xref: /freebsd/share/man/man9/bus_dma.9 (revision 74bf4e164ba5851606a27d4feff27717452583e5)
1.\" Copyright (c) 2002, 2003 Hiten M. Pandya.
2.\" All rights reserved.
3.\"
4.\" Redistribution and use in source and binary forms, with or without
5.\" modification, are permitted provided that the following conditions
6.\" are met:
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.
12.\"
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.
24.\"
25.\" Copyright (c) 1996, 1997, 1998, 2001 The NetBSD Foundation, Inc.
26.\" All rights reserved.
27.\"
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.
31.\"
32.\" Redistribution and use in source and binary forms, with or without
33.\" modification, are permitted provided that the following conditions
34.\" are met:
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.
47.\"
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.
59.\"
60.\" $FreeBSD$
61.\" $NetBSD: bus_dma.9,v 1.25 2002/10/14 13:43:16 wiz Exp $
62.\"
63.Dd May 28, 2003
64.Dt BUS_DMA 9
65.Os
66.Sh NAME
67.Nm bus_dma ,
68.Nm bus_dma_tag_create ,
69.Nm bus_dma_tag_destroy ,
70.Nm bus_dmamap_create ,
71.Nm bus_dmamap_destroy ,
72.Nm bus_dmamap_load ,
73.Nm bus_dmamap_load_mbuf ,
74.Nm bus_dmamap_load_uio ,
75.Nm bus_dmamap_unload ,
76.Nm bus_dmamap_sync ,
77.Nm bus_dmamem_alloc ,
78.Nm bus_dmamem_free
79.Nd Bus and Machine Independent DMA Mapping Interface
80.Sh SYNOPSIS
81.In machine/bus.h
82.Ft int
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"
88.Ft int
89.Fn bus_dma_tag_destroy "bus_dma_tag_t dmat"
90.Ft int
91.Fn bus_dmamap_create "bus_dma_tag_t dmat" "int flags" "bus_dmamap_t *mapp"
92.Ft int
93.Fn bus_dmamap_destroy "bus_dma_tag_t dmat" "bus_dmamap_t map"
94.Ft int
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" \
97"int flags"
98.Ft int
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" \
101"int flags"
102.Ft int
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" \
105"int flags"
106.Ft void
107.Fn bus_dmamap_unload "bus_dma_tag_t dmat" "bus_dmamap_t map"
108.Ft void
109.Fn bus_dmamap_sync "bus_dma_tag_t dmat" "bus_dmamap_t map" \
110"op"
111.Ft int
112.Fn bus_dmamem_alloc "bus_dma_tag_t dmat" "void **vaddr" \
113"int flags" "bus_dmamap_t *mapp"
114.Ft void
115.Fn bus_dmamem_free "bus_dma_tag_t dmat" "void *vaddr" \
116"bus_dmamap_t map"
117.Sh DESCRIPTION
118Direct Memory Access (DMA) is a method of transferring data
119without involving the CPU, thus providing higher performance.
120A DMA transaction can be achieved between device to memory,
121device to device, or memory to memory.
122.Pp
123The
124.Nm
125API is a bus, device, and machine-independent (MI) interface to
126DMA mechanisms.
127It provides the client with flexibility and simplicity by
128abstracting machine dependent issues like setting up
129DMA mappings, handling cache issues, bus specific features
130and limitations.
131.Sh STRUCTURES AND TYPES
132.Bl -tag -width compact
133.It Vt bus_dma_tag_t
134A machine-dependent (MD) opaque type that describes the
135characteristics of DMA transactions.
136DMA tags are organized into a hierarchy, with each child
137tag inheriting the restrictions of its parent.
138This allows all devices along the path of DMA transactions
139to contribute to the constraints of those transactions.
140.It Vt bus_dma_filter_t
141Client specified address filter having the format:
142.Bl -tag -width compact
143.It Ft int
144.Fn "client_filter" "void *filtarg" "bus_addr_t testaddr"
145.El
146.sp
147Address filters can be specified during tag creation to allow
148for devices who's DMA address restrictions cannot be specified
149by a single window.
150The
151.Fa filtarg
152is client specified during tag creation to be passed to all
153invocations of the callback.
