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 lowaddr and
368 less than or equal to highaddr.
369 For example, a device incapable of DMA above 4GB, would specify
371 .Dv BUS_SPACE_MAXADDR
373 .Dv BUS_SPACE_MAXADDR_32BIT .
374 Similarly a device that can only dma to addresses bellow 16MB would
375 specify a highaddr of
376 .Dv BUS_SPACE_MAXADDR
378 .Dv BUS_SPACE_MAXADDR_24BIT .
379 Some implementations requires that some region of device visible
380 address space, overlapping available host memory, be outside the
384 is used to bounce requests that would otherwise conflict with
385 the exclusion window.
387 Optional filter function (may be
389 to be called for any attempt to
390 map memory into the window described by
394 A filter function is only required when the single window described
399 cannot adequately describe the constraints of the device.
400 The filter function will be called for every machine page
401 that overlaps the exclusion window.
403 Argument passed to all calls to the filter function for this tag.
407 Maximum size, in bytes, of the sum of all segment lengths in a given
408 DMA mapping associated with this tag.
410 Number of discontinuities (scatter/gather segments) allowed
411 in a DMA mapped region.
412 If there is no restriction,
413 .Dv BUS_SPACE_UNRESTRICTED
416 Maximum size, in bytes, of a segment in any DMA mapped region associated
421 .Bl -tag -width ".Dv BUS_DMA_ALLOCNOW"
422 .It Dv BUS_DMA_ALLOCNOW
423 Pre-allocate enough resources to handle at least one map load operation on
425 If sufficient resources are not available,
428 This should not be used for tags that only describe buffers that will be
430 .Fn bus_dmamem_alloc .
431 Also, due to resource sharing with other tags, this flag does not guarantee
432 that resources will be allocated or reserved exclusively for this tag.
433 It should be treated only as a minor optimization.
436 Optional lock manipulation function (may be
438 to be called when busdma
439 needs to manipulate a lock on behalf of the client.
446 Optional argument to be passed to the function specified by
449 Pointer to a bus_dma_tag_t where the resulting DMA tag will
455 if sufficient memory is not available for tag creation
456 or allocating mapping resources.
457 .It Fn bus_dma_tag_destroy "dmat"
458 Deallocate the DMA tag
461 .Fn bus_dma_tag_create .
465 if any DMA maps remain associated with
470 .It Fn bus_dmamap_create "dmat" "flags" "*mapp"
471 Allocates and initializes a DMA map.
472 Arguments are as follows:
473 .Bl -tag -width ".Fa nsegments"
477 The value of this argument is currently undefined and should be
483 where the resulting DMA map will be stored.
488 if sufficient memory is not available for creating the
489 map or allocating mapping resources.
490 .It Fn bus_dmamap_destroy "dmat" "map"
491 Frees all resources associated with a given DMA map.
492 Arguments are as follows:
493 .Bl -tag -width ".Fa dmat"
495 DMA tag used to allocate
498 The DMA map to destroy.
503 if a mapping is still active for
505 .It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
506 "callback_arg" "flags"
507 Creates a mapping in device visible address space of
511 associated with the DMA map
513 This call will always return immediately and will not block for any reason.
514 Arguments are as follows:
515 .Bl -tag -width ".Fa buflen"
517 DMA tag used to allocate
520 A DMA map without a currently active mapping.
522 A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
523 mapped into device visible address space.
525 The size of the buffer.
526 .It Fa callback Fa callback_arg
527 The callback function, and its argument.
528 This function is called once sufficient mapping resources are available for
530 If resources are temporarily unavailable, this function will be deferred until
531 later, but the load operation will still return immediately to the caller.
532 Thus, callers should not assume that the callback will be called before the
533 load returns, and code should be structured appropriately to handle this.
534 See below for specific flags and error codes that control this behavior.
537 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
538 .It Dv BUS_DMA_NOWAIT
539 The load should not be deferred in case of insufficient mapping resources,
540 and instead should return immediately with an appropriate error.
544 Return values to the caller are as follows:
545 .Bl -tag -width ".Er EINPROGRESS"
547 The callback has been called and completed.
548 The status of the mapping has been delivered to the callback.
550 The mapping has been deferred for lack of resources.
551 The callback will be called as soon as resources are available.
552 Callbacks are serviced in FIFO order.
553 To ensure that ordering is guaranteed, all subsequent load requests will also
554 be deferred until all callbacks have been processed.
