xref: /freebsd/share/man/man9/bus_dma.9 (revision cc16dea626cf2fc80cde667ac4798065108e596c)
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 July 17, 2013
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_bio ,
74.Nm bus_dmamap_load_ccb ,
75.Nm bus_dmamap_load_mbuf ,
76.Nm bus_dmamap_load_mbuf_sg ,
77.Nm bus_dmamap_load_uio ,
78.Nm bus_dmamap_unload ,
79.Nm bus_dmamap_sync ,
80.Nm bus_dmamem_alloc ,
81.Nm bus_dmamem_free
82.Nd Bus and Machine Independent DMA Mapping Interface
83.Sh SYNOPSIS
84.In machine/bus.h
85.Ft int
86.Fn bus_dma_tag_create "bus_dma_tag_t parent" "bus_size_t alignment" \
87"bus_addr_t boundary" "bus_addr_t lowaddr" "bus_addr_t highaddr" \
88"bus_dma_filter_t *filtfunc" "void *filtfuncarg" "bus_size_t maxsize" \
89"int nsegments" "bus_size_t maxsegsz" "int flags" "bus_dma_lock_t *lockfunc" \
90"void *lockfuncarg" "bus_dma_tag_t *dmat"
91.Ft int
92.Fn bus_dma_tag_destroy "bus_dma_tag_t dmat"
93.Ft int
94.Fn bus_dmamap_create "bus_dma_tag_t dmat" "int flags" "bus_dmamap_t *mapp"
95.Ft int
96.Fn bus_dmamap_destroy "bus_dma_tag_t dmat" "bus_dmamap_t map"
97.Ft int
98.Fn bus_dmamap_load "bus_dma_tag_t dmat" "bus_dmamap_t map" "void *buf" \
99"bus_size_t buflen" "bus_dmamap_callback_t *callback" "void *callback_arg" \
100"int flags"
101.Ft int
102.Fn bus_dmamap_load_bio "bus_dma_tag_t dmat" "bus_dmamap_t map" \
103"struct bio *bio" "bus_dmamap_callback_t *callback" "void *callback_arg" \
104"int flags"
105.Ft int
106.Fn bus_dmamap_load_ccb "bus_dma_tag_t dmat" "bus_dmamap_t map" \
107"union ccb *ccb" "bus_dmamap_callback_t *callback" "void *callback_arg" \
108"int flags"
109.Ft int
110.Fn bus_dmamap_load_mbuf "bus_dma_tag_t dmat" "bus_dmamap_t map" \
111"struct mbuf *mbuf" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
112"int flags"
113.Ft int
114.Fn bus_dmamap_load_mbuf_sg "bus_dma_tag_t dmat" "bus_dmamap_t map" \
115"struct mbuf *mbuf" "bus_dma_segment_t *segs" "int *nsegs" "int flags"
116.Ft int
117.Fn bus_dmamap_load_uio "bus_dma_tag_t dmat" "bus_dmamap_t map" \
118"struct uio *uio" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
119"int flags"
120.Ft void
121.Fn bus_dmamap_unload "bus_dma_tag_t dmat" "bus_dmamap_t map"
122.Ft void
123.Fn bus_dmamap_sync "bus_dma_tag_t dmat" "bus_dmamap_t map" \
124"op"
125.Ft int
126.Fn bus_dmamem_alloc "bus_dma_tag_t dmat" "void **vaddr" \
127"int flags" "bus_dmamap_t *mapp"
128.Ft void
129.Fn bus_dmamem_free "bus_dma_tag_t dmat" "void *vaddr" \
130"bus_dmamap_t map"
131.Sh DESCRIPTION
132Direct Memory Access (DMA) is a method of transferring data
133without involving the CPU, thus providing higher performance.
134A DMA transaction can be achieved between device to memory,
135device to device, or memory to memory.
136.Pp
137The
138.Nm
139API is a bus, device, and machine-independent (MI) interface to
140DMA mechanisms.
141It provides the client with flexibility and simplicity by
142abstracting machine dependent issues like setting up
143DMA mappings, handling cache issues, bus specific features
144and limitations.
145.Sh STRUCTURES AND TYPES
146.Bl -tag -width indent
147.It Vt bus_dma_tag_t
148A machine-dependent (MD) opaque type that describes the
149characteristics of DMA transactions.
150DMA tags are organized into a hierarchy, with each child
151tag inheriting the restrictions of its parent.
