1 .\" Copyright (c) 2002, 2003 Hiten M. Pandya.
2 .\" All rights reserved.
4 .\" Redistribution and use in source and binary forms, with or without
5 .\" modification, are permitted provided that the following conditions
7 .\" 1. Redistributions of source code must retain the above copyright
8 .\" notice, this list of conditions, and the following disclaimer,
9 .\" without modification, immediately at the beginning of the file.
10 .\" 2. The name of the author may not be used to endorse or promote products
11 .\" derived from this software without specific prior written permission.
13 .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR, CONTRIBUTORS OR THE
17 .\" VOICES IN HITEN PANDYA'S HEAD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
18 .\" SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
19 .\" TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
20 .\" PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
21 .\" LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
22 .\" NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
23 .\" SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 .\" Copyright (c) 1996, 1997, 1998, 2001 The NetBSD Foundation, Inc.
26 .\" All rights reserved.
28 .\" This code is derived from software contributed to The NetBSD Foundation
29 .\" by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
30 .\" NASA Ames Research Center.
32 .\" Redistribution and use in source and binary forms, with or without
33 .\" modification, are permitted provided that the following conditions
35 .\" 1. Redistributions of source code must retain the above copyright
36 .\" notice, this list of conditions and the following disclaimer.
37 .\" 2. Redistributions in binary form must reproduce the above copyright
38 .\" notice, this list of conditions and the following disclaimer in the
39 .\" documentation and/or other materials provided with the distribution.
40 .\" 3. All advertising materials mentioning features or use of this software
41 .\" must display the following acknowledgment:
42 .\" This product includes software developed by the NetBSD
43 .\" Foundation, Inc. and its contributors.
44 .\" 4. Neither the name of The NetBSD Foundation nor the names of its
45 .\" contributors may be used to endorse or promote products derived
46 .\" from this software without specific prior written permission.
48 .\" THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
49 .\" ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
50 .\" TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
51 .\" PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
52 .\" BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
53 .\" CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
54 .\" SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
55 .\" INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
56 .\" CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
57 .\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
58 .\" POSSIBILITY OF SUCH DAMAGE.
61 .\" $NetBSD: bus_dma.9,v 1.25 2002/10/14 13:43:16 wiz Exp $
68 .Nm bus_dma_tag_create ,
69 .Nm bus_dma_tag_destroy ,
70 .Nm bus_dmamap_create ,
71 .Nm bus_dmamap_destroy ,
73 .Nm bus_dmamap_load_mbuf ,
74 .Nm bus_dmamap_load_mbuf_sg ,
75 .Nm bus_dmamap_load_uio ,
76 .Nm bus_dmamap_unload ,
78 .Nm bus_dmamem_alloc ,
80 .Nd Bus and Machine Independent DMA Mapping Interface
84 .Fn bus_dma_tag_create "bus_dma_tag_t parent" "bus_size_t alignment" \
85 "bus_size_t boundary" "bus_addr_t lowaddr" "bus_addr_t highaddr" \
86 "bus_dma_filter_t *filtfunc" "void *filtfuncarg" "bus_size_t maxsize" \
87 "int nsegments" "bus_size_t maxsegsz" "int flags" "bus_dma_lock_t *lockfunc" \
88 "void *lockfuncarg" "bus_dma_tag_t *dmat"
90 .Fn bus_dma_tag_destroy "bus_dma_tag_t dmat"
92 .Fn bus_dmamap_create "bus_dma_tag_t dmat" "int flags" "bus_dmamap_t *mapp"
94 .Fn bus_dmamap_destroy "bus_dma_tag_t dmat" "bus_dmamap_t map"
96 .Fn bus_dmamap_load "bus_dma_tag_t dmat" "bus_dmamap_t map" "void *buf" \
97 "bus_size_t buflen" "bus_dmamap_callback_t *callback" "void *callback_arg" \
100 .Fn bus_dmamap_load_mbuf "bus_dma_tag_t dmat" "bus_dmamap_t map" \
101 "struct mbuf *mbuf" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
104 .Fn bus_dmamap_load_mbuf_sg "bus_dma_tag_t dmat" "bus_dmamap_t map" \
105 "struct mbuf *mbuf" "bus_dma_segment_t *segs" "int *nsegs" "int flags"
107 .Fn bus_dmamap_load_uio "bus_dma_tag_t dmat" "bus_dmamap_t map" \
108 "struct uio *uio" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
111 .Fn bus_dmamap_unload "bus_dma_tag_t dmat" "bus_dmamap_t map"
113 .Fn bus_dmamap_sync "bus_dma_tag_t dmat" "bus_dmamap_t map" \
116 .Fn bus_dmamem_alloc "bus_dma_tag_t dmat" "void **vaddr" \
117 "int flags" "bus_dmamap_t *mapp"
119 .Fn bus_dmamem_free "bus_dma_tag_t dmat" "void *vaddr" \
122 Direct Memory Access (DMA) is a method of transferring data
123 without involving the CPU, thus providing higher performance.