154The
155.Fa testaddr
156argument contains a potential starting address of a DMA mapping.
157The filter function operates on the set of addresses from
158.Fa testaddr
159to
160.Ql trunc_page(testaddr) + PAGE_SIZE - 1 ,
161inclusive.
162The filter function should return zero for any mapping in this range
163that can be accommodated by the device and non-zero otherwise.
164.It Vt bus_dma_segment_t
165A machine-dependent type that describes individual
166DMA segments.
167.Bd -literal
168	bus_addr_t	ds_addr;
169	bus_size_t	ds_len;
170.Ed
171.sp
172The
173.Fa ds_addr
174field contains the device visible address of the DMA segment, and
175.Fa ds_len
176contains the length of the DMA segment.
177Although the DMA segments returned by a mapping call will adhere to
178all 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
180almost always required when presenting segment information to the device.
181.It Vt bus_dmamap_t
182A machine-dependent opaque type describing an individual mapping.
183Multiple DMA maps can be associated with one DMA tag.
184.It Vt bus_dmamap_callback_t
185Client specified callback for receiving mapping information resulting from
186the load of a
187.Vt bus_dmamap_t
188via
189.Fn bus_dmamap_load .
190Callbacks are of the format:
191.Bl -tag -width compact
192.It Ft void
193.Fn "client_callback" "void *callback_arg" "bus_dma_segment_t *segs" \
194"int nseg" "int error"
195.El
196.sp
197The
198.Fa callback_arg
199is the callback argument passed to dmamap load functions.
200The
201.Fa segs
202and
203.Fa nseg
204parameters describe an array of
205.Vt bus_dma_segment_t
206structures that represent the mapping.
207This array is only valid within the scope of the callback function.
208The success or failure of the mapping is indicated by the
209.Fa error
210parameter.
211More information on the use of callbacks can be found in the
212description of the individual dmamap load functions.
213.It Vt bus_dmamap_callback2_t
214Client specified callback for receiving mapping information resulting from
215the load of a
216.Vt bus_dmamap_t
217via
218.Fn bus_dmamap_load_uio
219or
220.Fn bus_dmamap_load_mbuf .
221.sp
222Callback2s are of the format:
223.Bl -tag -width compact
224.It Ft void
225.Fn "client_callback2" "void *callback_arg" "bus_dma_segment_t *segs" \
226"int nseg" "bus_size_t mapsize" "int error"
227.El
228.sp
229Callback2's behavior is the same as
230.Vt bus_dmamap_callback_t
231with the addition that the length of the data mapped is provided via
232.Fa mapsize .
233.It Vt bus_dmasync_op_t
234Memory synchronization operation specifier.
235Bus DMA requires explicit synchronization of memory with it's device
236visible mapping in order to guarantee memory coherency.
237The
238.Vt bus_dmasync_op_t
239allows the type of DMA operation that will be or has been performed
240to be communicated to the system so that the correct coherency measures
241are taken.
242All operations specified below are performed from the CPU's
243point of view (for a complete description, see the
244.Fn bus_dmamap_sync
245description below):
246.Bl -tag -width BUS_DMASYNC_POSTWRITE
247.It Dv BUS_DMASYNC_PREREAD
248Perform any synchronization required after an update of memory by the CPU
249but prior to DMA read operations.
250.It Dv BUS_DMASYNC_PREWRITE
251Perform any synchronization required after an update of memory by the CPU
252but prior to DMA write operations.
253.It Dv BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE
254Perform any synchronization required prior to a combination of DMA read
255and write operations.
256.It Dv BUS_DMASYNC_POSTREAD
257Perform any synchronization required after DMA read operations, but prior
258to CPU access of the memory.
259.It Dv BUS_DMASYNC_POSTWRITE
260Perform any synchronization required after DMA write operations, but prior
261to CPU access of the memory.
262.It Dv BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE
263Perform any synchronization required after a combination of DMA read
264and write operations.
265.El
266.It Vt bus_dma_lock_t
267Client specified lock/mutex manipulation method.
268This will be called from
269within busdma whenever a client lock needs to be manipulated.
270In its current form, the function will be called immediately before
271the callback for a dma load operation that has been deferred with
272.Dv BUS_DMA_LOCK
273and immediately after with
274.Dv BUS_DMA_UNLOCK .