152This allows all devices along the path of DMA transactions
153to contribute to the constraints of those transactions.
154.It Vt bus_dma_filter_t
155Client specified address filter having the format:
156.Bl -tag -width indent
157.It Ft int
158.Fn "client_filter" "void *filtarg" "bus_addr_t testaddr"
159.El
160.Pp
161Address filters can be specified during tag creation to allow
162for devices whose DMA address restrictions cannot be specified
163by a single window.
164The
165.Fa filtarg
166argument is specified by the client during tag creation to be passed to all
167invocations of the callback.
168The
169.Fa testaddr
170argument contains a potential starting address of a DMA mapping.
171The filter function operates on the set of addresses from
172.Fa testaddr
173to
174.Ql trunc_page(testaddr) + PAGE_SIZE - 1 ,
175inclusive.
176The filter function should return zero if any mapping in this range
177can be accommodated by the device and non-zero otherwise.
178.It Vt bus_dma_segment_t
179A machine-dependent type that describes individual
180DMA segments.
181It contains the following fields:
182.Bd -literal
183	bus_addr_t	ds_addr;
184	bus_size_t	ds_len;
185.Ed
186.Pp
187The
188.Fa ds_addr
189field contains the device visible address of the DMA segment, and
190.Fa ds_len
191contains the length of the DMA segment.
192Although the DMA segments returned by a mapping call will adhere to
193all restrictions necessary for a successful DMA operation, some conversion
194(e.g.\& a conversion from host byte order to the device's byte order) is
195almost always required when presenting segment information to the device.
196.It Vt bus_dmamap_t
197A machine-dependent opaque type describing an individual mapping.
198One map is used for each memory allocation that will be loaded.
199Maps can be reused once they have been unloaded.
200Multiple maps can be associated with one DMA tag.
201While the value of the map may evaluate to
202.Dv NULL
203on some platforms under certain conditions,
204it should never be assumed that it will be
205.Dv NULL
206in all cases.
207.It Vt bus_dmamap_callback_t
208Client specified callback for receiving mapping information resulting from
209the load of a
210.Vt bus_dmamap_t
211via
212.Fn bus_dmamap_load ,
213.Fn bus_dmamap_load_bio
214or
215.Fn bus_dmamap_load_ccb .
216Callbacks are of the format:
217.Bl -tag -width indent
218.It Ft void
219.Fn "client_callback" "void *callback_arg" "bus_dma_segment_t *segs" \
220"int nseg" "int error"
221.El
222.Pp
223The
224.Fa callback_arg
225is the callback argument passed to dmamap load functions.
226The
227.Fa segs
228and
229.Fa nseg
230arguments describe an array of
231.Vt bus_dma_segment_t
232structures that represent the mapping.
233This array is only valid within the scope of the callback function.
234The success or failure of the mapping is indicated by the
235.Fa error
236argument.
237More information on the use of callbacks can be found in the
238description of the individual dmamap load functions.
239.It Vt bus_dmamap_callback2_t
240Client specified callback for receiving mapping information resulting from
241the load of a
242.Vt bus_dmamap_t
243via
244.Fn bus_dmamap_load_uio
245or
246.Fn bus_dmamap_load_mbuf .
247.Pp
248Callback2s are of the format:
249.Bl -tag -width indent
250.It Ft void
251.Fn "client_callback2" "void *callback_arg" "bus_dma_segment_t *segs" \
252"int nseg" "bus_size_t mapsize" "int error"
253.El
254.Pp
255Callback2's behavior is the same as
256.Vt bus_dmamap_callback_t
257with the addition that the length of the data mapped is provided via
258.Fa mapsize .
259.It Vt bus_dmasync_op_t
260Memory synchronization operation specifier.
261Bus DMA requires explicit synchronization of memory with its device
262visible mapping in order to guarantee memory coherency.
263The
264.Vt bus_dmasync_op_t
265allows the type of DMA operation that will be or has been performed
266to be communicated to the system so that the correct coherency measures
267are taken.
268The operations are represented as bitfield flags that can be combined together,
269though it only makes sense to combine PRE flags or POST flags, not both.
270See the
271.Fn bus_dmamap_sync
272description below for more details on how to use these operations.
273.Pp
274All operations specified below are performed from the host memory point of view,
275where a read implies data coming from the device to the host memory, and a write
276implies data going from the host memory to the device.