124 A DMA transaction can be achieved between device to memory,
125 device to device, or memory to memory.
129 API is a bus, device, and machine-independent (MI) interface to
131 It provides the client with flexibility and simplicity by
132 abstracting machine dependent issues like setting up
133 DMA mappings, handling cache issues, bus specific features
135 .Sh STRUCTURES AND TYPES
136 .Bl -tag -width indent
138 A machine-dependent (MD) opaque type that describes the
139 characteristics of DMA transactions.
140 DMA tags are organized into a hierarchy, with each child
141 tag inheriting the restrictions of its parent.
142 This allows all devices along the path of DMA transactions
143 to contribute to the constraints of those transactions.
144 .It Vt bus_dma_filter_t
145 Client specified address filter having the format:
146 .Bl -tag -width indent
148 .Fn "client_filter" "void *filtarg" "bus_addr_t testaddr"
151 Address filters can be specified during tag creation to allow
152 for devices whose DMA address restrictions cannot be specified
156 argument is specified by the client during tag creation to be passed to all
157 invocations of the callback.
160 argument contains a potential starting address of a DMA mapping.
161 The filter function operates on the set of addresses from
164 .Ql trunc_page(testaddr) + PAGE_SIZE - 1 ,
166 The filter function should return zero if any mapping in this range
167 can be accommodated by the device and non-zero otherwise.
168 .It Vt bus_dma_segment_t
169 A machine-dependent type that describes individual
171 It contains the following fields:
179 field contains the device visible address of the DMA segment, and
181 contains the length of the DMA segment.
182 Although the DMA segments returned by a mapping call will adhere to
183 all restrictions necessary for a successful DMA operation, some conversion
184 (e.g.\& a conversion from host byte order to the device's byte order) is
185 almost always required when presenting segment information to the device.
187 A machine-dependent opaque type describing an individual mapping.
188 One map is used for each memory allocation that will be loaded.
189 Maps can be reused once they have been unloaded.
190 Multiple maps can be associated with one DMA tag.
191 While the value of the map may evaluate to
193 on some platforms under certain conditions,
194 it should never be assumed that it will be
197 .It Vt bus_dmamap_callback_t
198 Client specified callback for receiving mapping information resulting from
202 .Fn bus_dmamap_load .
203 Callbacks are of the format:
204 .Bl -tag -width indent
206 .Fn "client_callback" "void *callback_arg" "bus_dma_segment_t *segs" \
207 "int nseg" "int error"
212 is the callback argument passed to dmamap load functions.
217 arguments describe an array of
218 .Vt bus_dma_segment_t
219 structures that represent the mapping.
220 This array is only valid within the scope of the callback function.
221 The success or failure of the mapping is indicated by the
224 More information on the use of callbacks can be found in the
225 description of the individual dmamap load functions.
226 .It Vt bus_dmamap_callback2_t
227 Client specified callback for receiving mapping information resulting from
231 .Fn bus_dmamap_load_uio
233 .Fn bus_dmamap_load_mbuf .
235 Callback2s are of the format:
236 .Bl -tag -width indent
238 .Fn "client_callback2" "void *callback_arg" "bus_dma_segment_t *segs" \
239 "int nseg" "bus_size_t mapsize" "int error"
242 Callback2's behavior is the same as
243 .Vt bus_dmamap_callback_t
244 with the addition that the length of the data mapped is provided via
246 .It Vt bus_dmasync_op_t
247 Memory synchronization operation specifier.
248 Bus DMA requires explicit synchronization of memory with its device
249 visible mapping in order to guarantee memory coherency.
252 allows the type of DMA operation that will be or has been performed
253 to be communicated to the system so that the correct coherency measures
255 The operations are represented as bitfield flags that can be combined together,
256 though it only makes sense to combine PRE flags or POST flags, not both.
259 description below for more details on how to use these operations.