275If the load operation does not need to be deferred, then it
276will not be called since the function loading the map should
277be holding the appropriate locks.
278This method is of the format:
279.Bl -tag -width compact
280.It Ft void
281.Fn "lockfunc" "void *lockfunc_arg" "bus_dma_lock_op_t op"
282.El
283.sp
284Two
285.Vt lockfunc
286implementations are provided for convenience.
287.Fn busdma_lock_mutex
288performs standard mutex operations on the sleep mutex provided via the
289.Fa lockfuncarg .
290passed into
291.Fn bus_dma_tag_create .
292.Fn dflt_lock
293will generate a system panic if it is called.
294It is substituted into the tag when
295.Fa lockfunc
296is passed as NULL to
297.Fn bus_dma_tag_create .
298.It Vt bus_dma_lock_op_t
299Operations to be performed by the client-specified
300.Fn lockfunc .
301.Bl -tag -width BUS_DMA_UNLOCK
302.It Dv BUS_DMA_LOCK
303Acquires and/or locks the client locking primitive.
304.It Dv BUS_DMA_UNLOCK
305Releases and/or unlocks the client locking primitive.
306.El
307.El
308.sp
309.Sh FUNCTIONS
310.Bl -tag -width compact
311.It Fn bus_dma_tag_create "parent" "alignment" "boundary" "lowaddr" \
312"highaddr" "*filtfunc" "*filtfuncarg" "maxsize" "nsegments" "maxsegsz" \
313"flags" "lockfunc" "lockfuncarg" "*dmat"
314Allocates a device specific DMA tag, and initializes it according to
315the arguments provided:
316.Bl -tag -width *filtfuncarg -compact
317.It Fa parent
318Indicates restrictions between the parent bridge, CPU memory, and the
319device.
320May be NULL, if no DMA restrictions are to be inherited.
321.It Fa alignment
322Alignment constraint, in bytes, of any mappings created using this tag.
323The alignment must be a power of 2.
324Hardware that can DMA starting at any address would specify
325.Em 1
326for byte alignment.
327Hardware requiring DMA transfers to start on a multiple of 4K
328would specify
329.Em 4096.
330.It Fa boundary
331Boundary constraint, in bytes, of the target DMA memory region.
332The boundary indicates the set of addresses, all multiples of the
333boundary argument, that cannot be crossed by a single
334.Vt bus_dma_segment_t .
335The boundary must be a power of 2 and must be no smaller that the
336maximum segment size.
337.Ql 0
338indicates that there are no boundary restrictions.
339.It Fa lowaddr
340.It Fa highaddr
341Bounds of the window of bus address space that
342.Em cannot
343be directly accessed by the device.
344The window contains all address greater than lowaddr and
345less than or equal to highaddr.
346For example, a device incapable of DMA above 4GB, would specify
347a highaddr of
348.Dv BUS_SPACE_MAXADDR
349and a lowaddr of
350.Dv BUS_SPACE_MAXADDR_32BIT .
351Similarly a device that can only dma to addresses bellow 16MB would
352specify a highaddr of
353.Dv BUS_SPACE_MAXADDR
354and a lowaddr of
355.Dv BUS_SPACE_MAXADDR_24BIT .
356Some implementations requires that some region of device visible
357address space, overlapping available host memory, be outside the
358window.
359This area of
360.Ql safe memory
361is used to bounce requests that would otherwise conflict with
362the exclusion window.
363.It Fa filtfunc
364Optional filter function (may be NULL) to be called for any attempt to
365map memory into the window described by
366.Fa lowaddr
367and
368.Fa highaddr.
369A filter function is only required when the single window described
370by
371.Fa lowaddr
372and
373.Fa highaddr
374cannot adequately describe the constraints of the device.
375The filter function will be called for every machine page
376that overlaps the exclusion window.
377.It Fa filtfuncarg
378Argument passed to all calls to the filter function for this tag.
379May be NULL.
380.It Fa maxsize
381Maximum size, in bytes, of the sum of all segment lengths in a given
382DMA mapping associated with this tag.
