xref: /freebsd/share/man/man9/bus_dma.9 (revision 7750ad47a9a7dbc83f87158464170c8640723293)
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 12, 2009
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_addr_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
368.Fa lowaddr
369and less than or equal to
370.Fa highaddr .
371For example, a device incapable of DMA above 4GB, would specify a
372.Fa highaddr
373of
374.Dv BUS_SPACE_MAXADDR
375and a
376.Fa lowaddr
377of
378.Dv BUS_SPACE_MAXADDR_32BIT .
379Similarly a device that can only perform DMA to addresses below
38016MB would specify a
381.Fa highaddr
382of
383.Dv BUS_SPACE_MAXADDR
384and a
385.Fa lowaddr
386of
387.Dv BUS_SPACE_MAXADDR_24BIT .
388Some implementations requires that some region of device visible
389address space, overlapping available host memory, be outside the
390window.
391This area of
392.Ql safe memory
393is used to bounce requests that would otherwise conflict with
394the exclusion window.
395.It Fa filtfunc
396Optional filter function (may be
397.Dv NULL )
398to be called for any attempt to
399map memory into the window described by
400.Fa lowaddr
401and
402.Fa highaddr .
403A filter function is only required when the single window described
404by
405.Fa lowaddr
406and
407.Fa highaddr
408cannot adequately describe the constraints of the device.
409The filter function will be called for every machine page
410that overlaps the exclusion window.
411.It Fa filtfuncarg
412Argument passed to all calls to the filter function for this tag.
413May be
414.Dv NULL .
415.It Fa maxsize
416Maximum size, in bytes, of the sum of all segment lengths in a given
417DMA mapping associated with this tag.
418.It Fa nsegments
419Number of discontinuities (scatter/gather segments) allowed
420in a DMA mapped region.
421If there is no restriction,
422.Dv BUS_SPACE_UNRESTRICTED
423may be specified.
424.It Fa maxsegsz
425Maximum size, in bytes, of a segment in any DMA mapped region associated
426with
427.Fa dmat .
428.It Fa flags
429Are as follows:
430.Bl -tag -width ".Dv BUS_DMA_ALLOCNOW"
431.It Dv BUS_DMA_ALLOCNOW
432Pre-allocate enough resources to handle at least one map load operation on
433this tag.
434If sufficient resources are not available,
435.Er ENOMEM
436is returned.
437This should not be used for tags that only describe buffers that will be
438allocated with
439.Fn bus_dmamem_alloc .
440Also, due to resource sharing with other tags, this flag does not guarantee
441that resources will be allocated or reserved exclusively for this tag.
442It should be treated only as a minor optimization.
443.El
444.It Fa lockfunc
445Optional lock manipulation function (may be
446.Dv NULL )
447to be called when busdma
448needs to manipulate a lock on behalf of the client.
449If
450.Dv NULL
451is specified,
452.Fn dflt_lock
453is used.
454.It Fa lockfuncarg
455Optional argument to be passed to the function specified by
456.Fa lockfunc .
457.It Fa dmat
458Pointer to a bus_dma_tag_t where the resulting DMA tag will
459be stored.
460.El
461.Pp
462Returns
463.Er ENOMEM
464if sufficient memory is not available for tag creation
465or allocating mapping resources.
466.It Fn bus_dma_tag_destroy "dmat"
467Deallocate the DMA tag
468.Fa dmat
469that was created by
470.Fn bus_dma_tag_create .
471.Pp
472Returns
473.Er EBUSY
474if any DMA maps remain associated with
475.Fa dmat
476or
477.Ql 0
478on success.
479.It Fn bus_dmamap_create "dmat" "flags" "*mapp"
480Allocates and initializes a DMA map.
481Arguments are as follows:
482.Bl -tag -width ".Fa nsegments"
483.It Fa dmat
484DMA tag.
485.It Fa flags
486Are as follows:
487.Bl -tag -width ".Dv BUS_DMA_COHERENT"
488.It Dv BUS_DMA_COHERENT
489Attempt to map the memory loaded with this map such that cache sync
490operations are as cheap as possible.
