1 .\" Copyright (c) 2002, 2003, 2004 Hiten Pandya <hmp@backplane.com>
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.
60 .\" $FreeBSD: /repoman/r/ncvs/src/share/man/man9/bus_dma.9,v 1.7 2003/07/27 14:05:29 mux Exp $
61 .\" $NetBSD: bus_dma.9,v 1.25 2002/10/14 13:43:16 wiz Exp $
62 .\" $DragonFly: src/share/man/man9/bus_dma.9,v 1.2 2004/05/14 16:23:17 hmp Exp $
69 .Nm bus_dma_tag_create ,
70 .Nm bus_dma_tag_destroy ,
71 .Nm bus_dmamap_create ,
72 .Nm bus_dmamap_destroy ,
74 .Nm bus_dmamap_load_mbuf ,
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_tag_t *dmat"
89 .Fn bus_dma_tag_destroy "bus_dma_tag_t dmat"
91 .Fn bus_dmamap_create "bus_dma_tag_t dmat" "int flags" "bus_dmamap_t *mapp"
93 .Fn bus_dmamap_destroy "bus_dma_tag_t dmat" "bus_dmamap_t map"
95 .Fn bus_dmamap_load "bus_dma_tag_t dmat" "bus_dmamap_t map" "void *buf" \
96 "bus_size_t buflen" "bus_dmamap_callback_t *callback" "void *callback_arg" \
99 .Fn bus_dmamap_load_mbuf "bus_dma_tag_t dmat" "bus_dmamap_t map" \
100 "struct mbuf *mbuf" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
103 .Fn bus_dmamap_load_uio "bus_dma_tag_t dmat" "bus_dmamap_t map" \
104 "struct uio *uio" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
107 .Fn bus_dmamem_alloc "bus_dma_tag_t dmat" "void **vaddr" \
108 "int flags" "bus_dmamap_t *mapp"
110 .Fn bus_dmamap_unload "bus_dma_tag_t dmat" "bus_dmamap_t map"
112 .Fn bus_dmamap_sync "bus_dma_tag_t dmat" "bus_dmamap_t map" \
115 .Fn bus_dmamem_free "bus_dma_tag_t dmat" "void *vaddr" \
118 Direct Memory Access (DMA) is a method of transferring data
119 without involving the CPU, thus providing higher performance.
120 A DMA transaction can be achieved between device to memory,
121 device to device, or memory to memory.
125 API is a bus, device, and machine-independent (MI) interface to
127 It provides the client with flexibility and simplicity by
128 abstracting machine dependent issues like setting up
129 DMA mappings, handling cache issues, bus specific features
131 .Sh STRUCTURES AND TYPES
132 .Bl -tag -width compact
134 A machine-dependent (MD) opaque type that describes the
135 characteristics of DMA transactions.
136 DMA tags are organized into a hierarchy, with each child
137 tag inheriting the restrictions of its parent.
138 This allows all devices along the path of DMA transactions
139 to contribute to the constraints of those transactions.
140 .It Vt bus_dma_filter_t
141 Client specified address filter having the format:
142 .Bl -tag -width compact
144 .Fn "client_filter" "void *filtarg" "bus_addr_t testaddr"
147 Address filters can be specified during tag creation to allow
148 for devices who's DMA address restrictions cannot be specified
152 is client specified during tag creation to be passed to all
153 invocations of the callback.
156 argument contains a potential starting address of a DMA mapping.
157 The filter function operates on the set of addresses from
160 .Ql trunc_page(testaddr) + PAGE_SIZE - 1 ,
162 The filter function should return zero for any mapping in this range
163 that can be accommodated by the device and non-zero otherwise.
164 .It Vt bus_dma_segment_t
165 A machine-dependent type that describes individual
174 field contains the device visible address of the DMA segment, and
176 contains the length of the DMA segment.
177 Although the DMA segments returned by a mapping call will adhere to
178 all restrictions necessary for a successful DMA operation, some conversion
179 (e.g. a conversion from host byte order to the device's byte order) is
180 almost always required when presenting segment information to the device.