383.It Fa nsegments
384Number of discontinuities (scatter/gather segments) allowed
385in a DMA mapped region.
386If there is no restriction,
387.Dv BUS_SPACE_UNRESTRICTED
388may be specified.
389.It Fa maxsegsz
390Maximum size, in bytes, of a segment in any DMA mapped region associated
391with
392.Fa dmat .
393.It Fa flags
394Are as follows:
395.Bl -tag -width "BUS_DMA_ALLOCNOW" -compact
396.It Dv BUS_DMA_ALLOCNOW
397Allocate the resources necessary to guarantee that all map load
398operations associated with this tag will not block.
399If sufficient resources are not available,
400.Er ENOMEM
401is returned.
402.El
403.It Fa lockfunc
404Optional lock manipulation function (may be NULL) to be called when busdma
405needs to manipulate a lock on behalf of the client.
406If NULL is specified,
407.Fn dflt_lock
408is used.
409.It Fa lockfuncarg
410Optional argument to be passed to the function specified by
411.Fa lockfunc .
412.It Fa dmat
413Pointer to a bus_dma_tag_t where the resulting DMA tag will
414be stored.
415.El
416.Pp
417Returns
418.Er ENOMEM
419if sufficient memory is not available for tag creation
420or allocating mapping resources.
421.It Fn bus_dma_tag_destroy "dmat"
422Deallocate the DMA tag
423.Fa dmat
424that was created by
425.Fn bus_dma_tag_create .
426.Pp
427Returns
428.Er EBUSY
429if any DMA maps remain associated with
430.Fa dmat
431or
432.Ql 0
433on success.
434.It Fn bus_dmamap_create "dmat" "flags" "*mapp"
435Allocates and initializes a DMA map.
436Arguments are as follows:
437.Bl -tag -width nsegments -compact
438.It Fa dmat
439DMA tag.
440.It Fa flags
441The value of this argument is currently undefined and should be
442specified as
443.Ql 0 .
444.It Fa mapp
445Pointer to a
446.Vt bus_dmamap_t
447where the resulting DMA map will be stored.
448.El
449.Pp
450Returns
451.Er ENOMEM
452if sufficient memory is not available for creating the
453map or allocating mapping resources.
454.It Fn bus_dmamap_destroy "dmat" "map"
455Frees all resources associated with a given DMA map.
456Arguments are as follows:
457.Bl -tag -width dmat -compact
458.It Fa dmat
459DMA tag used to allocate
460.Fa map .
461.It Fa map
462The DMA map to destroy.
463.El
464.Pp
465Returns
466.Er EBUSY
467if a mapping is still active for
468.Fa map .
469.It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
470"callback_arg" "flags"
471Creates a mapping in device visible address space of
472.Fa buflen
473bytes of
474.Fa buf ,
475associated with the DMA map
476.Fa map.
477Arguments are as follows:
478.Bl -tag -width buflen -compact
479.It Fa dmat
480DMA tag used to allocate
481.Fa map.
482.It Fa map
483A DMA map without a currently active mapping.
484.It Fa buf
485A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
486mapped into device visible address space.
487.It Fa buflen
488The size of the buffer.
489.It Fa callback Fa callback_arg
490The callback function, and its argument.
491.It Fa flags
492The value of this argument is currently undefined, and should be
493specified as
494.Ql 0 .
495.El
496.Pp
497Return values to the caller are as follows:
498.Bl -tag -width EINPROGRESS -compact
499.It 0
500The callback has been called and completed.
501The status of the mapping has been delivered to the callback.
502.It Er EINPROGRESS
503The mapping has been deferred for lack of resources.
504The callback will be called as soon as resources are available.
505Callbacks are serviced in FIFO order.
506DMA maps created from DMA tags that are allocated with
507the
508.Dv BUS_DMA_ALLOCNOW
509flag will never return this status for a load operation.
510.It Er EINVAL
511The load request was invalid.
512The callback has not, and will not be called.
513This error value may indicate that
514.Fa dmat ,
515.Fa map ,
516.Fa buf ,
517or
518.Fa callback
519were invalid, or
520.Fa buslen
521was larger than the
522.Fa maxsize
523argument used to create the dma tag
524.Fa dmat .