261 All operations specified below are performed from the host memory point of view,
262 where a read implies data coming from the device to the host memory, and a write
263 implies data going from the host memory to the device.
264 Alternatively, the operations can be thought of in terms of driver operations,
265 where reading a network packet or storage sector corresponds to a read operation
268 .Bl -tag -width ".Dv BUS_DMASYNC_POSTWRITE"
269 .It Dv BUS_DMASYNC_PREREAD
270 Perform any synchronization required prior to an update of host memory by the
272 .It Dv BUS_DMASYNC_PREWRITE
273 Perform any synchronization required after an update of host memory by the CPU
274 and prior to device access to host memory.
275 .It Dv BUS_DMASYNC_POSTREAD
276 Perform any synchronization required after an update of host memory by the
277 device and prior to CPU access to host memory.
278 .It Dv BUS_DMASYNC_POSTWRITE
279 Perform any synchronization required after device access to host memory.
281 .It Vt bus_dma_lock_t
282 Client specified lock/mutex manipulation method.
283 This will be called from
284 within busdma whenever a client lock needs to be manipulated.
285 In its current form, the function will be called immediately before
286 the callback for a DMA load operation that has been deferred with
288 and immediately after with
290 If the load operation does not need to be deferred, then it
291 will not be called since the function loading the map should
292 be holding the appropriate locks.
293 This method is of the format:
294 .Bl -tag -width indent
296 .Fn "lockfunc" "void *lockfunc_arg" "bus_dma_lock_op_t op"
301 argument is specified by the client during tag creation to be passed to all
302 invocations of the callback.
305 argument specifies the lock operation to perform.
309 implementations are provided for convenience.
310 .Fn busdma_lock_mutex
311 performs standard mutex operations on the sleep mutex provided via
314 will generate a system panic if it is called.
315 It is substituted into the tag when
320 .Fn bus_dma_tag_create
321 and is useful for tags that should not be used with deferred load operations.
322 .It Vt bus_dma_lock_op_t
323 Operations to be performed by the client-specified
325 .Bl -tag -width ".Dv BUS_DMA_UNLOCK"
327 Acquires and/or locks the client locking primitive.
328 .It Dv BUS_DMA_UNLOCK
329 Releases and/or unlocks the client locking primitive.
333 .Bl -tag -width indent
334 .It Fn bus_dma_tag_create "parent" "alignment" "boundary" "lowaddr" \
335 "highaddr" "*filtfunc" "*filtfuncarg" "maxsize" "nsegments" "maxsegsz" \
336 "flags" "lockfunc" "lockfuncarg" "*dmat"
337 Allocates a device specific DMA tag, and initializes it according to
338 the arguments provided:
339 .Bl -tag -width ".Fa filtfuncarg"
341 Indicates restrictions between the parent bridge, CPU memory, and the
343 Each device must use a master parent tag by calling
344 .Fn bus_get_dma_tag .
346 Alignment constraint, in bytes, of any mappings created using this tag.
347 The alignment must be a power of 2.
348 Hardware that can DMA starting at any address would specify
351 Hardware requiring DMA transfers to start on a multiple of 4K
355 Boundary constraint, in bytes, of the target DMA memory region.
356 The boundary indicates the set of addresses, all multiples of the
357 boundary argument, that cannot be crossed by a single
358 .Vt bus_dma_segment_t .
359 The boundary must be a power of 2 and must be no smaller than the
360 maximum segment size.
362 indicates that there are no boundary restrictions.
363 .It Fa lowaddr , highaddr
364 Bounds of the window of bus address space that
366 be directly accessed by the device.
367 The window contains all addresses greater than
369 and less than or equal to
371 For example, a device incapable of DMA above 4GB, would specify a
374 .Dv BUS_SPACE_MAXADDR
378 .Dv BUS_SPACE_MAXADDR_32BIT .
379 Similarly a device that can only perform DMA to addresses below
383 .Dv BUS_SPACE_MAXADDR
387 .Dv BUS_SPACE_MAXADDR_24BIT .
388 Some implementations requires that some region of device visible
389 address space, overlapping available host memory, be outside the
393 is used to bounce requests that would otherwise conflict with
394 the exclusion window.
396 Optional filter function (may be
398 to be called for any attempt to
399 map memory into the window described by
403 A filter function is only required when the single window described
408 cannot adequately describe the constraints of the device.
409 The filter function will be called for every machine page
410 that overlaps the exclusion window.