277Alternatively, the operations can be thought of in terms of driver operations,
278where reading a network packet or storage sector corresponds to a read operation
279in
280.Nm .
281.Bl -tag -width ".Dv BUS_DMASYNC_POSTWRITE"
282.It Dv BUS_DMASYNC_PREREAD
283Perform any synchronization required prior to an update of host memory by the
284device.
285.It Dv BUS_DMASYNC_PREWRITE
286Perform any synchronization required after an update of host memory by the CPU
287and prior to device access to host memory.
288.It Dv BUS_DMASYNC_POSTREAD
289Perform any synchronization required after an update of host memory by the
290device and prior to CPU access to host memory.
291.It Dv BUS_DMASYNC_POSTWRITE
292Perform any synchronization required after device access to host memory.
293.El
294.It Vt bus_dma_lock_t
295Client specified lock/mutex manipulation method.
296This will be called from
297within busdma whenever a client lock needs to be manipulated.
298In its current form, the function will be called immediately before
299the callback for a DMA load operation that has been deferred with
300.Dv BUS_DMA_LOCK
301and immediately after with
302.Dv BUS_DMA_UNLOCK .
303If the load operation does not need to be deferred, then it
304will not be called since the function loading the map should
305be holding the appropriate locks.
306This method is of the format:
307.Bl -tag -width indent
308.It Ft void
309.Fn "lockfunc" "void *lockfunc_arg" "bus_dma_lock_op_t op"
310.El
311.Pp
312The
313.Fa lockfuncarg
314argument is specified by the client during tag creation to be passed to all
315invocations of the callback.
316The
317.Fa op
318argument specifies the lock operation to perform.
319.Pp
320Two
321.Vt lockfunc
322implementations are provided for convenience.
323.Fn busdma_lock_mutex
324performs standard mutex operations on the sleep mutex provided via
325.Fa lockfuncarg .
326.Fn dflt_lock
327will generate a system panic if it is called.
328It is substituted into the tag when
329.Fa lockfunc
330is passed as
331.Dv NULL
332to
333.Fn bus_dma_tag_create
334and is useful for tags that should not be used with deferred load operations.
335.It Vt bus_dma_lock_op_t
336Operations to be performed by the client-specified
337.Fn lockfunc .
338.Bl -tag -width ".Dv BUS_DMA_UNLOCK"
339.It Dv BUS_DMA_LOCK
340Acquires and/or locks the client locking primitive.
341.It Dv BUS_DMA_UNLOCK
342Releases and/or unlocks the client locking primitive.
343.El
344.El
345.Sh FUNCTIONS
346.Bl -tag -width indent
347.It Fn bus_dma_tag_create "parent" "alignment" "boundary" "lowaddr" \
348"highaddr" "*filtfunc" "*filtfuncarg" "maxsize" "nsegments" "maxsegsz" \
349"flags" "lockfunc" "lockfuncarg" "*dmat"
350Allocates a device specific DMA tag, and initializes it according to
351the arguments provided:
352.Bl -tag -width ".Fa filtfuncarg"
353.It Fa parent
354Indicates restrictions between the parent bridge, CPU memory, and the
355device.
356Each device must use a master parent tag by calling
357.Fn bus_get_dma_tag .
358.It Fa alignment
359Alignment constraint, in bytes, of any mappings created using this tag.
360The alignment must be a power of 2.
361Hardware that can DMA starting at any address would specify
362.Em 1
363for byte alignment.
364Hardware requiring DMA transfers to start on a multiple of 4K
365would specify
366.Em 4096 .
367.It Fa boundary
368Boundary constraint, in bytes, of the target DMA memory region.
369The boundary indicates the set of addresses, all multiples of the
370boundary argument, that cannot be crossed by a single
371.Vt bus_dma_segment_t .
372The boundary must be a power of 2 and must be no smaller than the
373maximum segment size.
374.Ql 0
375indicates that there are no boundary restrictions.
376.It Fa lowaddr , highaddr
377Bounds of the window of bus address space that
378.Em cannot
379be directly accessed by the device.
380The window contains all addresses greater than
381.Fa lowaddr
382and less than or equal to
383.Fa highaddr .
384For example, a device incapable of DMA above 4GB, would specify a
385.Fa highaddr
386of
387.Dv BUS_SPACE_MAXADDR
388and a
389.Fa lowaddr
390of
391.Dv BUS_SPACE_MAXADDR_32BIT .
