xref: /freebsd/share/man/man9/bus_dma.9 (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
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 March 6, 2007
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_mbuf_sg ,
75.Nm bus_dmamap_load_uio ,
76.Nm bus_dmamap_unload ,
77.Nm bus_dmamap_sync ,
78.Nm bus_dmamem_alloc ,
79.Nm bus_dmamem_free
80.Nd Bus and Machine Independent DMA Mapping Interface
81.Sh SYNOPSIS
82.In machine/bus.h
83.Ft int
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"
89.Ft int
90.Fn bus_dma_tag_destroy "bus_dma_tag_t dmat"
91.Ft int
92.Fn bus_dmamap_create "bus_dma_tag_t dmat" "int flags" "bus_dmamap_t *mapp"
93.Ft int
94.Fn bus_dmamap_destroy "bus_dma_tag_t dmat" "bus_dmamap_t map"
95.Ft int
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" \
98"int flags"
99.Ft int
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" \
102"int flags"
103.Ft int
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"
106.Ft int
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" \
109"int flags"
110.Ft void
111.Fn bus_dmamap_unload "bus_dma_tag_t dmat" "bus_dmamap_t map"
112.Ft void
113.Fn bus_dmamap_sync "bus_dma_tag_t dmat" "bus_dmamap_t map" \
114"op"
115.Ft int
116.Fn bus_dmamem_alloc "bus_dma_tag_t dmat" "void **vaddr" \
117"int flags" "bus_dmamap_t *mapp"
118.Ft void
119.Fn bus_dmamem_free "bus_dma_tag_t dmat" "void *vaddr" \
120"bus_dmamap_t map"
121.Sh DESCRIPTION
122Direct Memory Access (DMA) is a method of transferring data
123without involving the CPU, thus providing higher performance.
124A DMA transaction can be achieved between device to memory,
125device to device, or memory to memory.
126.Pp
127The
128.Nm
129API is a bus, device, and machine-independent (MI) interface to
130DMA mechanisms.
131It provides the client with flexibility and simplicity by
132abstracting machine dependent issues like setting up
133DMA mappings, handling cache issues, bus specific features
134and limitations.
135.Sh STRUCTURES AND TYPES
136.Bl -tag -width indent
137.It Vt bus_dma_tag_t
138A machine-dependent (MD) opaque type that describes the
139characteristics of DMA transactions.
140DMA tags are organized into a hierarchy, with each child
141tag inheriting the restrictions of its parent.
142This allows all devices along the path of DMA transactions
143to contribute to the constraints of those transactions.
144.It Vt bus_dma_filter_t
145Client specified address filter having the format:
146.Bl -tag -width indent
147.It Ft int
148.Fn "client_filter" "void *filtarg" "bus_addr_t testaddr"
149.El
150.Pp
151Address filters can be specified during tag creation to allow
152for devices whose DMA address restrictions cannot be specified
153by a single window.
154The
155.Fa filtarg
156argument is specified by the client during tag creation to be passed to all
157invocations of the callback.
158The
159.Fa testaddr
160argument contains a potential starting address of a DMA mapping.
161The filter function operates on the set of addresses from
162.Fa testaddr
163to
164.Ql trunc_page(testaddr) + PAGE_SIZE - 1 ,
165inclusive.
166The filter function should return zero if any mapping in this range
167can be accommodated by the device and non-zero otherwise.
168.It Vt bus_dma_segment_t
169A machine-dependent type that describes individual
170DMA segments.
171It contains the following fields:
172.Bd -literal
173	bus_addr_t	ds_addr;
174	bus_size_t	ds_len;
175.Ed
176.Pp
177The
178.Fa ds_addr
179field contains the device visible address of the DMA segment, and
180.Fa ds_len
181contains the length of the DMA segment.
182Although the DMA segments returned by a mapping call will adhere to
183all 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
185almost always required when presenting segment information to the device.
186.It Vt bus_dmamap_t
187A machine-dependent opaque type describing an individual mapping.
188One map is used for each memory allocation that will be loaded.
189Maps can be reused once they have been unloaded.
190Multiple maps can be associated with one DMA tag.