412 Argument passed to all calls to the filter function for this tag.
416 Maximum size, in bytes, of the sum of all segment lengths in a given
417 DMA mapping associated with this tag.
419 Number of discontinuities (scatter/gather segments) allowed
420 in a DMA mapped region.
421 If there is no restriction,
422 .Dv BUS_SPACE_UNRESTRICTED
425 Maximum size, in bytes, of a segment in any DMA mapped region associated
430 .Bl -tag -width ".Dv BUS_DMA_ALLOCNOW"
431 .It Dv BUS_DMA_ALLOCNOW
432 Pre-allocate enough resources to handle at least one map load operation on
434 If sufficient resources are not available,
437 This should not be used for tags that only describe buffers that will be
439 .Fn bus_dmamem_alloc .
440 Also, due to resource sharing with other tags, this flag does not guarantee
441 that resources will be allocated or reserved exclusively for this tag.
442 It should be treated only as a minor optimization.
445 Optional lock manipulation function (may be
447 to be called when busdma
448 needs to manipulate a lock on behalf of the client.
455 Optional argument to be passed to the function specified by
458 Pointer to a bus_dma_tag_t where the resulting DMA tag will
464 if sufficient memory is not available for tag creation
465 or allocating mapping resources.
466 .It Fn bus_dma_tag_destroy "dmat"
467 Deallocate the DMA tag
470 .Fn bus_dma_tag_create .
474 if any DMA maps remain associated with
479 .It Fn bus_dmamap_create "dmat" "flags" "*mapp"
480 Allocates and initializes a DMA map.
481 Arguments are as follows:
482 .Bl -tag -width ".Fa nsegments"
487 .Bl -tag -width ".Dv BUS_DMA_COHERENT"
488 .It Dv BUS_DMA_COHERENT
489 Attempt to map the memory loaded with this map such that cache sync
490 operations are as cheap as possible.
491 This flag is typically set on maps when the memory loaded with these will
492 be accessed by both a CPU and a DMA engine, frequently such as control data
493 and as opposed to streamable data such as receive and transmit buffers.
494 Use of this flag does not remove the requirement of using
495 .Fn bus_dmamap_sync ,
496 but it may reduce the cost of performing these operations.
498 .Fn bus_dmamap_create ,
501 flag is currently implemented on sparc64.
506 where the resulting DMA map will be stored.
511 if sufficient memory is not available for creating the
512 map or allocating mapping resources.
513 .It Fn bus_dmamap_destroy "dmat" "map"
514 Frees all resources associated with a given DMA map.
515 Arguments are as follows:
516 .Bl -tag -width ".Fa dmat"
518 DMA tag used to allocate
521 The DMA map to destroy.
526 if a mapping is still active for
528 .It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
529 "callback_arg" "flags"
530 Creates a mapping in device visible address space of
534 associated with the DMA map
536 This call will always return immediately and will not block for any reason.
537 Arguments are as follows:
538 .Bl -tag -width ".Fa buflen"
540 DMA tag used to allocate
543 A DMA map without a currently active mapping.
545 A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
546 mapped into device visible address space.
548 The size of the buffer.
549 .It Fa callback Fa callback_arg
550 The callback function, and its argument.
551 This function is called once sufficient mapping resources are available for
553 If resources are temporarily unavailable, this function will be deferred until
554 later, but the load operation will still return immediately to the caller.
555 Thus, callers should not assume that the callback will be called before the
556 load returns, and code should be structured appropriately to handle this.
557 See below for specific flags and error codes that control this behavior.
560 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
561 .It Dv BUS_DMA_NOWAIT
562 The load should not be deferred in case of insufficient mapping resources,
563 and instead should return immediately with an appropriate error.
564 .It Dv BUS_DMA_NOCACHE
565 The generated transactions to and from the virtual page are non-cacheable.
567 .Fn bus_dmamap_load ,
570 flag is currently implemented on sparc64.
574 Return values to the caller are as follows:
575 .Bl -tag -width ".Er EINPROGRESS"
577 The callback has been called and completed.
578 The status of the mapping has been delivered to the callback.
580 The mapping has been deferred for lack of resources.
581 The callback will be called as soon as resources are available.
582 Callbacks are serviced in FIFO order.
584 Note that subsequent load operations for the same tag that do not require
585 extra resources will still succeed.
586 This may result in out-of-order processing of requests.
587 If the caller requires the order of requests to be preserved,
588 then the caller is required to stall subsequent requests until a pending
589 request's callback is invoked.