525.El
526.Pp
527When the callback is called, it is presented with an error value
528indicating the disposition of the mapping.
529Error may be one of the following:
530.Bl -tag -width EINPROGRESS -compact
531.It 0
532The mapping was successful and the
533.Fa dm_segs
534callback argument contains an array of
535.Vt bus_dma_segment_t
536elements describing the mapping.
537This array is only valid during the scope of the callback function.
538.It Er EFBIG
539A mapping could not be achieved within the segment constraints provided
540in the tag even though the requested allocation size was less than maxsize.
541.El
542.It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
543"flags"
544This is a variation of
545.Fn bus_dmamap_load
546which maps mbuf chains
547for DMA transfers.
548A
549.Vt bus_size_t
550argument is also passed to the callback routine, which
551contains the mbuf chain's packet header length.
552.Pp
553Mbuf chains are assumed to be in kernel virtual address space.
554.Pp
555Returns
556.Er EINVAL
557if the size of the mbuf chain exceeds the maximum limit of the
558DMA tag.
559.It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
560This is a variation of
561.Fn bus_dmamap_load
562which maps buffers pointed to by
563.Fa uio
564for DMA transfers.
565A
566.Vt bus_size_t
567argument is also passed to the callback routine, which contains the size of
568.Fa uio ,
569i.e.
570.Fa uio->uio_resid .
571.Pp
572If
573.Fa uio->uio_segflg
574is
575.Dv UIO_USERSPACE ,
576then it is assumed that the buffer,
577.Fa uio
578is in
579.Fa "uio->uio_td->td_proc" Ns 's
580address space.
581User space memory must be in-core and wired prior to attempting a map
582load operation.
583Pages may be locked using
584.Xr vslock 9 .
585.It Fn bus_dmamap_unload "dmat" "map"
586Unloads a DMA map.
587Arguments are as follows:
588.Bl -tag -width dmam -compact
589.It Fa dmat
590DMA tag used to allocate
591.Fa map .
592.It Fa map
593The DMA map that is to be unloaded.
594.El
595.Pp
596.Fn bus_dmamap_unload
597will not perform any implicit synchronization of DMA buffers.
598This must be done explicitly by a call to
599.Fn bus_dmamap_sync
600prior to unloading the map.
601.It Fn bus_dmamap_sync "dmat" "map" "op"
602Performs synchronization of a device visible mapping with the CPU visible
603memory referenced by that mapping.
604Arguments are as follows:
605.Bl -tag -width dmat -compact
606.It Fa dmat
607DMA tag used to allocate
608.Fa map .
609.It Fa map
610The DMA mapping to be synchronized.
611.It Fa op
612Type of synchronization operation to perform.
613See the definition of
614.Vt bus_dmasync_op_t
615for a description of the acceptable values for
616.Fa op .
617.El
618.Pp
619.Fn bus_dmamap_sync
620is the method used to ensure that CPU and device DMA access to shared
621memory is coherent.
622For example, the CPU might be used to setup the contents of a buffer
623that is to be DMA'ed into a device.
624To ensure that the data are visible via the device's mapping of that
625memory, the buffer must be loaded and a dma sync operation of
626.Dv BUS_DMASYNC_PREREAD
627must be performed.
628Additional sync operations must be performed after every CPU write
629to this memory if additional DMA reads are to be performed.
630Conversely, for the DMA write case, the buffer must be loaded,
631and a dma sync operation of
632.Dv BUS_DMASYNC_PREWRITE
633must be performed.
634The CPU will only be able to see the results of this DMA write
635once the DMA has completed and a
636.Dv BUS_DMASYNC_POSTWRITE
637operation has been performed.
638.Pp
639If DMA read and write operations are not preceded and followed by the
640appropriate synchronization operations, behavior is undefined.
641.It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "mapp"
642Allocates memory that is mapped into KVA at the address returned
643in
644.Fa vaddr
645that is permanently loaded into the newly created
646.Vt bus_dmamap_t
647returned via
648.Fa mapp .
649Arguments are as follows:
650.Bl -tag -width alignment -compact
651.It Fa dmat
652DMA tag describing the constraints of the DMA mapping.
653.It Fa vaddr
654Pointer to a pointer that will hold the returned KVA mapping of
655the allocated region.