182 A machine-dependent opaque type describing an individual mapping.
183 Multiple DMA maps can be associated with one DMA tag.
184 .It Vt bus_dmamap_callback_t
185 Client specified callback for receiving mapping information resulting from
189 .Fn bus_dmamap_load .
190 Callbacks are of the format:
191 .Bl -tag -width compact
193 .Fn "client_callback" "void *callback_arg" "bus_dma_segment_t *segs" \
194 "int nseg" "int error"
199 is the callback argument passed to dmamap load functions.
204 parameters describe an array of
205 .Vt bus_dma_segment_t
206 structures that represent the mapping.
207 This array is only valid within the scope of the callback function.
208 The success or failure of the mapping is indicated by the
211 More information on the use of callbacks can be found in the
212 description of the individual dmamap load functions.
213 .It Vt bus_dmamap_callback2_t
214 Client specified callback for receiving mapping information resulting from
218 .Fn bus_dmamap_load_uio
220 .Fn bus_dmamap_load_mbuf .
222 Callback2s are of the format:
223 .Bl -tag -width compact
225 .Fn "client_callback2" "void *callback_arg" "bus_dma_segment_t *segs" \
226 "int nseg" "bus_size_t mapsize" "int error"
229 Callback2's behavior is the same as
230 .Vt bus_dmamap_callback_t
231 with the addition that the length of the data mapped is provided via
233 .It Vt bus_dmasync_op_t
234 Memory synchronization operation specifier.
235 Bus DMA requires explicit synchronization of memory with it's device
236 visible mapping in order to guarantee memory coherency.
239 allows the type of DMA operation that will be or has been performed
240 to be communicated to the system so that the correct coherency measures
242 All operations specified below are performed from the DMA engine's
244 .Bl -tag -width BUS_DMASYNC_POSTWRITE
245 .It Dv BUS_DMASYNC_PREREAD
246 Perform any synchronization required after an update of memory by the CPU
247 but prior to DMA read operations.
248 .It Dv BUS_DMASYNC_PREWRITE
249 Perform any synchronization required after an update of memory by the CPU
250 but prior to DMA write operations.
251 .It Dv BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE
252 Perform any synchronization required prior to a combination of DMA read
253 and write operations.
254 .It Dv BUS_DMASYNC_POSTREAD
255 Perform any synchronization required after DMA read operations, but prior
256 to CPU access of the memory.
257 .It Dv BUS_DMASYNC_POSTWRITE
258 Perform any synchronization required after DMA write operations, but prior
259 to CPU access of the memory.
260 .It Dv BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE
261 Perform any synchronization required after a combination of DMA read
262 and write operations.
267 .Bl -tag -width compact
268 .It Fn bus_dma_tag_create "parent" "alignment" "boundary" "lowaddr" \
269 "highaddr" "*filtfunc" "*filtfuncarg" "maxsize" "nsegments" "maxsegsz" \
271 Allocates a device specific DMA tag, and initializes it according to
272 the arguments provided:
273 .Bl -tag -width *filtfuncarg -compact
275 Indicates restrictions between the parent bridge, CPU memory, and the
277 May be NULL, if no DMA restrictions are to be inherited.
279 Alignment constraint, in bytes, of any mappings created using this tag.
280 The alignment must be a power of 2.
281 Hardware that can DMA starting at any address would specify
284 Hardware requiring DMA transfers to start on a multiple of 4K
288 Boundary constraint, in bytes, of the target DMA memory region.
289 The boundary indicates the set of addresses, all multiples of the
290 boundary argument, that cannot be crossed by a single
291 .Vt bus_dma_segment_t .
292 The boundary must be either a power of 2 or 0.
294 indicates that there are no boundary restrictions.
297 Bounds of the window of bus address space that
299 be directly accessed by the device.
300 The window contains all address greater than lowaddr and
301 less than or equal to highaddr.
302 For example, a device incapable of DMA above 4GB, would specify
304 .Dv BUS_SPACE_MAXADDR
306 .Dv BUS_SPACE_MAXADDR_32BIT .
307 Similarly a device that can only dma to addresses bellow 16MB would
308 specify a highaddr of
309 .Dv BUS_SPACE_MAXADDR
311 .Dv BUS_SPACE_MAXADDR_24BIT .