392Similarly a device that can only perform DMA to addresses below
39316MB would specify a
394.Fa highaddr
395of
396.Dv BUS_SPACE_MAXADDR
397and a
398.Fa lowaddr
399of
400.Dv BUS_SPACE_MAXADDR_24BIT .
401Some implementations requires that some region of device visible
402address space, overlapping available host memory, be outside the
403window.
404This area of
405.Ql safe memory
406is used to bounce requests that would otherwise conflict with
407the exclusion window.
408.It Fa filtfunc
409Optional filter function (may be
410.Dv NULL )
411to be called for any attempt to
412map memory into the window described by
413.Fa lowaddr
414and
415.Fa highaddr .
416A filter function is only required when the single window described
417by
418.Fa lowaddr
419and
420.Fa highaddr
421cannot adequately describe the constraints of the device.
422The filter function will be called for every machine page
423that overlaps the exclusion window.
424.It Fa filtfuncarg
425Argument passed to all calls to the filter function for this tag.
426May be
427.Dv NULL .
428.It Fa maxsize
429Maximum size, in bytes, of the sum of all segment lengths in a given
430DMA mapping associated with this tag.
431.It Fa nsegments
432Number of discontinuities (scatter/gather segments) allowed
433in a DMA mapped region.
434If there is no restriction,
435.Dv BUS_SPACE_UNRESTRICTED
436may be specified.
437.It Fa maxsegsz
438Maximum size, in bytes, of a segment in any DMA mapped region associated
439with
440.Fa dmat .
441.It Fa flags
442Are as follows:
443.Bl -tag -width ".Dv BUS_DMA_ALLOCNOW"
444.It Dv BUS_DMA_ALLOCNOW
445Pre-allocate enough resources to handle at least one map load operation on
446this tag.
447If sufficient resources are not available,
448.Er ENOMEM
449is returned.
450This should not be used for tags that only describe buffers that will be
451allocated with
452.Fn bus_dmamem_alloc .
453Also, due to resource sharing with other tags, this flag does not guarantee
454that resources will be allocated or reserved exclusively for this tag.
455It should be treated only as a minor optimization.
456.El
457.It Fa lockfunc
458Optional lock manipulation function (may be
459.Dv NULL )
460to be called when busdma
461needs to manipulate a lock on behalf of the client.
462If
463.Dv NULL
464is specified,
465.Fn dflt_lock
466is used.
467.It Fa lockfuncarg
468Optional argument to be passed to the function specified by
469.Fa lockfunc .
470.It Fa dmat
471Pointer to a bus_dma_tag_t where the resulting DMA tag will
472be stored.
473.El
474.Pp
475Returns
476.Er ENOMEM
477if sufficient memory is not available for tag creation
478or allocating mapping resources.
479.It Fn bus_dma_tag_destroy "dmat"
480Deallocate the DMA tag
481.Fa dmat
482that was created by
483.Fn bus_dma_tag_create .
484.Pp
485Returns
486.Er EBUSY
487if any DMA maps remain associated with
488.Fa dmat
489or
490.Ql 0
491on success.
492.It Fn bus_dmamap_create "dmat" "flags" "*mapp"
493Allocates and initializes a DMA map.
494Arguments are as follows:
495.Bl -tag -width ".Fa nsegments"
496.It Fa dmat
497DMA tag.
498.It Fa flags
499Are as follows:
500.Bl -tag -width ".Dv BUS_DMA_COHERENT"
501.It Dv BUS_DMA_COHERENT
502Attempt to map the memory loaded with this map such that cache sync
503operations are as cheap as possible.
504This flag is typically set on maps when the memory loaded with these will
505be accessed by both a CPU and a DMA engine, frequently such as control data
506and as opposed to streamable data such as receive and transmit buffers.
507Use of this flag does not remove the requirement of using
508.Fn bus_dmamap_sync ,
509but it may reduce the cost of performing these operations.
510For
511.Fn bus_dmamap_create ,
512the
513.Dv BUS_DMA_COHERENT
514flag is currently implemented on sparc64.
515.El
516.It Fa mapp
517Pointer to a
518.Vt bus_dmamap_t
519where the resulting DMA map will be stored.
520.El
521.Pp
522Returns
523.Er ENOMEM
524if sufficient memory is not available for creating the
525map or allocating mapping resources.
526.It Fn bus_dmamap_destroy "dmat" "map"
527Frees all resources associated with a given DMA map.
