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.
583 To ensure that ordering is guaranteed, all subsequent load requests will also
584 be deferred until all callbacks have been processed.
586 The load request has failed due to insufficient resources, and the caller
587 specifically used the
591 The load request was invalid.
592 The callback has been called and has been provided the same error.
593 This error value may indicate that
603 argument used to create the dma tag
607 When the callback is called, it is presented with an error value
608 indicating the disposition of the mapping.
609 Error may be one of the following:
610 .Bl -tag -width ".Er EINPROGRESS"
612 The mapping was successful and the
614 callback argument contains an array of
615 .Vt bus_dma_segment_t
616 elements describing the mapping.
617 This array is only valid during the scope of the callback function.
619 A mapping could not be achieved within the segment constraints provided
620 in the tag even though the requested allocation size was less than maxsize.
622 .It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
624 This is a variation of
626 which maps mbuf chains
630 argument is also passed to the callback routine, which
631 contains the mbuf chain's packet header length.
634 flag is implied, thus no callback deferral will happen.
636 Mbuf chains are assumed to be in kernel virtual address space.
638 Beside the error values listed for
639 .Fn bus_dmamap_load ,
641 will be returned if the size of the mbuf chain exceeds the maximum limit of the
643 .It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
645 .Fn bus_dmamap_load_mbuf
646 except that it returns immediately without calling a callback function.
647 It is provided for efficiency.
648 The scatter/gather segment array
650 is provided by the caller and filled in directly by the function.
653 argument is returned with the number of segments filled in.
654 Returns the same errors as
655 .Fn bus_dmamap_load_mbuf .
656 .It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
657 This is a variation of
659 which maps buffers pointed to by
664 argument is also passed to the callback routine, which contains the size of
670 flag is implied, thus no callback deferral will happen.
671 Returns the same errors as
672 .Fn bus_dmamap_load .
678 then it is assumed that the buffer,
681 .Fa "uio->uio_td->td_proc" Ns 's
683 User space memory must be in-core and wired prior to attempting a map
685 Pages may be locked using
687 .It Fn bus_dmamap_unload "dmat" "map"
689 Arguments are as follows:
690 .Bl -tag -width ".Fa dmam"
692 DMA tag used to allocate
695 The DMA map that is to be unloaded.
698 .Fn bus_dmamap_unload
699 will not perform any implicit synchronization of DMA buffers.
700 This must be done explicitly by a call to
702 prior to unloading the map.
703 .It Fn bus_dmamap_sync "dmat" "map" "op"
704 Performs synchronization of a device visible mapping with the CPU visible
705 memory referenced by that mapping.
706 Arguments are as follows:
707 .Bl -tag -width ".Fa dmat"
709 DMA tag used to allocate
712 The DMA mapping to be synchronized.
714 Type of synchronization operation to perform.
715 See the definition of
717 for a description of the acceptable values for
724 is the method used to ensure that CPU's and device's direct
725 memory access (DMA) to shared
727 For example, the CPU might be used to set up the contents of a buffer
728 that is to be made available to a device.
729 To ensure that the data are visible via the device's mapping of that
730 memory, the buffer must be loaded and a DMA sync operation of
731 .Dv BUS_DMASYNC_PREWRITE
732 must be performed after the CPU has updated the buffer and before the device
734 If the CPU modifies this buffer again later, another
735 .Dv BUS_DMASYNC_PREWRITE
736 sync operation must be performed before an additional device
738 Conversely, suppose a device updates memory that is to be read by a CPU.
739 In this case, the buffer must be loaded, and a DMA sync operation of
740 .Dv BUS_DMASYNC_PREREAD
741 must be performed before the device access is initiated.
742 The CPU will only be able to see the results of this memory update
743 once the DMA operation has completed and a
744 .Dv BUS_DMASYNC_POSTREAD
745 sync operation has been performed.
747 If read and write operations are not preceded and followed by the
748 appropriate synchronization operations, behavior is undefined.
749 .It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
750 Allocates memory that is mapped into KVA at the address returned
753 and that is permanently loaded into the newly created
757 Arguments are as follows:
758 .Bl -tag -width ".Fa alignment"
760 DMA tag describing the constraints of the DMA mapping.
762 Pointer to a pointer that will hold the returned KVA mapping of
763 the allocated region.