191While the value of the map may evaluate to
192.Dv NULL
193on some platforms under certain conditions,
194it should never be assumed that it will be
195.Dv NULL
196in all cases.
197.It Vt bus_dmamap_callback_t
198Client specified callback for receiving mapping information resulting from
199the load of a
200.Vt bus_dmamap_t
201via
202.Fn bus_dmamap_load .
203Callbacks are of the format:
204.Bl -tag -width indent
205.It Ft void
206.Fn "client_callback" "void *callback_arg" "bus_dma_segment_t *segs" \
207"int nseg" "int error"
208.El
209.Pp
210The
211.Fa callback_arg
212is the callback argument passed to dmamap load functions.
213The
214.Fa segs
215and
216.Fa nseg
217arguments describe an array of
218.Vt bus_dma_segment_t
219structures that represent the mapping.
220This array is only valid within the scope of the callback function.
221The success or failure of the mapping is indicated by the
222.Fa error
223argument.
224More information on the use of callbacks can be found in the
225description of the individual dmamap load functions.
226.It Vt bus_dmamap_callback2_t
227Client specified callback for receiving mapping information resulting from
228the load of a
229.Vt bus_dmamap_t
230via
231.Fn bus_dmamap_load_uio
232or
233.Fn bus_dmamap_load_mbuf .
234.Pp
235Callback2s are of the format:
236.Bl -tag -width indent
237.It Ft void
238.Fn "client_callback2" "void *callback_arg" "bus_dma_segment_t *segs" \
239"int nseg" "bus_size_t mapsize" "int error"
240.El
241.Pp
242Callback2's behavior is the same as
243.Vt bus_dmamap_callback_t
244with the addition that the length of the data mapped is provided via
245.Fa mapsize .
246.It Vt bus_dmasync_op_t
247Memory synchronization operation specifier.
248Bus DMA requires explicit synchronization of memory with its device
249visible mapping in order to guarantee memory coherency.
250The
251.Vt bus_dmasync_op_t
252allows the type of DMA operation that will be or has been performed
253to be communicated to the system so that the correct coherency measures
254are taken.
255The operations are represented as bitfield flags that can be combined together,
256though it only makes sense to combine PRE flags or POST flags, not both.
257See the
258.Fn bus_dmamap_sync
259description below for more details on how to use these operations.
260.Pp
261All operations specified below are performed from the host memory point of view,
262where a read implies data coming from the device to the host memory, and a write
263implies data going from the host memory to the device.
264Alternatively, the operations can be thought of in terms of driver operations,
265where reading a network packet or storage sector corresponds to a read operation
266in
267.Nm .
268.Bl -tag -width ".Dv BUS_DMASYNC_POSTWRITE"
269.It Dv BUS_DMASYNC_PREREAD
270Perform any synchronization required prior to an update of host memory by the
271device.
272.It Dv BUS_DMASYNC_PREWRITE
273Perform any synchronization required after an update of host memory by the CPU
274and prior to device access to host memory.
275.It Dv BUS_DMASYNC_POSTREAD
276Perform any synchronization required after an update of host memory by the
277device and prior to CPU access to host memory.
278.It Dv BUS_DMASYNC_POSTWRITE
279Perform any synchronization required after device access to host memory.
280.El
281.It Vt bus_dma_lock_t
282Client specified lock/mutex manipulation method.
283This will be called from
284within busdma whenever a client lock needs to be manipulated.
285In its current form, the function will be called immediately before
286the callback for a dma load operation that has been deferred with
287.Dv BUS_DMA_LOCK
288and immediately after with
289.Dv BUS_DMA_UNLOCK .
290If the load operation does not need to be deferred, then it
291will not be called since the function loading the map should
292be holding the appropriate locks.
293This method is of the format:
294.Bl -tag -width indent
295.It Ft void
296.Fn "lockfunc" "void *lockfunc_arg" "bus_dma_lock_op_t op"
297.El
298.Pp
299The
300.Fa lockfuncarg
301argument is specified by the client during tag creation to be passed to all
302invocations of the callback.
303The
304.Fa op
305argument specifies the lock operation to perform.
306.Pp
307Two
308.Vt lockfunc
309implementations are provided for convenience.
