->rx_len,
 663                                                DMA_FROM_DEVICE);
 665                        /* Now it is safe to examine the buffer. */
 666                        hp = (struct my_card_header *) cp->rx_buf;
 667                        if (header_is_ok(hp)) {
 668                                dma_unmap_single(&cp->dev, cp->rx_dma, cp->rx_len,
 669                                                 DMA_FROM_DEVICE);
 670                                pass_to_upper_layers(cp->rx_buf);
 671                                make_and_setup_new_rx_buf(cp);
 672                        } else {
 673                                /* CPU should not write to
 674                                 * DMA_FROM_DEVICE-mapped area,
 675                                 * so dma_sync_single_for_device() is
 676                                 * not needed here. It would be required
 677                                 * for DMA_BIDIRECTIONAL mapping if
 678                                 * the memory was modified.
 679                                 */
 680                                give_rx_buf_to_card(cp);
 681                        }
 682                }
 683        }
 685Drivers converted fully to this interface should not use virt_to_bus any
 686longer, nor should they use bus_to_virt. Some drivers have to be changed a
 687little bit, because there is no longer an equivalent to bus_to_virt in the
 688dynamic DMA mapping scheme - you have to always store the DMA addresses
 689returned by the dma_alloc_coherent, dma_pool_alloc, and dma_map_single
 690calls (dma_map_sg stores them in the scatterlist itself if the platform
 691supports dynamic DMA mapping in hardware) in your driver structures and/or
 692in the card registers.
 694All drivers should be using these interfaces with no exceptions.  It
 695is planned to completely remove virt_to_bus() and bus_to_virt() as
 696they are entirely deprecated.  Some ports already do not provide these
 697as it is impossible to correctly support them.
 699                        Handling Errors
 701DMA address space is limited on some architectures and an allocation
 702failure can be determined by:
 704- checking if dma_alloc_coherent returns NULL or dma_map_sg returns 0
 706- checking the returned dma_addr_t of dma_map_single and dma_map_page
 707  by using dma_mapping_error():
 709        dma_addr_t dma_handle;
 711        dma_handle = dma_map_single(dev, addr, size, direction);
 712        if (dma_mapping_error(dev, dma_handle)) {
 713                /*
 714                 * reduce current DMA mapping usage,
 715                 * delay and try again later or
 716                 * reset driver.
 717                 */
 718                goto map_error_handling;
 719        }
 721- unmap pages that are already mapped, when mapping error occurs in the middle
 722  of a multiple page mapping attempt. These example are applicable to
 723  dma_map_page() as well.
 725Example 1:
 726        dma_addr_t dma_handle1;
 727        dma_addr_t dma_handle2;
 729        dma_handle1 = dma_map_single(dev, addr, size, direction);
 730        if (dma_mapping_error(dev, dma_handle1)) {
 731                /*
 732                 * reduce current DMA mapping usage,
 733                 * delay and try again later or
 734                 * reset driver.
 735                 */
 736                goto map_error_handling1;
 737        }
 738        dma_handle2 = dma_map_single(dev, addr, size, direction);
 739        if (dma_mapping_error(dev, dma_handle2)) {
 740                /*
 741                 * reduce current DMA mapping usage,
 742                 * delay and try again later or
 743                 * reset driver.
 744                 */
 745                goto map_error_handling2;
 746        }
 748        ...
 750        map_error_handling2:
 751                dma_unmap_single(dma_handle1);
 752        map_error_handling1:
 754Example 2: (if buffers are allocated a loop, unmap all mapped buffers when
 755            mapping error is detected in the middle)
 757        dma_addr_t dma_addr;
 758        dma_addr_t array[DMA_BUFFERS];
 759        int save_index = 0;
 761        for (i = 0; i < DMA_BUFFERS; i++) {
 763                ...
 765                dma_addr = dma_map_single(dev, addr, size, direction);
 766                if (dma_mapping_error(dev, dma_addr)) {
 767                        /*
 768                         * reduce current DMA mapping usage,
 769                         * delay and try again later or
 770                         * reset driver.
 771                         */
 772                        goto map_error_handling;
 773                }
 774                array[i].dma_addr = dma_addr;
 775                save_index++;
 776        }
 778        ...
 780        map_error_handling:
 782        for (i = 0; i < save_index; i++) {
 784                ...