591 The load request has failed due to insufficient resources, and the caller
592 specifically used the
596 The load request was invalid.
597 The callback has been called and has been provided the same error.
598 This error value may indicate that
608 argument used to create the dma tag
612 When the callback is called, it is presented with an error value
613 indicating the disposition of the mapping.
614 Error may be one of the following:
615 .Bl -tag -width ".Er EINPROGRESS"
617 The mapping was successful and the
619 callback argument contains an array of
620 .Vt bus_dma_segment_t
621 elements describing the mapping.
622 This array is only valid during the scope of the callback function.
624 A mapping could not be achieved within the segment constraints provided
625 in the tag even though the requested allocation size was less than maxsize.
627 .It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
629 This is a variation of
631 which maps mbuf chains
635 argument is also passed to the callback routine, which
636 contains the mbuf chain's packet header length.
639 flag is implied, thus no callback deferral will happen.
641 Mbuf chains are assumed to be in kernel virtual address space.
643 Beside the error values listed for
644 .Fn bus_dmamap_load ,
646 will be returned if the size of the mbuf chain exceeds the maximum limit of the
648 .It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
650 .Fn bus_dmamap_load_mbuf
651 except that it returns immediately without calling a callback function.
652 It is provided for efficiency.
653 The scatter/gather segment array
655 is provided by the caller and filled in directly by the function.
658 argument is returned with the number of segments filled in.
659 Returns the same errors as
660 .Fn bus_dmamap_load_mbuf .
661 .It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
662 This is a variation of
664 which maps buffers pointed to by
669 argument is also passed to the callback routine, which contains the size of
675 flag is implied, thus no callback deferral will happen.
676 Returns the same errors as
677 .Fn bus_dmamap_load .
683 then it is assumed that the buffer,
686 .Fa "uio->uio_td->td_proc" Ns 's
688 User space memory must be in-core and wired prior to attempting a map
690 Pages may be locked using
692 .It Fn bus_dmamap_unload "dmat" "map"
694 Arguments are as follows:
695 .Bl -tag -width ".Fa dmam"
697 DMA tag used to allocate
700 The DMA map that is to be unloaded.
703 .Fn bus_dmamap_unload
704 will not perform any implicit synchronization of DMA buffers.
705 This must be done explicitly by a call to
707 prior to unloading the map.
708 .It Fn bus_dmamap_sync "dmat" "map" "op"
709 Performs synchronization of a device visible mapping with the CPU visible
710 memory referenced by that mapping.
711 Arguments are as follows:
712 .Bl -tag -width ".Fa dmat"
714 DMA tag used to allocate
717 The DMA mapping to be synchronized.
719 Type of synchronization operation to perform.
720 See the definition of
722 for a description of the acceptable values for
729 is the method used to ensure that CPU's and device's direct
730 memory access (DMA) to shared
732 For example, the CPU might be used to set up the contents of a buffer
733 that is to be made available to a device.
734 To ensure that the data are visible via the device's mapping of that
735 memory, the buffer must be loaded and a DMA sync operation of
736 .Dv BUS_DMASYNC_PREWRITE
737 must be performed after the CPU has updated the buffer and before the device
739 If the CPU modifies this buffer again later, another
740 .Dv BUS_DMASYNC_PREWRITE
741 sync operation must be performed before an additional device
743 Conversely, suppose a device updates memory that is to be read by a CPU.
744 In this case, the buffer must be loaded, and a DMA sync operation of
745 .Dv BUS_DMASYNC_PREREAD
746 must be performed before the device access is initiated.
747 The CPU will only be able to see the results of this memory update
748 once the DMA operation has completed and a
749 .Dv BUS_DMASYNC_POSTREAD
750 sync operation has been performed.
752 If read and write operations are not preceded and followed by the
753 appropriate synchronization operations, behavior is undefined.
754 .It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
755 Allocates memory that is mapped into KVA at the address returned
758 and that is permanently loaded into the newly created
762 Arguments are as follows:
763 .Bl -tag -width ".Fa alignment"
765 DMA tag describing the constraints of the DMA mapping.
767 Pointer to a pointer that will hold the returned KVA mapping of
768 the allocated region.