556 The load request has failed due to insufficient resources, and the caller
557 specifically used the
561 The load request was invalid.
562 The callback has been called and has been provided the same error.
563 This error value may indicate that
573 argument used to create the dma tag
577 When the callback is called, it is presented with an error value
578 indicating the disposition of the mapping.
579 Error may be one of the following:
580 .Bl -tag -width ".Er EINPROGRESS"
582 The mapping was successful and the
584 callback argument contains an array of
585 .Vt bus_dma_segment_t
586 elements describing the mapping.
587 This array is only valid during the scope of the callback function.
589 A mapping could not be achieved within the segment constraints provided
590 in the tag even though the requested allocation size was less than maxsize.
592 .It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
594 This is a variation of
596 which maps mbuf chains
600 argument is also passed to the callback routine, which
601 contains the mbuf chain's packet header length.
604 flag is implied, thus no callback deferral will happen.
606 Mbuf chains are assumed to be in kernel virtual address space.
608 Beside the error values listed for
609 .Fn bus_dmamap_load ,
611 will be returned if the size of the mbuf chain exceeds the maximum limit of the
613 .It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
615 .Fn bus_dmamap_load_mbuf
616 except that it returns immediately without calling a callback function.
617 It is provided for efficiency.
618 The scatter/gather segment array
620 is provided by the caller and filled in directly by the function.
623 argument is returned with the number of segments filled in.
624 Returns the same errors as
625 .Fn bus_dmamap_load_mbuf .
626 .It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
627 This is a variation of
629 which maps buffers pointed to by
634 argument is also passed to the callback routine, which contains the size of
640 flag is implied, thus no callback deferral will happen.
641 Returns the same errors as
642 .Fn bus_dmamap_load .
648 then it is assumed that the buffer,
651 .Fa "uio->uio_td->td_proc" Ns 's
653 User space memory must be in-core and wired prior to attempting a map
655 Pages may be locked using
657 .It Fn bus_dmamap_unload "dmat" "map"
659 Arguments are as follows:
660 .Bl -tag -width ".Fa dmam"
662 DMA tag used to allocate
665 The DMA map that is to be unloaded.
668 .Fn bus_dmamap_unload
669 will not perform any implicit synchronization of DMA buffers.
670 This must be done explicitly by a call to
672 prior to unloading the map.
673 .It Fn bus_dmamap_sync "dmat" "map" "op"
674 Performs synchronization of a device visible mapping with the CPU visible
675 memory referenced by that mapping.
676 Arguments are as follows:
677 .Bl -tag -width ".Fa dmat"
679 DMA tag used to allocate
682 The DMA mapping to be synchronized.
684 Type of synchronization operation to perform.
685 See the definition of
687 for a description of the acceptable values for
694 is the method used to ensure that CPU's and device's direct
695 memory access (DMA) to shared
697 For example, the CPU might be used to set up the contents of a buffer
698 that is to be made available to a device.
699 To ensure that the data are visible via the device's mapping of that
700 memory, the buffer must be loaded and a DMA sync operation of
701 .Dv BUS_DMASYNC_PREWRITE
702 must be performed after the CPU has updated the buffer and before the device
704 If the CPU modifies this buffer again later, another
705 .Dv BUS_DMASYNC_PREWRITE
706 sync operation must be performed before an additional device
708 Conversely, suppose a device updates memory that is to be read by a CPU.
709 In this case, the buffer must be loaded, and a DMA sync operation of
710 .Dv BUS_DMASYNC_PREREAD
711 must be performed before the device access is initiated.
712 The CPU will only be able to see the results of this memory update
713 once the DMA operation has completed and a
714 .Dv BUS_DMASYNC_POSTREAD
715 sync operation has been performed.
717 If read and write operations are not preceded and followed by the
718 appropriate synchronization operations, behavior is undefined.
719 .It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
720 Allocates memory that is mapped into KVA at the address returned
723 and that is permanently loaded into the newly created
727 Arguments are as follows:
728 .Bl -tag -width ".Fa alignment"
730 DMA tag describing the constraints of the DMA mapping.
732 Pointer to a pointer that will hold the returned KVA mapping of
733 the allocated region.
