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_uio ,
75 .Nm bus_dmamap_unload ,
77 .Nm bus_dmamem_alloc ,
79 .Nd Bus and Machine Independent DMA Mapping Interface
83 .Fn bus_dma_tag_create "bus_dma_tag_t parent" "bus_size_t alignment" \
84 "bus_size_t boundary" "bus_addr_t lowaddr" "bus_addr_t highaddr" \
85 "bus_dma_filter_t *filtfunc" "void *filtfuncarg" "bus_size_t maxsize" \
86 "int nsegments" "bus_size_t maxsegsz" "int flags" "bus_dma_lock_t *lockfunc" \
87 "void *lockfuncarg" "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.
264 .It Vt bus_dma_lock_t
265 Client specified lock/mutex manipulation method.
266 This will be called from
267 within busdma whenever a client lock needs to be manipulated.
268 This method is of the format:
269 .Bl -tag -width compact
271 .Fn "lockfunc" "void *lockfunc_arg" "bus_dma_lock_op_t op"
276 implementations are provided for convenience.
277 .Fn busdma_lock_mutex
278 performs standard mutex operations on the sleep mutex provided via the
281 .Fn bus_dma_tag_create .
283 will generate a system panic if it is called.
284 It is substituted into the tag when
287 .Fn bus_dma_tag_create .
288 .It Vt bus_dma_lock_op_t
289 Operations to be performed by the client-specified
291 .Bl -tag -width BUS_DMA_UNLOCK
293 Aquires and/or locks the client locking primitive.
294 .It Dv BUS_DMA_UNLOCK
295 Releases and/or unlocks the client locking primitive.
300 .Bl -tag -width compact
301 .It Fn bus_dma_tag_create "parent" "alignment" "boundary" "lowaddr" \
302 "highaddr" "*filtfunc" "*filtfuncarg" "maxsize" "nsegments" "maxsegsz" \
303 "flags" "lockfunc" "lockfuncarg" "*dmat"
304 Allocates a device specific DMA tag, and initializes it according to
305 the arguments provided:
306 .Bl -tag -width *filtfuncarg -compact
308 Indicates restrictions between the parent bridge, CPU memory, and the
310 May be NULL, if no DMA restrictions are to be inherited.
312 Alignment constraint, in bytes, of any mappings created using this tag.
313 The alignment must be a power of 2.
314 Hardware that can DMA starting at any address would specify
317 Hardware requiring DMA transfers to start on a multiple of 4K
321 Boundary constraint, in bytes, of the target DMA memory region.
322 The boundary indicates the set of addresses, all multiples of the
323 boundary argument, that cannot be crossed by a single
324 .Vt bus_dma_segment_t .
325 The boundary must be either a power of 2 or 0.
327 indicates that there are no boundary restrictions.
330 Bounds of the window of bus address space that
332 be directly accessed by the device.
333 The window contains all address greater than lowaddr and
334 less than or equal to highaddr.
335 For example, a device incapable of DMA above 4GB, would specify
337 .Dv BUS_SPACE_MAXADDR
339 .Dv BUS_SPACE_MAXADDR_32BIT .
340 Similarly a device that can only dma to addresses bellow 16MB would
341 specify a highaddr of
342 .Dv BUS_SPACE_MAXADDR
344 .Dv BUS_SPACE_MAXADDR_24BIT .
345 Some implementations requires that some region of device visible
346 address space, overlapping available host memory, be outside the
350 is used to bounce requests that would otherwise conflict with
351 the exclusion window.
353 Optional filter function (may be NULL) to be called for any attempt to
354 map memory into the window described by
358 A filter function is only required when the single window described
363 cannot adequately describe the constraints of the device.
364 The filter function will be called for every machine page
365 that overlaps the exclusion window.
367 Argument passed to all calls to the filter function for this tag.
370 Maximum size, in bytes, of the sum of all segment lengths in a given
371 DMA mapping associated with this tag.
373 Number of discontinuities (scatter/gather segments) allowed
374 in a DMA mapped region.