312 Some implementations requires that some region of device visible
313 address space, overlapping available host memory, be outside the
317 is used to bounce requests that would otherwise conflict with
318 the exclusion window.
320 Optional filter function (may be NULL) to be called for any attempt to
321 map memory into the window described by
325 A filter function is only required when the single window described
330 cannot adequately describe the constraints of the device.
331 The filter function will be called for every machine page
332 that overlaps the exclusion window.
334 Argument passed to all calls to the filter function for this tag.
337 Maximum size, in bytes, of the sum of all segment lengths in a given
338 DMA mapping associated with this tag.
340 Number of discontinuities (scatter/gather segments) allowed
341 in a DMA mapped region.
342 If there is no restriction,
343 .Dv BUS_SPACE_UNRESTRICTED
346 Maximum size, in bytes, of a segment in any DMA mapped region associated
351 .Bl -tag -width "BUS_DMA_ALLOCNOW" -compact
352 .It Dv BUS_DMA_ALLOCNOW
353 Allocate the resources necessary to guarantee that all map load
354 operations associated with this tag will not block.
355 If sufficient resources are not available,
360 Pointer to a bus_dma_tag_t where the resulting DMA tag will
366 if sufficient memory is not available for tag creation
367 or allocating mapping resources.
368 .It Fn bus_dma_tag_destroy "dmat"
369 Deallocate the DMA tag
372 .Fn bus_dma_tag_create .
376 if any DMA maps remain associated with
381 .It Fn bus_dmamap_create "dmat" "flags" "*mapp"
382 Allocates and initializes a DMA map.
383 Arguments are as follows:
384 .Bl -tag -width nsegments -compact
388 The value of this argument is currently undefined and should be
394 where the resulting DMA map will be stored.
399 if sufficient memory is not available for creating the
400 map or allocating mapping resources.
401 .It Fn bus_dmamap_destroy "dmat" "map"
402 Frees all resources associated with a given DMA map.
403 Arguments are as follows:
404 .Bl -tag -width dmat -compact
406 DMA tag used to allocate
409 The DMA map to destroy.
414 if a mapping is still active for
416 .It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" "..."
417 Creates a mapping in device visible address space of
421 associated with the DMA map
423 Arguments are as follows:
424 .Bl -tag -width buflen -compact
426 DMA tag used to allocate
429 A DMA map without a currently active mapping.
431 A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
432 mapped into device visible address space.
434 The size of the buffer.
435 .It Fa callback Fa callback_arg
436 The callback function, and its argument.
438 The value of this argument is currently undefined, and should be
443 Return values to the caller are as follows:
444 .Bl -tag -width EINPROGRESS -compact
446 The callback has been called and completed.
447 The status of the mapping has been delivered to the callback.
449 The mapping has been deferred for lack of resources.
450 The callback will be called as soon as resources are available.
451 Callbacks are serviced in FIFO order.
452 DMA maps created from DMA tags that are allocated with
455 flag will never return this status for a load operation.
457 The load request was invalid.
458 The callback has not, and will not be called.
459 This error value may indicate that
469 argument used to create the dma tag
473 When the callback is called, it is presented with an error value
474 indicating the disposition of the mapping.
475 Error may be one of the following:
476 .Bl -tag -width EINPROGRESS -compact
478 The mapping was successful and the
480 callback argument contains an array of
481 .Vt bus_dma_segment_t
482 elements describing the mapping.
483 This array is only valid during the scope of the callback function.
485 A mapping could not be achieved within the segment constraints provided
486 in the tag even though the requested allocation size was less than maxsize.
488 .It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
490 This is a variation of
492 which maps mbuf chains
496 argument is also passed to the callback routine, which
497 contains the mbuf chain's packet header length.
499 Mbuf chains are assumed to be in kernel virtual address space.
503 if the size of the mbuf chain exceeds the maximum limit of the
505 .It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
506 This is a variation of
508 which maps buffers pointed to by
513 argument is also passed to the callback routine, which contains the size of
522 then it is assumed that the buffer,
525 .Fa "uio->uio_td->td_proc" Ns 's
527 User space memory must be in-core and wired prior to attempting a map
529 .It Fn bus_dmamap_unload "dmat" "map"
531 Arguments are as follows:
532 .Bl -tag -width dmam -compact
534 DMA tag used to allocate
537 The DMA map that is to be unloaded.
540 .Fn bus_dmamap_unload
541 will not perform any implicit synchronization of DMA buffers.