770 Flags are defined as follows:
771 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
772 .It Dv BUS_DMA_WAITOK
773 The routine can safely wait (sleep) for resources.
774 .It Dv BUS_DMA_NOWAIT
775 The routine is not allowed to wait for resources.
776 If resources are not available,
779 .It Dv BUS_DMA_COHERENT
780 Attempt to map this memory in a coherent fashion.
782 .Fn bus_dmamap_create
783 above for a description of this flag.
785 .Fn bus_dmamem_alloc ,
788 flag is currently implemented on arm and sparc64.
790 Causes the allocated memory to be set to all zeros.
791 .It Dv BUS_DMA_NOCACHE
792 The allocated memory will not be cached in the processor caches.
793 All memory accesses appear on the bus and are executed
796 .Fn bus_dmamem_alloc ,
799 flag is currently implemented on amd64 and i386 where it results in the
800 Strong Uncacheable PAT to be set for the allocated virtual address range.
805 where the resulting DMA map will be stored.
808 The size of memory to be allocated is
810 as specified in the call to
811 .Fn bus_dma_tag_create
815 The current implementation of
817 will allocate all requests as a single segment.
819 An initial load operation is required to obtain the bus address of the allocated
820 memory, and an unload operation is required before freeing the memory, as
822 .Fn bus_dmamem_free .
823 Maps are automatically handled by this function and should not be explicitly
824 allocated or destroyed.
826 Although an explicit load is not required for each access to the memory
827 referenced by the returned map, the synchronization requirements
830 section still apply and should be used to achieve portability on architectures
831 without coherent buses.
835 if sufficient memory is not available for completing
837 .It Fn bus_dmamem_free "dmat" "*vaddr" "map"
838 Frees memory previously allocated by
839 .Fn bus_dmamem_alloc .
842 Arguments are as follows:
843 .Bl -tag -width ".Fa vaddr"
847 Kernel virtual address of the memory.
849 DMA map to be invalidated.
853 Behavior is undefined if invalid arguments are passed to
854 any of the above functions.
855 If sufficient resources cannot be allocated for a given
860 routines that are not of type
862 will return 0 on success or an error
863 code on failure as discussed above.
867 routines will succeed if provided with valid arguments.
869 Two locking protocols are used by
871 The first is a private global lock that is used to synchronize access to the
872 bounce buffer pool on the architectures that make use of them.
873 This lock is strictly a leaf lock that is only used internally to
875 and is not exposed to clients of the API.
877 The second protocol involves protecting various resources stored in the tag.
880 operations are done through requests from the driver that created the tag,
881 the most efficient way to protect the tag resources is through the lock that
885 acts on its own without being called by the driver, the lock primitive
886 specified in the tag is acquired and released automatically.
887 An example of this is when the
889 callback function is called from a deferred context instead of the driver
891 This means that certain
893 functions must always be called with the same lock held that is specified in the
895 These functions include:
897 .Bl -item -offset indent -compact
901 .Fn bus_dmamap_load_uio
903 .Fn bus_dmamap_load_mbuf
905 .Fn bus_dmamap_load_mbuf_sg
907 .Fn bus_dmamap_unload
912 There is one exception to this rule.
913 It is common practice to call some of these functions during driver start-up
914 without any locks held.
915 So long as there is a guarantee of no possible concurrent use of the tag by
916 different threads during this operation, it is safe to not hold a lock for
921 operations should not be called with the driver lock held, either because
922 they are already protected by an internal lock, or because they might sleep
923 due to memory or resource allocation.
924 The following functions must not be
925 called with any non-sleepable locks held:
927 .Bl -item -offset indent -compact
929 .Fn bus_dma_tag_create
931 .Fn bus_dmamap_create
936 All other functions do not have a locking protocol and can thus be
937 called with or without any system or driver locks held.
946 .%A "Jason R. Thorpe"
947 .%T "A Machine-Independent DMA Framework for NetBSD"
948 .%J "Proceedings of the Summer 1998 USENIX Technical Conference"
949 .%Q "USENIX Association"
955 interface first appeared in
962 for use in the CAM SCSI subsystem.
963 The alterations to the original API were aimed to remove the need for
965 .Vt bus_dma_segment_t
968 while allowing callers to queue up on scarce resources.
972 interface was designed and implemented by
974 of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
975 Additional input on the
977 design was provided by
979 .An Chris Demetriou ,
991 benefits from the contributions of
992 .An Justin T. Gibbs ,
995 .An Matthew N. Dodd ,
998 .An Jake Burkholder ,
999 .An Takahashi Yoshihiro ,
1003 This manual page was written by
1006 .An Justin T. Gibbs .