 786                dma_unmap_single(array[i].dma_addr);
 787        }
 789Networking drivers must call dev_kfree_skb to free the socket buffer
 790and return NETDEV_TX_OK if the DMA mapping fails on the transmit hook
 791(ndo_start_xmit). This means that the socket buffer is just dropped in
 792the failure case.
 794SCSI drivers must return SCSI_MLQUEUE_HOST_BUSY if the DMA mapping
 795fails in the queuecommand hook. This means that the SCSI subsystem
 796passes the command to the driver again later.
 798                Optimizing Unmap State Space Consumption
 800On many platforms, dma_unmap_{single,page}() is simply a nop.
 801Therefore, keeping track of the mapping address and length is a waste
 802of space.  Instead of filling your drivers up with ifdefs and the like
 803to "work around" this (which would defeat the whole purpose of a
 804portable API) the following facilities are provided.
 806Actually, instead of describing the macros one by one, we'll
 807transform some example code.
 8091) Use DEFINE_DMA_UNMAP_{ADDR,LEN} in state saving structures.
 810   Example, before:
 812        struct ring_state {
 813                struct sk_buff *skb;
 814                dma_addr_t mapping;
 815                __u32 len;
 816        };
 818   after:
 820        struct ring_state {
 821                struct sk_buff *skb;
 822                DEFINE_DMA_UNMAP_ADDR(mapping);
 823                DEFINE_DMA_UNMAP_LEN(len);
 824        };
 8262) Use dma_unmap_{addr,len}_set to set these values.
 827   Example, before:
 829        ringp->mapping = FOO;
 830        ringp->len = BAR;
 832   after:
 834        dma_unmap_addr_set(ringp, mapping, FOO);
 835        dma_unmap_len_set(ringp, len, BAR);
 8373) Use dma_unmap_{addr,len} to access these values.
 838   Example, before:
 840        dma_unmap_single(dev, ringp->mapping, ringp->len,
 841                         DMA_FROM_DEVICE);
 843   after:
 845        dma_unmap_single(dev,
 846                         dma_unmap_addr(ringp, mapping),
 847                         dma_unmap_len(ringp, len),
 848                         DMA_FROM_DEVICE);
 850It really should be self-explanatory.  We treat the ADDR and LEN
 851separately, because it is possible for an implementation to only
 852need the address in order to perform the unmap operation.
 854                        Platform Issues
 856If you are just writing drivers for Linux and do not maintain
 857an architecture port for the kernel, you can safely skip down
 858to "Closing".
 8601) Struct scatterlist requirements.
 862   Don't invent the architecture specific struct scatterlist; just use
 863   <asm-generic/scatterlist.h>. You need to enable
 864   CONFIG_NEED_SG_DMA_LENGTH if the architecture supports IOMMUs
 865   (including software IOMMU).
 869   Architectures must ensure that kmalloc'ed buffer is
 870   DMA-safe. Drivers and subsystems depend on it. If an architecture
 871   isn't fully DMA-coherent (i.e. hardware doesn't ensure that data in
 872   the CPU cache is identical to data in main memory),
 873   ARCH_DMA_MINALIGN must be set so that the memory allocator
 874   makes sure that kmalloc'ed buffer doesn't share a cache line with
 875   the others. See arch/arm/include/asm/cache.h as an example.
 877   Note that ARCH_DMA_MINALIGN is about DMA memory alignment
 878   constraints. You don't need to worry about the architecture data
 879   alignment constraints (e.g. the alignment constraints about 64-bit
 880   objects).
 8823) Supporting multiple types of IOMMUs
 884   If your architecture needs to support multiple types of IOMMUs, you
 885   can use include/linux/asm-generic/dma-mapping-common.h. It's a
 886   library to support the DMA API with multiple types of IOMMUs. Lots
 887   of architectures (x86, powerpc, sh, alpha, ia64, microblaze and
 888   sparc) use it. Choose one to see how it can be used. If you need to
 889   support multiple types of IOMMUs in a single system, the example of
 890   x86 or powerpc helps.
 892                           Closing
 894This document, and the API itself, would not be in its current
 895form without the feedback and suggestions from numerous individuals.
 896We would like to specifically mention, in no particular order, the
 897following people:
 899        Russell King <>
 900        Leo Dagum <>
 901        Ralf Baechle <>
 902        Grant Grundler <>
 903        Jay Estabrook <>
 904        Thomas Sailer <>
 905        Andrea Arcangeli <>
 906        Jens Axboe <>
 907        David Mosberger-Tang <>
 908 kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.