375 If there is no restriction,
376 .Dv BUS_SPACE_UNRESTRICTED
379 Maximum size, in bytes, of a segment in any DMA mapped region associated
384 .Bl -tag -width "BUS_DMA_ALLOCNOW" -compact
385 .It Dv BUS_DMA_ALLOCNOW
386 Allocate the resources necessary to guarantee that all map load
387 operations associated with this tag will not block.
388 If sufficient resources are not available,
393 Optional lock manipulation function (may be NULL) to be called when busdma
394 needs to manipulate a lock on behalf of the client.
395 If NULL is specified,
399 Optional argument to be passed to the function specified by
402 Pointer to a bus_dma_tag_t where the resulting DMA tag will
408 if sufficient memory is not available for tag creation
409 or allocating mapping resources.
410 .It Fn bus_dma_tag_destroy "dmat"
411 Deallocate the DMA tag
414 .Fn bus_dma_tag_create .
418 if any DMA maps remain associated with
423 .It Fn bus_dmamap_create "dmat" "flags" "*mapp"
424 Allocates and initializes a DMA map.
425 Arguments are as follows:
426 .Bl -tag -width nsegments -compact
430 The value of this argument is currently undefined and should be
436 where the resulting DMA map will be stored.
441 if sufficient memory is not available for creating the
442 map or allocating mapping resources.
443 .It Fn bus_dmamap_destroy "dmat" "map"
444 Frees all resources associated with a given DMA map.
445 Arguments are as follows:
446 .Bl -tag -width dmat -compact
448 DMA tag used to allocate
451 The DMA map to destroy.
456 if a mapping is still active for
458 .It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" "..."
459 Creates a mapping in device visible address space of
463 associated with the DMA map
465 Arguments are as follows:
466 .Bl -tag -width buflen -compact
468 DMA tag used to allocate
471 A DMA map without a currently active mapping.
473 A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
474 mapped into device visible address space.
476 The size of the buffer.
477 .It Fa callback Fa callback_arg
478 The callback function, and its argument.
480 The value of this argument is currently undefined, and should be
485 Return values to the caller are as follows:
486 .Bl -tag -width EINPROGRESS -compact
488 The callback has been called and completed.
489 The status of the mapping has been delivered to the callback.
491 The mapping has been deferred for lack of resources.
492 The callback will be called as soon as resources are available.
493 Callbacks are serviced in FIFO order.
494 DMA maps created from DMA tags that are allocated with
497 flag will never return this status for a load operation.
499 The load request was invalid.
500 The callback has not, and will not be called.
501 This error value may indicate that
511 argument used to create the dma tag
515 When the callback is called, it is presented with an error value
516 indicating the disposition of the mapping.
517 Error may be one of the following:
518 .Bl -tag -width EINPROGRESS -compact
520 The mapping was successful and the
522 callback argument contains an array of
523 .Vt bus_dma_segment_t
524 elements describing the mapping.
525 This array is only valid during the scope of the callback function.
527 A mapping could not be achieved within the segment constraints provided
528 in the tag even though the requested allocation size was less than maxsize.
530 .It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
532 This is a variation of
534 which maps mbuf chains
538 argument is also passed to the callback routine, which
539 contains the mbuf chain's packet header length.
541 Mbuf chains are assumed to be in kernel virtual address space.
545 if the size of the mbuf chain exceeds the maximum limit of the
547 .It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
548 This is a variation of
550 which maps buffers pointed to by
555 argument is also passed to the callback routine, which contains the size of
564 then it is assumed that the buffer,
567 .Fa "uio->uio_td->td_proc" Ns 's
569 User space memory must be in-core and wired prior to attempting a map
571 .It Fn bus_dmamap_unload "dmat" "map"
573 Arguments are as follows:
574 .Bl -tag -width dmam -compact
576 DMA tag used to allocate
579 The DMA map that is to be unloaded.
582 .Fn bus_dmamap_unload
583 will not perform any implicit synchronization of DMA buffers.
584 This must be done explicitly by a call to
586 prior to unloading the map.