491This flag is typically set on maps when the memory loaded with these will
492be accessed by both a CPU and a DMA engine, frequently such as control data
493and as opposed to streamable data such as receive and transmit buffers.
494Use of this flag does not remove the requirement of using
495.Fn bus_dmamap_sync ,
496but it may reduce the cost of performing these operations.
497For
498.Fn bus_dmamap_create ,
499the
500.Dv BUS_DMA_COHERENT
501flag is currently implemented on sparc64.
502.El
503.It Fa mapp
504Pointer to a
505.Vt bus_dmamap_t
506where the resulting DMA map will be stored.
507.El
508.Pp
509Returns
510.Er ENOMEM
511if sufficient memory is not available for creating the
512map or allocating mapping resources.
513.It Fn bus_dmamap_destroy "dmat" "map"
514Frees all resources associated with a given DMA map.
515Arguments are as follows:
516.Bl -tag -width ".Fa dmat"
517.It Fa dmat
518DMA tag used to allocate
519.Fa map .
520.It Fa map
521The DMA map to destroy.
522.El
523.Pp
524Returns
525.Er EBUSY
526if a mapping is still active for
527.Fa map .
528.It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
529"callback_arg" "flags"
530Creates a mapping in device visible address space of
531.Fa buflen
532bytes of
533.Fa buf ,
534associated with the DMA map
535.Fa map .
536This call will always return immediately and will not block for any reason.
537Arguments are as follows:
538.Bl -tag -width ".Fa buflen"
539.It Fa dmat
540DMA tag used to allocate
541.Fa map .
542.It Fa map
543A DMA map without a currently active mapping.
544.It Fa buf
545A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
546mapped into device visible address space.
547.It Fa buflen
548The size of the buffer.
549.It Fa callback Fa callback_arg
550The callback function, and its argument.
551This function is called once sufficient mapping resources are available for
552the DMA operation.
553If resources are temporarily unavailable, this function will be deferred until
554later, but the load operation will still return immediately to the caller.
555Thus, callers should not assume that the callback will be called before the
556load returns, and code should be structured appropriately to handle this.
557See below for specific flags and error codes that control this behavior.
558.It Fa flags
559Are as follows:
560.Bl -tag -width ".Dv BUS_DMA_NOWAIT"
561.It Dv BUS_DMA_NOWAIT
562The load should not be deferred in case of insufficient mapping resources,
563and instead should return immediately with an appropriate error.
564.It Dv BUS_DMA_NOCACHE
565The generated transactions to and from the virtual page are non-cacheable.
566For
567.Fn bus_dmamap_load ,
568the
569.Dv BUS_DMA_NOCACHE
570flag is currently implemented on sparc64.
571.El
572.El
573.Pp
574Return values to the caller are as follows:
575.Bl -tag -width ".Er EINPROGRESS"
576.It 0
577The callback has been called and completed.
578The status of the mapping has been delivered to the callback.
579.It Er EINPROGRESS
580The mapping has been deferred for lack of resources.
581The callback will be called as soon as resources are available.
582Callbacks are serviced in FIFO order.
583To ensure that ordering is guaranteed, all subsequent load requests will also
584be deferred until all callbacks have been processed.
585.It Er ENOMEM
586The load request has failed due to insufficient resources, and the caller
587specifically used the
588.Dv BUS_DMA_NOWAIT
589flag.
590.It Er EINVAL
591The load request was invalid.
592The callback has been called and has been provided the same error.
593This error value may indicate that
594.Fa dmat ,
595.Fa map ,
596.Fa buf ,
597or
598.Fa callback
599were invalid, or
600.Fa buflen
601was larger than the
602.Fa maxsize
603argument used to create the dma tag
604.Fa dmat .
605.El
606.Pp
607When the callback is called, it is presented with an error value
608indicating the disposition of the mapping.
609Error may be one of the following:
610.Bl -tag -width ".Er EINPROGRESS"
611.It 0
612The mapping was successful and the
613.Fa dm_segs
614callback argument contains an array of
615.Vt bus_dma_segment_t
616elements describing the mapping.