310.Fn busdma_lock_mutex
311performs standard mutex operations on the sleep mutex provided via
312.Fa lockfuncarg .
313.Fn dflt_lock
314will generate a system panic if it is called.
315It is substituted into the tag when
316.Fa lockfunc
317is passed as
318.Dv NULL
319to
320.Fn bus_dma_tag_create
321and is useful for tags that should not be used with deferred load operations.
322.It Vt bus_dma_lock_op_t
323Operations to be performed by the client-specified
324.Fn lockfunc .
325.Bl -tag -width ".Dv BUS_DMA_UNLOCK"
326.It Dv BUS_DMA_LOCK
327Acquires and/or locks the client locking primitive.
328.It Dv BUS_DMA_UNLOCK
329Releases and/or unlocks the client locking primitive.
330.El
331.El
332.Sh FUNCTIONS
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"
337Allocates a device specific DMA tag, and initializes it according to
338the arguments provided:
339.Bl -tag -width ".Fa filtfuncarg"
340.It Fa parent
341Indicates restrictions between the parent bridge, CPU memory, and the
342device.
343Each device must use a master parent tag by calling
344.Fn bus_get_dma_tag .
345.It Fa alignment
346Alignment constraint, in bytes, of any mappings created using this tag.
347The alignment must be a power of 2.
348Hardware that can DMA starting at any address would specify
349.Em 1
350for byte alignment.
351Hardware requiring DMA transfers to start on a multiple of 4K
352would specify
353.Em 4096 .
354.It Fa boundary
355Boundary constraint, in bytes, of the target DMA memory region.
356The boundary indicates the set of addresses, all multiples of the
357boundary argument, that cannot be crossed by a single
358.Vt bus_dma_segment_t .
359The boundary must be a power of 2 and must be no smaller than the
360maximum segment size.
361.Ql 0
362indicates that there are no boundary restrictions.
363.It Fa lowaddr , highaddr
364Bounds of the window of bus address space that
365.Em cannot
366be directly accessed by the device.
367The window contains all addresses greater than lowaddr and
368less than or equal to highaddr.
369For example, a device incapable of DMA above 4GB, would specify
370a highaddr of
371.Dv BUS_SPACE_MAXADDR
372and a lowaddr of
373.Dv BUS_SPACE_MAXADDR_32BIT .
374Similarly a device that can only dma to addresses bellow 16MB would
375specify a highaddr of
376.Dv BUS_SPACE_MAXADDR
377and a lowaddr of
378.Dv BUS_SPACE_MAXADDR_24BIT .
379Some implementations requires that some region of device visible
380address space, overlapping available host memory, be outside the
381window.
382This area of
383.Ql safe memory
384is used to bounce requests that would otherwise conflict with
385the exclusion window.
386.It Fa filtfunc
387Optional filter function (may be
388.Dv NULL )
389to be called for any attempt to
390map memory into the window described by
391.Fa lowaddr
392and
393.Fa highaddr .
394A filter function is only required when the single window described
395by
396.Fa lowaddr
397and
398.Fa highaddr
399cannot adequately describe the constraints of the device.
400The filter function will be called for every machine page
401that overlaps the exclusion window.
402.It Fa filtfuncarg
403Argument passed to all calls to the filter function for this tag.
404May be
405.Dv NULL .
406.It Fa maxsize
407Maximum size, in bytes, of the sum of all segment lengths in a given
408DMA mapping associated with this tag.
409.It Fa nsegments
410Number of discontinuities (scatter/gather segments) allowed
411in a DMA mapped region.
412If there is no restriction,
413.Dv BUS_SPACE_UNRESTRICTED
414may be specified.
415.It Fa maxsegsz
416Maximum size, in bytes, of a segment in any DMA mapped region associated
417with
418.Fa dmat .
419.It Fa flags
420Are as follows:
421.Bl -tag -width ".Dv BUS_DMA_ALLOCNOW"
422.It Dv BUS_DMA_ALLOCNOW
423Pre-allocate enough resources to handle at least one map load operation on
424this tag.
425If sufficient resources are not available,
426.Er ENOMEM
427is returned.