656.It Fa flags
657Flags are defined as follows:
658.Bl -tag -width BUS_DMA_NOWAIT -compact
659.It Dv BUS_DMA_WAITOK
660The routine can safely wait (sleep) for resources.
661.It Dv BUS_DMA_NOWAIT
662The routine is not allowed to wait for resources.
663If resources are not available,
664.Dv ENOMEM
665is returned.
666.It Dv BUS_DMA_COHERENT
667Attempt to map this memory such that cache sync operations are
668as cheap as possible.
669This flag is typically set on memory that will be accessed by both
670a CPU and a DMA engine, frequently.
671Use of this flag does not remove the requirement of using
672bus_dmamap_sync, but it may reduce the cost of performing
673these operations.
674.It Dv BUS_DMA_ZERO
675Causes the allocated memory to be set to all zeros.
676.El
677.It Fa mapp
678Pointer to storage for the returned DMA map.
679.El
680.Pp
681The size of memory to be allocated is
682.Fa maxsize
683as specified in
684.Fa dmat .
685.Pp
686The current implementation of
687.Fn bus_dmamem_alloc
688will allocate all requests as a single segment.
689.Pp
690Although no explicit loading is required to access the memory
691referenced by the returned map, the synchronization requirements
692as described in the
693.Fn bus_dmamap_sync
694section still apply.
695.Pp
696Returns
697.Er ENOMEM
698if sufficient memory is not available for completing
699the operation.
700.It Fn bus_dmamem_free "dmat" "*vaddr" "map"
701Frees memory previously allocated by
702.Fn bus_dmamem_alloc .
703Any mappings
704will be invalidated.
705Arguments are as follows:
706.Bl -tag -width vaddr -compact
707.It Fa dmat
708DMA tag.
709.It Fa vaddr
710Kernel virtual address of the memory.
711.It Fa map
712DMA map to be invalidated.
713.El
714.El
715.Sh RETURN VALUES
716Behavior is undefined if invalid arguments are passed to
717any of the above functions.
718If sufficient resources cannot be allocated for a given
719transaction,
720.Er ENOMEM
721is returned.
722All
723routines that are not of type,
724.Vt void ,
725will return 0 on success or an error
726code, as discussed above.
727.Pp
728All
729.Vt void
730routines will succeed if provided with valid arguments.
731.Sh SEE ALSO
732.Xr devclass 9 ,
733.Xr device 9 ,
734.Xr driver 9 ,
735.Xr rman 9 ,
736.Xr vslock 9
737.Pp
738.Rs
739.%A "Jason R. Thorpe"
740.%T "A Machine-Independent DMA Framework for NetBSD"
741.%J "Proceedings of the Summer 1998 USENIX Technical Conference"
742.%Q "USENIX Association"
743.%D "June 1998"
744.Re
745.Sh HISTORY
746The
747.Nm
748interface first appeared in
749.Nx 1.3 .
750.Pp
751The
752.Nm
753API was adopted from
754.Nx
755for use in the CAM SCSI subsystem.
756The alterations to the original API were aimed to remove the need for
757a
758.Vt bus_dma_segment_t
759array stored in each
760.Vt bus_dmamap_t
761while allowing callers to queue up on scarce resources.
762.Sh AUTHORS
763The
764.Nm
765interface was designed and implemented by
766.An Jason R. Thorpe
767of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
768Additional input on the
769.Nm
770design was provided by
771.An -nosplit
772.An Chris Demetriou ,
773.An Charles Hannum ,
774.An Ross Harvey ,
775.An Matthew Jacob ,
776.An Jonathan Stone ,
777and
778.An Matt Thomas .
779.Pp
780The
781.Nm
782interface in
783.Fx
784benefits from the contributions of
785.An Justin T. Gibbs ,
786.An Peter Wemm ,
787.An Doug Rabson ,
788.An Matthew N. Dodd ,
789.An Sam Leffler ,
790.An Maxime Henrion ,
791.An Jake Burkholder ,
792.An Takahashi Yoshihiro ,
793.An Scott Long
794and many others.
795.Pp
796This manual page was written by
797.An Hiten M. Pandya
798and
799.An Justin T. Gibbs .
800