617This array is only valid during the scope of the callback function.
618.It Er EFBIG
619A mapping could not be achieved within the segment constraints provided
620in the tag even though the requested allocation size was less than maxsize.
621.El
622.It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
623"flags"
624This is a variation of
625.Fn bus_dmamap_load
626which maps mbuf chains
627for DMA transfers.
628A
629.Vt bus_size_t
630argument is also passed to the callback routine, which
631contains the mbuf chain's packet header length.
632The
633.Dv BUS_DMA_NOWAIT
634flag is implied, thus no callback deferral will happen.
635.Pp
636Mbuf chains are assumed to be in kernel virtual address space.
637.Pp
638Beside the error values listed for
639.Fn bus_dmamap_load ,
640.Er EINVAL
641will be returned if the size of the mbuf chain exceeds the maximum limit of the
642DMA tag.
643.It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
644This is just like
645.Fn bus_dmamap_load_mbuf
646except that it returns immediately without calling a callback function.
647It is provided for efficiency.
648The scatter/gather segment array
649.Va segs
650is provided by the caller and filled in directly by the function.
651The
652.Va nsegs
653argument is returned with the number of segments filled in.
654Returns the same errors as
655.Fn bus_dmamap_load_mbuf .
656.It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
657This is a variation of
658.Fn bus_dmamap_load
659which maps buffers pointed to by
660.Fa uio
661for DMA transfers.
662A
663.Vt bus_size_t
664argument is also passed to the callback routine, which contains the size of
665.Fa uio ,
666i.e.
667.Fa uio->uio_resid .
668The
669.Dv BUS_DMA_NOWAIT
670flag is implied, thus no callback deferral will happen.
671Returns the same errors as
672.Fn bus_dmamap_load .
673.Pp
674If
675.Fa uio->uio_segflg
676is
677.Dv UIO_USERSPACE ,
678then it is assumed that the buffer,
679.Fa uio
680is in
681.Fa "uio->uio_td->td_proc" Ns 's
682address space.
683User space memory must be in-core and wired prior to attempting a map
684load operation.
685Pages may be locked using
686.Xr vslock 9 .
687.It Fn bus_dmamap_unload "dmat" "map"
688Unloads a DMA map.
689Arguments are as follows:
690.Bl -tag -width ".Fa dmam"
691.It Fa dmat
692DMA tag used to allocate
693.Fa map .
694.It Fa map
695The DMA map that is to be unloaded.
696.El
697.Pp
698.Fn bus_dmamap_unload
699will not perform any implicit synchronization of DMA buffers.
700This must be done explicitly by a call to
701.Fn bus_dmamap_sync
702prior to unloading the map.
703.It Fn bus_dmamap_sync "dmat" "map" "op"
704Performs synchronization of a device visible mapping with the CPU visible
705memory referenced by that mapping.
706Arguments are as follows:
707.Bl -tag -width ".Fa dmat"
708.It Fa dmat
709DMA tag used to allocate
710.Fa map .
711.It Fa map
712The DMA mapping to be synchronized.
713.It Fa op
714Type of synchronization operation to perform.
715See the definition of
716.Vt bus_dmasync_op_t
717for a description of the acceptable values for
718.Fa op .
719.El
720.Pp
721The
722.Fn bus_dmamap_sync
723function
724is the method used to ensure that CPU's and device's direct
725memory access (DMA) to shared
726memory is coherent.
727For example, the CPU might be used to set up the contents of a buffer
728that is to be made available to a device.
729To ensure that the data are visible via the device's mapping of that
730memory, the buffer must be loaded and a DMA sync operation of
731.Dv BUS_DMASYNC_PREWRITE
732must be performed after the CPU has updated the buffer and before the device
733access is initiated.
734If the CPU modifies this buffer again later, another
735.Dv BUS_DMASYNC_PREWRITE
736sync operation must be performed before an additional device
737access.
738Conversely, suppose a device updates memory that is to be read by a CPU.