428This should not be used for tags that only describe buffers that will be
429allocated with
430.Fn bus_dmamem_alloc .
431Also, due to resource sharing with other tags, this flag does not guarantee
432that resources will be allocated or reserved exclusively for this tag.
433It should be treated only as a minor optimization.
434.El
435.It Fa lockfunc
436Optional lock manipulation function (may be
437.Dv NULL )
438to be called when busdma
439needs to manipulate a lock on behalf of the client.
440If
441.Dv NULL
442is specified,
443.Fn dflt_lock
444is used.
445.It Fa lockfuncarg
446Optional argument to be passed to the function specified by
447.Fa lockfunc .
448.It Fa dmat
449Pointer to a bus_dma_tag_t where the resulting DMA tag will
450be stored.
451.El
452.Pp
453Returns
454.Er ENOMEM
455if sufficient memory is not available for tag creation
456or allocating mapping resources.
457.It Fn bus_dma_tag_destroy "dmat"
458Deallocate the DMA tag
459.Fa dmat
460that was created by
461.Fn bus_dma_tag_create .
462.Pp
463Returns
464.Er EBUSY
465if any DMA maps remain associated with
466.Fa dmat
467or
468.Ql 0
469on success.
470.It Fn bus_dmamap_create "dmat" "flags" "*mapp"
471Allocates and initializes a DMA map.
472Arguments are as follows:
473.Bl -tag -width ".Fa nsegments"
474.It Fa dmat
475DMA tag.
476.It Fa flags
477The value of this argument is currently undefined and should be
478specified as
479.Ql 0 .
480.It Fa mapp
481Pointer to a
482.Vt bus_dmamap_t
483where the resulting DMA map will be stored.
484.El
485.Pp
486Returns
487.Er ENOMEM
488if sufficient memory is not available for creating the
489map or allocating mapping resources.
490.It Fn bus_dmamap_destroy "dmat" "map"
491Frees all resources associated with a given DMA map.
492Arguments are as follows:
493.Bl -tag -width ".Fa dmat"
494.It Fa dmat
495DMA tag used to allocate
496.Fa map .
497.It Fa map
498The DMA map to destroy.
499.El
500.Pp
501Returns
502.Er EBUSY
503if a mapping is still active for
504.Fa map .
505.It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
506"callback_arg" "flags"
507Creates a mapping in device visible address space of
508.Fa buflen
509bytes of
510.Fa buf ,
511associated with the DMA map
512.Fa map .
513This call will always return immediately and will not block for any reason.
514Arguments are as follows:
515.Bl -tag -width ".Fa buflen"
516.It Fa dmat
517DMA tag used to allocate
518.Fa map .
519.It Fa map
520A DMA map without a currently active mapping.
521.It Fa buf
522A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
523mapped into device visible address space.
524.It Fa buflen
525The size of the buffer.
526.It Fa callback Fa callback_arg
527The callback function, and its argument.
528This function is called once sufficient mapping resources are available for
529the DMA operation.
530If resources are temporarily unavailable, this function will be deferred until
531later, but the load operation will still return immediately to the caller.
532Thus, callers should not assume that the callback will be called before the
533load returns, and code should be structured appropriately to handle this.
534See below for specific flags and error codes that control this behavior.
535.It Fa flags
536Are as follows:
537.Bl -tag -width ".Dv BUS_DMA_NOWAIT"
538.It Dv BUS_DMA_NOWAIT
539The load should not be deferred in case of insufficient mapping resources,
540and instead should return immediately with an appropriate error.
541.El
542.El
543.Pp
544Return values to the caller are as follows:
545.Bl -tag -width ".Er EINPROGRESS"
546.It 0
547The callback has been called and completed.
548The status of the mapping has been delivered to the callback.
549.It Er EINPROGRESS
550The mapping has been deferred for lack of resources.
551The callback will be called as soon as resources are available.
552Callbacks are serviced in FIFO order.
553To ensure that ordering is guaranteed, all subsequent load requests will also
554be deferred until all callbacks have been processed.
555.It Er ENOMEM
556The load request has failed due to insufficient resources, and the caller
557specifically used the
558.Dv BUS_DMA_NOWAIT
559flag.