765 Flags are defined as follows:
766 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
767 .It Dv BUS_DMA_WAITOK
768 The routine can safely wait (sleep) for resources.
769 .It Dv BUS_DMA_NOWAIT
770 The routine is not allowed to wait for resources.
771 If resources are not available,
774 .It Dv BUS_DMA_COHERENT
775 Attempt to map this memory in a coherent fashion.
777 .Fn bus_dmamap_create
778 above for a description of this flag.
780 .Fn bus_dmamem_alloc ,
783 flag is currently implemented on arm and sparc64.
785 Causes the allocated memory to be set to all zeros.
786 .It Dv BUS_DMA_NOCACHE
787 The allocated memory will not be cached in the processor caches.
788 All memory accesses appear on the bus and are executed
791 .Fn bus_dmamem_alloc ,
794 flag is currently implemented on amd64 and i386 where it results in the
795 Strong Uncacheable PAT to be set for the allocated virtual address range.
800 where the resulting DMA map will be stored.
803 The size of memory to be allocated is
805 as specified in the call to
806 .Fn bus_dma_tag_create
810 The current implementation of
812 will allocate all requests as a single segment.
814 An initial load operation is required to obtain the bus address of the allocated
815 memory, and an unload operation is required before freeing the memory, as
817 .Fn bus_dmamem_free .
818 Maps are automatically handled by this function and should not be explicitly
819 allocated or destroyed.
821 Although an explicit load is not required for each access to the memory
822 referenced by the returned map, the synchronization requirements
825 section still apply and should be used to achieve portability on architectures
826 without coherent buses.
830 if sufficient memory is not available for completing
832 .It Fn bus_dmamem_free "dmat" "*vaddr" "map"
833 Frees memory previously allocated by
834 .Fn bus_dmamem_alloc .
837 Arguments are as follows:
838 .Bl -tag -width ".Fa vaddr"
842 Kernel virtual address of the memory.
844 DMA map to be invalidated.
848 Behavior is undefined if invalid arguments are passed to
849 any of the above functions.
850 If sufficient resources cannot be allocated for a given
855 routines that are not of type
857 will return 0 on success or an error
858 code on failure as discussed above.
862 routines will succeed if provided with valid arguments.
864 Two locking protocols are used by
866 The first is a private global lock that is used to synchronize access to the
867 bounce buffer pool on the architectures that make use of them.
868 This lock is strictly a leaf lock that is only used internally to
870 and is not exposed to clients of the API.
872 The second protocol involves protecting various resources stored in the tag.
875 operations are done through requests from the driver that created the tag,
876 the most efficient way to protect the tag resources is through the lock that
880 acts on its own without being called by the driver, the lock primitive
881 specified in the tag is acquired and released automatically.
882 An example of this is when the
884 callback function is called from a deferred context instead of the driver
886 This means that certain
888 functions must always be called with the same lock held that is specified in the
890 These functions include:
892 .Bl -item -offset indent -compact
896 .Fn bus_dmamap_load_uio
898 .Fn bus_dmamap_load_mbuf
900 .Fn bus_dmamap_load_mbuf_sg
902 .Fn bus_dmamap_unload
907 There is one exception to this rule.
908 It is common practice to call some of these functions during driver start-up
909 without any locks held.
910 So long as there is a guarantee of no possible concurrent use of the tag by
911 different threads during this operation, it is safe to not hold a lock for
916 operations should not be called with the driver lock held, either because
917 they are already protected by an internal lock, or because they might sleep
918 due to memory or resource allocation.
919 The following functions must not be
920 called with any non-sleepable locks held:
922 .Bl -item -offset indent -compact
924 .Fn bus_dma_tag_create
926 .Fn bus_dmamap_create
931 All other functions do not have a locking protocol and can thus be
932 called with or without any system or driver locks held.
941 .%A "Jason R. Thorpe"
942 .%T "A Machine-Independent DMA Framework for NetBSD"
943 .%J "Proceedings of the Summer 1998 USENIX Technical Conference"
944 .%Q "USENIX Association"
950 interface first appeared in
957 for use in the CAM SCSI subsystem.
958 The alterations to the original API were aimed to remove the need for
960 .Vt bus_dma_segment_t
963 while allowing callers to queue up on scarce resources.
967 interface was designed and implemented by
969 of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
970 Additional input on the
972 design was provided by
974 .An Chris Demetriou ,
986 benefits from the contributions of
987 .An Justin T. Gibbs ,
990 .An Matthew N. Dodd ,
993 .An Jake Burkholder ,
994 .An Takahashi Yoshihiro ,
998 This manual page was written by
1001 .An Justin T. Gibbs .