587 .It Fn bus_dmamap_sync "dmat" "map" "op"
588 Performs synchronization of a device visible mapping with the CPU visible
589 memory referenced by that mapping.
590 Arguments are as follows:
591 .Bl -tag -width dmat -compact
593 DMA tag used to allocate
596 The DMA mapping to be synchronized.
598 Type of synchronization operation to perform.
599 See the definition of
601 for a description of the acceptable values for
606 is the method used to ensure that CPU and device DMA access to shared
608 For example, the CPU might be used to setup the contents of a buffer
609 that is to be DMA'ed into a device.
610 To ensure that the data are visible via the device's mapping of that
611 memory, the buffer must be loaded and a dma sync operation of
612 .Dv BUS_DMASYNC_PREREAD
614 Additional sync operations must be performed after every CPU write
615 to this memory if additional DMA reads are to be performed.
616 Conversely, for the DMA write case, the buffer must be loaded,
617 and a dma sync operation of
618 .Dv BUS_DMASYNC_PREWRITE
620 The CPU will only be able to see the results of this DMA write
621 once the DMA has completed and a
622 .Dv BUS_DMASYNC_POSTWRITE
623 operation has been performed.
625 If DMA read and write operations are not preceded and followed by the
626 appropriate synchronization operations, behavior is undefined.
627 .It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "mapp"
628 Allocates memory that is mapped into KVA at the address returned
631 that is permanently loaded into the newly created
635 Arguments are as follows:
636 .Bl -tag -width alignment -compact
638 DMA tag describing the constraints of the DMA mapping.
640 Pointer to a pointer that will hold the returned KVA mapping of
641 the allocated region.
643 Flags are defined as follows:
644 .Bl -tag -width BUS_DMA_NOWAIT -compact
645 .It Dv BUS_DMA_WAITOK
646 The routine can safely wait (sleep) for resources.
647 .It Dv BUS_DMA_NOWAIT
648 The routine is not allowed to wait for resources.
649 If resources are not available,
652 .It Dv BUS_DMA_COHERENT
653 Attempt to map this memory such that cache sync operations are
654 as cheap as possible.
655 This flag is typically set on memory that will be accessed by both
656 a CPU and a DMA engine, frequently.
657 Use of this flag does not remove the requirement of using
658 bus_dmamap_sync, but it may reduce the cost of performing
661 Causes the allocated memory to be set to all zeros.
664 Pointer to storage for the returned DMA map.
667 The size of memory to be allocated is
672 The current implementation of
674 will allocate all requests as a single segment.
676 Although no explicit loading is required to access the memory
677 referenced by the returned map, the synchronization requirements
684 if sufficient memory is not available for completing
686 .It Fn bus_dmamem_free "dmat" "*vaddr" "map"
687 Frees memory previously allocated by
688 .Fn bus_dmamem_alloc .
691 Arguments are as follows:
692 .Bl -tag -width vaddr -compact
696 Kernel virtual address of the memory.
698 DMA map to be invalidated.
702 Behavior is undefined if invalid arguments are passed to
703 any of the above functions.
704 If sufficient resources cannot be allocated for a given
709 routines that are not of type,
711 will return 0 on success or an error
712 code, as discussed above.
716 routines will succeed if provided with valid arguments.
724 .%A "Jason R. Thorpe"
725 .%T "A Machine-Independent DMA Framework for NetBSD"
726 .%J "Proceedings of the Summer 1998 USENIX Technical Conference"
727 .%Q "USENIX Association"
733 interface first appeared in
740 for use in the CAM SCSI subsystem.
741 The alterations to the original API were aimed to remove the need for
743 .Vt bus_dma_segment_t
746 while allowing callers to queue up on scarce resources.
750 interface was designed and implemented by
752 of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
753 Additional input on the
755 design was provided by
757 .An Chris Demetriou ,
769 benefits from the contributions of
770 .An Justin T. Gibbs ,
773 .An Matthew N. Dodd ,
776 .An Jake Burkholder ,
777 .An Takahashi Yoshihiro ,
781 This manual page was written by
784 .An Justin T. Gibbs .