560.It Er EINVAL
561The load request was invalid.
562The callback has been called and has been provided the same error.
563This error value may indicate that
564.Fa dmat ,
565.Fa map ,
566.Fa buf ,
567or
568.Fa callback
569were invalid, or
570.Fa buflen
571was larger than the
572.Fa maxsize
573argument used to create the dma tag
574.Fa dmat .
575.El
576.Pp
577When the callback is called, it is presented with an error value
578indicating the disposition of the mapping.
579Error may be one of the following:
580.Bl -tag -width ".Er EINPROGRESS"
581.It 0
582The mapping was successful and the
583.Fa dm_segs
584callback argument contains an array of
585.Vt bus_dma_segment_t
586elements describing the mapping.
587This array is only valid during the scope of the callback function.
588.It Er EFBIG
589A mapping could not be achieved within the segment constraints provided
590in the tag even though the requested allocation size was less than maxsize.
591.El
592.It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
593"flags"
594This is a variation of
595.Fn bus_dmamap_load
596which maps mbuf chains
597for DMA transfers.
598A
599.Vt bus_size_t
600argument is also passed to the callback routine, which
601contains the mbuf chain's packet header length.
602The
603.Dv BUS_DMA_NOWAIT
604flag is implied, thus no callback deferral will happen.
605.Pp
606Mbuf chains are assumed to be in kernel virtual address space.
607.Pp
608Beside the error values listed for
609.Fn bus_dmamap_load ,
610.Er EINVAL
611will be returned if the size of the mbuf chain exceeds the maximum limit of the
612DMA tag.
613.It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
614This is just like
615.Fn bus_dmamap_load_mbuf
616except that it returns immediately without calling a callback function.
617It is provided for efficiency.
618The scatter/gather segment array
619.Va segs
620is provided by the caller and filled in directly by the function.
621The
622.Va nsegs
623argument is returned with the number of segments filled in.
624Returns the same errors as
625.Fn bus_dmamap_load_mbuf .
626.It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
627This is a variation of
628.Fn bus_dmamap_load
629which maps buffers pointed to by
630.Fa uio
631for DMA transfers.
632A
633.Vt bus_size_t
634argument is also passed to the callback routine, which contains the size of
635.Fa uio ,
636i.e.
637.Fa uio->uio_resid .
638The
639.Dv BUS_DMA_NOWAIT
640flag is implied, thus no callback deferral will happen.
641Returns the same errors as
642.Fn bus_dmamap_load .
643.Pp
644If
645.Fa uio->uio_segflg
646is
647.Dv UIO_USERSPACE ,
648then it is assumed that the buffer,
649.Fa uio
650is in
651.Fa "uio->uio_td->td_proc" Ns 's
652address space.
653User space memory must be in-core and wired prior to attempting a map
654load operation.
655Pages may be locked using
656.Xr vslock 9 .
657.It Fn bus_dmamap_unload "dmat" "map"
658Unloads a DMA map.
659Arguments are as follows:
660.Bl -tag -width ".Fa dmam"
661.It Fa dmat
662DMA tag used to allocate
663.Fa map .
664.It Fa map
665The DMA map that is to be unloaded.
666.El
667.Pp
668.Fn bus_dmamap_unload
669will not perform any implicit synchronization of DMA buffers.
670This must be done explicitly by a call to
671.Fn bus_dmamap_sync
672prior to unloading the map.
673.It Fn bus_dmamap_sync "dmat" "map" "op"
674Performs synchronization of a device visible mapping with the CPU visible
675memory referenced by that mapping.
676Arguments are as follows:
677.Bl -tag -width ".Fa dmat"
678.It Fa dmat
679DMA tag used to allocate
680.Fa map .
681.It Fa map
682The DMA mapping to be synchronized.
683.It Fa op
684Type of synchronization operation to perform.
685See the definition of
686.Vt bus_dmasync_op_t
687for a description of the acceptable values for
688.Fa op .
689.El
690.Pp
691The
692.Fn bus_dmamap_sync
693function
694is the method used to ensure that CPU's and device's direct
695memory access (DMA) to shared
696memory is coherent.