739In this case, the buffer must be loaded, and a DMA sync operation of
740.Dv BUS_DMASYNC_PREREAD
741must be performed before the device access is initiated.
742The CPU will only be able to see the results of this memory update
743once the DMA operation has completed and a
744.Dv BUS_DMASYNC_POSTREAD
745sync operation has been performed.
746.Pp
747If read and write operations are not preceded and followed by the
748appropriate synchronization operations, behavior is undefined.
749.It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
750Allocates memory that is mapped into KVA at the address returned
751in
752.Fa vaddr
753and that is permanently loaded into the newly created
754.Vt bus_dmamap_t
755returned via
756.Fa mapp .
757Arguments are as follows:
758.Bl -tag -width ".Fa alignment"
759.It Fa dmat
760DMA tag describing the constraints of the DMA mapping.
761.It Fa vaddr
762Pointer to a pointer that will hold the returned KVA mapping of
763the allocated region.
764.It Fa flags
765Flags are defined as follows:
766.Bl -tag -width ".Dv BUS_DMA_NOWAIT"
767.It Dv BUS_DMA_WAITOK
768The routine can safely wait (sleep) for resources.
769.It Dv BUS_DMA_NOWAIT
770The routine is not allowed to wait for resources.
771If resources are not available,
772.Dv ENOMEM
773is returned.
774.It Dv BUS_DMA_COHERENT
775Attempt to map this memory in a coherent fashion.
776See
777.Fn bus_dmamap_create
778above for a description of this flag.
779For
780.Fn bus_dmamem_alloc ,
781the
782.Dv BUS_DMA_COHERENT
783flag is currently implemented on arm and sparc64.
784.It Dv BUS_DMA_ZERO
785Causes the allocated memory to be set to all zeros.
786.It Dv BUS_DMA_NOCACHE
787The allocated memory will not be cached in the processor caches.
788All memory accesses appear on the bus and are executed
789without reordering.
790For
791.Fn bus_dmamem_alloc ,
792the
793.Dv BUS_DMA_NOCACHE
794flag is currently implemented on amd64 and i386 where it results in the
795Strong Uncacheable PAT to be set for the allocated virtual address range.
796.El
797.It Fa mapp
798Pointer to a
799.Vt bus_dmamap_t
800where the resulting DMA map will be stored.
801.El
802.Pp
803The size of memory to be allocated is
804.Fa maxsize
805as specified in the call to
806.Fn bus_dma_tag_create
807for
808.Fa dmat .
809.Pp
810The current implementation of
811.Fn bus_dmamem_alloc
812will allocate all requests as a single segment.
813.Pp
814An initial load operation is required to obtain the bus address of the allocated
815memory, and an unload operation is required before freeing the memory, as
816described below in
817.Fn bus_dmamem_free .
818Maps are automatically handled by this function and should not be explicitly
819allocated or destroyed.
820.Pp
821Although an explicit load is not required for each access to the memory
822referenced by the returned map, the synchronization requirements
823as described in the
824.Fn bus_dmamap_sync
825section still apply and should be used to achieve portability on architectures
826without coherent buses.
827.Pp
828Returns
829.Er ENOMEM
830if sufficient memory is not available for completing
831the operation.
832.It Fn bus_dmamem_free "dmat" "*vaddr" "map"
833Frees memory previously allocated by
834.Fn bus_dmamem_alloc .
835Any mappings
836will be invalidated.
837Arguments are as follows:
838.Bl -tag -width ".Fa vaddr"
839.It Fa dmat
840DMA tag.
841.It Fa vaddr
842Kernel virtual address of the memory.
843.It Fa map
844DMA map to be invalidated.
845.El
846.El
847.Sh RETURN VALUES
848Behavior is undefined if invalid arguments are passed to
849any of the above functions.
850If sufficient resources cannot be allocated for a given
851transaction,
852.Er ENOMEM
853is returned.
854All
855routines that are not of type
856.Vt void
857will return 0 on success or an error
858code on failure as discussed above.