735 Flags are defined as follows:
736 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
737 .It Dv BUS_DMA_WAITOK
738 The routine can safely wait (sleep) for resources.
739 .It Dv BUS_DMA_NOWAIT
740 The routine is not allowed to wait for resources.
741 If resources are not available,
744 .It Dv BUS_DMA_COHERENT
745 Attempt to map this memory such that cache sync operations are
746 as cheap as possible.
747 This flag is typically set on memory that will be accessed by both
748 a CPU and a DMA engine, frequently.
749 Use of this flag does not remove the requirement of using
750 bus_dmamap_sync, but it may reduce the cost of performing
754 flag is currently implemented on sparc64 and arm.
756 Causes the allocated memory to be set to all zeros.
761 where the resulting DMA map will be stored.
764 The size of memory to be allocated is
766 as specified in the call to
767 .Fn bus_dma_tag_create
771 The current implementation of
773 will allocate all requests as a single segment.
775 An initial load operation is required to obtain the bus address of the allocated
776 memory, and an unload operation is required before freeing the memory, as
778 .Fn bus_dmamem_free .
779 Maps are automatically handled by this function and should not be explicitly
780 allocated or destroyed.
782 Although an explicit load is not required for each access to the memory
783 referenced by the returned map, the synchronization requirements
786 section still apply and should be used to achieve portability on architectures
787 without coherent buses.
791 if sufficient memory is not available for completing
793 .It Fn bus_dmamem_free "dmat" "*vaddr" "map"
794 Frees memory previously allocated by
795 .Fn bus_dmamem_alloc .
798 Arguments are as follows:
799 .Bl -tag -width ".Fa vaddr"
803 Kernel virtual address of the memory.
805 DMA map to be invalidated.
809 Behavior is undefined if invalid arguments are passed to
810 any of the above functions.
811 If sufficient resources cannot be allocated for a given
816 routines that are not of type
818 will return 0 on success or an error
819 code on failure as discussed above.
823 routines will succeed if provided with valid arguments.
825 Two locking protocols are used by
827 The first is a private global lock that is used to synchronize access to the
828 bounce buffer pool on the architectures that make use of them.
829 This lock is strictly a leaf lock that is only used internally to
831 and is not exposed to clients of the API.
833 The second protocol involves protecting various resources stored in the tag.
836 operations are done through requests from the driver that created the tag,
837 the most efficient way to protect the tag resources is through the lock that
841 acts on its own without being called by the driver, the lock primitive
842 specified in the tag is acquired and released automatically.
843 An example of this is when the
845 callback function is called from a deferred context instead of the driver
847 This means that certain
849 functions must always be called with the same lock held that is specified in the
851 These functions include:
853 .Bl -item -offset indent -compact
857 .Fn bus_dmamap_load_uio
859 .Fn bus_dmamap_load_mbuf
861 .Fn bus_dmamap_load_mbuf_sg
863 .Fn bus_dmamap_unload
868 There is one exception to this rule.
869 It is common practice to call some of these functions during driver start-up
870 without any locks held.
871 So long as there is a guarantee of no possible concurrent use of the tag by
872 different threads during this operation, it is safe to not hold a lock for
877 operations should not be called with the driver lock held, either because
878 they are already protected by an internal lock, or because they might sleep
879 due to memory or resource allocation.
880 The following functions must not be
881 called with any non-sleepable locks held:
883 .Bl -item -offset indent -compact
885 .Fn bus_dma_tag_create
887 .Fn bus_dmamap_create
892 All other functions do not have a locking protocol and can thus be
893 called with or without any system or driver locks held.
902 .%A "Jason R. Thorpe"
903 .%T "A Machine-Independent DMA Framework for NetBSD"
904 .%J "Proceedings of the Summer 1998 USENIX Technical Conference"
905 .%Q "USENIX Association"
911 interface first appeared in
918 for use in the CAM SCSI subsystem.
919 The alterations to the original API were aimed to remove the need for
921 .Vt bus_dma_segment_t
924 while allowing callers to queue up on scarce resources.
928 interface was designed and implemented by
930 of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
931 Additional input on the
933 design was provided by
935 .An Chris Demetriou ,
947 benefits from the contributions of
948 .An Justin T. Gibbs ,
951 .An Matthew N. Dodd ,
954 .An Jake Burkholder ,
955 .An Takahashi Yoshihiro ,
959 This manual page was written by
962 .An Justin T. Gibbs .