697For example, the CPU might be used to set up the contents of a buffer
698that is to be made available to a device.
699To ensure that the data are visible via the device's mapping of that
700memory, the buffer must be loaded and a DMA sync operation of
701.Dv BUS_DMASYNC_PREWRITE
702must be performed after the CPU has updated the buffer and before the device
703access is initiated.
704If the CPU modifies this buffer again later, another
705.Dv BUS_DMASYNC_PREWRITE
706sync operation must be performed before an additional device
707access.
708Conversely, suppose a device updates memory that is to be read by a CPU.
709In this case, the buffer must be loaded, and a DMA sync operation of
710.Dv BUS_DMASYNC_PREREAD
711must be performed before the device access is initiated.
712The CPU will only be able to see the results of this memory update
713once the DMA operation has completed and a
714.Dv BUS_DMASYNC_POSTREAD
715sync operation has been performed.
716.Pp
717If read and write operations are not preceded and followed by the
718appropriate synchronization operations, behavior is undefined.
719.It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
720Allocates memory that is mapped into KVA at the address returned
721in
722.Fa vaddr
723and that is permanently loaded into the newly created
724.Vt bus_dmamap_t
725returned via
726.Fa mapp .
727Arguments are as follows:
728.Bl -tag -width ".Fa alignment"
729.It Fa dmat
730DMA tag describing the constraints of the DMA mapping.
731.It Fa vaddr
732Pointer to a pointer that will hold the returned KVA mapping of
733the allocated region.
734.It Fa flags
735Flags are defined as follows:
736.Bl -tag -width ".Dv BUS_DMA_NOWAIT"
737.It Dv BUS_DMA_WAITOK
738The routine can safely wait (sleep) for resources.
739.It Dv BUS_DMA_NOWAIT
740The routine is not allowed to wait for resources.
741If resources are not available,
742.Dv ENOMEM
743is returned.
744.It Dv BUS_DMA_COHERENT
745Attempt to map this memory such that cache sync operations are
746as cheap as possible.
747This flag is typically set on memory that will be accessed by both
748a CPU and a DMA engine, frequently.
749Use of this flag does not remove the requirement of using
750bus_dmamap_sync, but it may reduce the cost of performing
751these operations.
752The
753.Dv BUS_DMA_COHERENT
754flag is currently implemented on sparc64 and arm.
755.It Dv BUS_DMA_ZERO
756Causes the allocated memory to be set to all zeros.
757.El
758.It Fa mapp
759Pointer to a
760.Vt bus_dmamap_t
761where the resulting DMA map will be stored.
762.El
763.Pp
764The size of memory to be allocated is
765.Fa maxsize
766as specified in the call to
767.Fn bus_dma_tag_create
768for
769.Fa dmat .
770.Pp
771The current implementation of
772.Fn bus_dmamem_alloc
773will allocate all requests as a single segment.
774.Pp
775An initial load operation is required to obtain the bus address of the allocated
776memory, and an unload operation is required before freeing the memory, as
777described below in
778.Fn bus_dmamem_free .
779Maps are automatically handled by this function and should not be explicitly
780allocated or destroyed.
781.Pp
782Although an explicit load is not required for each access to the memory
783referenced by the returned map, the synchronization requirements
784as described in the
785.Fn bus_dmamap_sync
786section still apply and should be used to achieve portability on architectures
787without coherent buses.
788.Pp
789Returns
790.Er ENOMEM
791if sufficient memory is not available for completing
792the operation.
793.It Fn bus_dmamem_free "dmat" "*vaddr" "map"
794Frees memory previously allocated by
795.Fn bus_dmamem_alloc .
796Any mappings
797will be invalidated.
798Arguments are as follows:
799.Bl -tag -width ".Fa vaddr"
800.It Fa dmat
801DMA tag.
802.It Fa vaddr
803Kernel virtual address of the memory.
804.It Fa map
805DMA map to be invalidated.
806.El
807.El
808.Sh RETURN VALUES
809Behavior is undefined if invalid arguments are passed to
810any of the above functions.
811If sufficient resources cannot be allocated for a given
812transaction,
813.Er ENOMEM
814is returned.