859.Pp
860All
861.Vt void
862routines will succeed if provided with valid arguments.
863.Sh LOCKING
864Two locking protocols are used by
865.Nm .
866The first is a private global lock that is used to synchronize access to the
867bounce buffer pool on the architectures that make use of them.
868This lock is strictly a leaf lock that is only used internally to
869.Nm
870and is not exposed to clients of the API.
871.Pp
872The second protocol involves protecting various resources stored in the tag.
873Since almost all
874.Nm
875operations are done through requests from the driver that created the tag,
876the most efficient way to protect the tag resources is through the lock that
877the driver uses.
878In cases where
879.Nm
880acts on its own without being called by the driver, the lock primitive
881specified in the tag is acquired and released automatically.
882An example of this is when the
883.Fn bus_dmamap_load
884callback function is called from a deferred context instead of the driver
885context.
886This means that certain
887.Nm
888functions must always be called with the same lock held that is specified in the
889tag.
890These functions include:
891.Pp
892.Bl -item -offset indent -compact
893.It
894.Fn bus_dmamap_load
895.It
896.Fn bus_dmamap_load_uio
897.It
898.Fn bus_dmamap_load_mbuf
899.It
900.Fn bus_dmamap_load_mbuf_sg
901.It
902.Fn bus_dmamap_unload
903.It
904.Fn bus_dmamap_sync
905.El
906.Pp
907There is one exception to this rule.
908It is common practice to call some of these functions during driver start-up
909without any locks held.
910So long as there is a guarantee of no possible concurrent use of the tag by
911different threads during this operation, it is safe to not hold a lock for
912these functions.
913.Pp
914Certain
915.Nm
916operations should not be called with the driver lock held, either because
917they are already protected by an internal lock, or because they might sleep
918due to memory or resource allocation.
919The following functions must not be
920called with any non-sleepable locks held:
921.Pp
922.Bl -item -offset indent -compact
923.It
924.Fn bus_dma_tag_create
925.It
926.Fn bus_dmamap_create
927.It
928.Fn bus_dmamem_alloc
929.El
930.Pp
931All other functions do not have a locking protocol and can thus be
932called with or without any system or driver locks held.
933.Sh SEE ALSO
934.Xr devclass 9 ,
935.Xr device 9 ,
936.Xr driver 9 ,
937.Xr rman 9 ,
938.Xr vslock 9
939.Pp
940.Rs
941.%A "Jason R. Thorpe"
942.%T "A Machine-Independent DMA Framework for NetBSD"
943.%J "Proceedings of the Summer 1998 USENIX Technical Conference"
944.%Q "USENIX Association"
945.%D "June 1998"
946.Re
947.Sh HISTORY
948The
949.Nm
950interface first appeared in
951.Nx 1.3 .
952.Pp
953The
954.Nm
955API was adopted from
956.Nx
957for use in the CAM SCSI subsystem.
958The alterations to the original API were aimed to remove the need for
959a
960.Vt bus_dma_segment_t
961array stored in each
962.Vt bus_dmamap_t
963while allowing callers to queue up on scarce resources.
964.Sh AUTHORS
965The
966.Nm
967interface was designed and implemented by
968.An Jason R. Thorpe
969of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
970Additional input on the
971.Nm
972design was provided by
973.An -nosplit
974.An Chris Demetriou ,
975.An Charles Hannum ,
976.An Ross Harvey ,
977.An Matthew Jacob ,
978.An Jonathan Stone ,
979and
980.An Matt Thomas .
981.Pp
982The
983.Nm
984interface in
985.Fx
986benefits from the contributions of
987.An Justin T. Gibbs ,
988.An Peter Wemm ,
989.An Doug Rabson ,
990.An Matthew N. Dodd ,
991.An Sam Leffler ,
992.An Maxime Henrion ,
993.An Jake Burkholder ,
994.An Takahashi Yoshihiro ,
995.An Scott Long
996and many others.
997.Pp
998This manual page was written by
999.An Hiten M. Pandya
1000and
1001.An Justin T. Gibbs .
1002