542 This must be done explicitly by a call to
544 prior to unloading the map.
545 .It Fn bus_dmamap_sync "dmat" "map" "op"
546 Performs synchronization of a device visible mapping with the CPU visible
547 memory referenced by that mapping.
548 Arguments are as follows:
549 .Bl -tag -width dmat -compact
551 DMA tag used to allocate
554 The DMA mapping to be synchronized.
556 Type of synchronization operation to perform.
557 See the definition of
559 for a description of the acceptable values for
564 is the method used to ensure that CPU and device DMA access to shared
566 For example, the CPU might be used to setup the contents of a buffer
567 that is to be DMA'ed into a device.
568 To ensure that the data are visible via the device's mapping of that
569 memory, the buffer must be loaded and a dma sync operation of
570 .Dv BUS_DMASYNC_PREREAD
572 Additional sync operations must be performed after every CPU write
573 to this memory if additional DMA reads are to be performed.
574 Conversely, for the DMA write case, the buffer must be loaded,
575 and a dma sync operation of
576 .Dv BUS_DMASYNC_PREWRITE
578 The CPU will only be able to see the results of this DMA write
579 once the DMA has completed and a
580 .Dv BUS_DMASYNC_POSTWRITE
581 operation has been performed.
583 If DMA read and write operations are not preceded and followed by the
584 appropriate synchronization operations, behavior is undefined.
585 .It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "mapp"
586 Allocates memory that is mapped into KVA at the address returned
589 that is permanently loaded into the newly created
593 Arguments are as follows:
594 .Bl -tag -width alignment -compact
596 DMA tag describing the constraints of the DMA mapping.
598 Pointer to a pointer that will hold the returned KVA mapping of
599 the allocated region.
601 Flags are defined as follows:
602 .Bl -tag -width BUS_DMA_NOWAIT -compact
603 .It Dv BUS_DMA_WAITOK
604 The routine can safely wait (sleep) for resources.
605 .It Dv BUS_DMA_NOWAIT
606 The routine is not allowed to wait for resources.
607 If resources are not available,
610 .It Dv BUS_DMA_COHERENT
611 Attempt to map this memory such that cache sync operations are
612 as cheap as possible.
613 This flag is typically set on memory that will be accessed by both
614 a CPU and a DMA engine, frequently.
615 Use of this flag does not remove the requirement of using
616 bus_dmamap_sync, but it may reduce the cost of performing
619 Causes the allocated memory to be set to all zeros.
622 Pointer to storage for the returned DMA map.
625 The size of memory to be allocated is
630 The current implementation of
632 will allocate all requests as a single segment.
634 Although no explicit loading is required to access the memory
635 referenced by the returned map, the synchronization requirements
642 if sufficient memory is not available for completing
644 .It Fn bus_dmamem_free "dmat" "*vaddr" "map"
645 Frees memory previously allocated by
646 .Fn bus_dmamem_alloc .
649 Arguments are as follows:
650 .Bl -tag -width vaddr -compact
654 Kernel virtual address of the memory.
656 DMA map to be invalidated.
660 Behavior is undefined if invalid arguments are passed to
661 any of the above functions.
662 If sufficient resources cannot be allocated for a given
667 routines that are not of type,
669 will return 0 on success or an error
670 code, as discussed above.
674 routines will succeed if provided with valid arguments.
682 .%A "Jason R. Thorpe"
683 .%T "A Machine-Independent DMA Framework for NetBSD"
684 .%J "Proceedings of the Summer 1998 USENIX Technical Conference"
685 .%Q "USENIX Association"
691 interface first appeared in
698 for use in the CAM SCSI subsystem.
699 The alterations to the original API were aimed to remove the need for
701 .Vt bus_dma_segment_t
704 while allowing callers to queue up on scarce resources.
708 interface was designed and implemented by
710 of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
711 Additional input on the
713 design was provided by
715 .An Chris Demetriou ,
723 This manual page was written by
726 .An Justin T. Gibbs .