815All
816routines that are not of type
817.Vt void
818will return 0 on success or an error
819code on failure as discussed above.
820.Pp
821All
822.Vt void
823routines will succeed if provided with valid arguments.
824.Sh LOCKING
825Two locking protocols are used by
826.Nm .
827The first is a private global lock that is used to synchronize access to the
828bounce buffer pool on the architectures that make use of them.
829This lock is strictly a leaf lock that is only used internally to
830.Nm
831and is not exposed to clients of the API.
832.Pp
833The second protocol involves protecting various resources stored in the tag.
834Since almost all
835.Nm
836operations are done through requests from the driver that created the tag,
837the most efficient way to protect the tag resources is through the lock that
838the driver uses.
839In cases where
840.Nm
841acts on its own without being called by the driver, the lock primitive
842specified in the tag is acquired and released automatically.
843An example of this is when the
844.Fn bus_dmamap_load
845callback function is called from a deferred context instead of the driver
846context.
847This means that certain
848.Nm
849functions must always be called with the same lock held that is specified in the
850tag.
851These functions include:
852.Pp
853.Bl -item -offset indent -compact
854.It
855.Fn bus_dmamap_load
856.It
857.Fn bus_dmamap_load_uio
858.It
859.Fn bus_dmamap_load_mbuf
860.It
861.Fn bus_dmamap_load_mbuf_sg
862.It
863.Fn bus_dmamap_unload
864.It
865.Fn bus_dmamap_sync
866.El
867.Pp
868There is one exception to this rule.
869It is common practice to call some of these functions during driver start-up
870without any locks held.
871So long as there is a guarantee of no possible concurrent use of the tag by
872different threads during this operation, it is safe to not hold a lock for
873these functions.
874.Pp
875Certain
876.Nm
877operations should not be called with the driver lock held, either because
878they are already protected by an internal lock, or because they might sleep
879due to memory or resource allocation.
880The following functions must not be
881called with any non-sleepable locks held:
882.Pp
883.Bl -item -offset indent -compact
884.It
885.Fn bus_dma_tag_create
886.It
887.Fn bus_dmamap_create
888.It
889.Fn bus_dmamem_alloc
890.El
891.Pp
892All other functions do not have a locking protocol and can thus be
893called with or without any system or driver locks held.
894.Sh SEE ALSO
895.Xr devclass 9 ,
896.Xr device 9 ,
897.Xr driver 9 ,
898.Xr rman 9 ,
899.Xr vslock 9
900.Pp
901.Rs
902.%A "Jason R. Thorpe"
903.%T "A Machine-Independent DMA Framework for NetBSD"
904.%J "Proceedings of the Summer 1998 USENIX Technical Conference"
905.%Q "USENIX Association"
906.%D "June 1998"
907.Re
908.Sh HISTORY
909The
910.Nm
911interface first appeared in
912.Nx 1.3 .
913.Pp
914The
915.Nm
916API was adopted from
917.Nx
918for use in the CAM SCSI subsystem.
919The alterations to the original API were aimed to remove the need for
920a
921.Vt bus_dma_segment_t
922array stored in each
923.Vt bus_dmamap_t
924while allowing callers to queue up on scarce resources.
925.Sh AUTHORS
926The
927.Nm
928interface was designed and implemented by
929.An Jason R. Thorpe
930of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
931Additional input on the
932.Nm
933design was provided by
934.An -nosplit
935.An Chris Demetriou ,
936.An Charles Hannum ,
937.An Ross Harvey ,
938.An Matthew Jacob ,
939.An Jonathan Stone ,
940and
941.An Matt Thomas .
942.Pp
943The
944.Nm
945interface in
946.Fx
947benefits from the contributions of
948.An Justin T. Gibbs ,
949.An Peter Wemm ,
950.An Doug Rabson ,
951.An Matthew N. Dodd ,
952.An Sam Leffler ,
953.An Maxime Henrion ,
954.An Jake Burkholder ,
955.An Takahashi Yoshihiro ,
956.An Scott Long
957and many others.
958.Pp
959This manual page was written by
960.An Hiten M. Pandya
961and
962.An Justin T. Gibbs .
963