528Arguments are as follows:
529.Bl -tag -width ".Fa dmat"
530.It Fa dmat
531DMA tag used to allocate
532.Fa map .
533.It Fa map
534The DMA map to destroy.
535.El
536.Pp
537Returns
538.Er EBUSY
539if a mapping is still active for
540.Fa map .
541.It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
542"callback_arg" "flags"
543Creates a mapping in device visible address space of
544.Fa buflen
545bytes of
546.Fa buf ,
547associated with the DMA map
548.Fa map .
549This call will always return immediately and will not block for any reason.
550Arguments are as follows:
551.Bl -tag -width ".Fa buflen"
552.It Fa dmat
553DMA tag used to allocate
554.Fa map .
555.It Fa map
556A DMA map without a currently active mapping.
557.It Fa buf
558A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
559mapped into device visible address space.
560.It Fa buflen
561The size of the buffer.
562.It Fa callback Fa callback_arg
563The callback function, and its argument.
564This function is called once sufficient mapping resources are available for
565the DMA operation.
566If resources are temporarily unavailable, this function will be deferred until
567later, but the load operation will still return immediately to the caller.
568Thus, callers should not assume that the callback will be called before the
569load returns, and code should be structured appropriately to handle this.
570See below for specific flags and error codes that control this behavior.
571.It Fa flags
572Are as follows:
573.Bl -tag -width ".Dv BUS_DMA_NOWAIT"
574.It Dv BUS_DMA_NOWAIT
575The load should not be deferred in case of insufficient mapping resources,
576and instead should return immediately with an appropriate error.
577.It Dv BUS_DMA_NOCACHE
578The generated transactions to and from the virtual page are non-cacheable.
579For
580.Fn bus_dmamap_load ,
581the
582.Dv BUS_DMA_NOCACHE
583flag is currently implemented on sparc64.
584.El
585.El
586.Pp
587Return values to the caller are as follows:
588.Bl -tag -width ".Er EINPROGRESS"
589.It 0
590The callback has been called and completed.
591The status of the mapping has been delivered to the callback.
592.It Er EINPROGRESS
593The mapping has been deferred for lack of resources.
594The callback will be called as soon as resources are available.
595Callbacks are serviced in FIFO order.
596.Pp
597Note that subsequent load operations for the same tag that do not require
598extra resources will still succeed.
599This may result in out-of-order processing of requests.
600If the caller requires the order of requests to be preserved,
601then the caller is required to stall subsequent requests until a pending
602request's callback is invoked.
603.It Er ENOMEM
604The load request has failed due to insufficient resources, and the caller
605specifically used the
606.Dv BUS_DMA_NOWAIT
607flag.
608.It Er EINVAL
609The load request was invalid.
610The callback has been called and has been provided the same error.
611This error value may indicate that
612.Fa dmat ,
613.Fa map ,
614.Fa buf ,
615or
616.Fa callback
617were invalid, or
618.Fa buflen
619was larger than the
620.Fa maxsize
621argument used to create the dma tag
622.Fa dmat .
623.El
624.Pp
625When the callback is called, it is presented with an error value
626indicating the disposition of the mapping.
627Error may be one of the following:
628.Bl -tag -width ".Er EINPROGRESS"
629.It 0
630The mapping was successful and the
631.Fa dm_segs
632callback argument contains an array of
633.Vt bus_dma_segment_t
634elements describing the mapping.
635This array is only valid during the scope of the callback function.
636.It Er EFBIG
637A mapping could not be achieved within the segment constraints provided
638in the tag even though the requested allocation size was less than maxsize.
639.El
640.It Fn bus_dmamap_load_bio "dmat" "map" "bio" "callback" "callback_arg" "flags"
641This is a variation of
642.Fn bus_dmamap_load
643which maps buffers pointed to by
644.Fa bio
645for DMA transfers.
646.Fa bio
647may point to either a mapped or unmapped buffer.
648.It Fn bus_dmamap_load_ccb "dmat" "map" "ccb" "callback" "callback_arg" "flags"
649This is a variation of
650.Fn bus_dmamap_load
651which maps data pointed to by
652.Fa ccb
653for DMA transfers.
654The data for
655.Fa ccb
656may be any of the following types:
657.Bl -tag -width ".Er CAM_DATA_SG_PADDR"
658.It CAM_DATA_VADDR
659The data is a single KVA buffer.
660.It CAM_DATA_PADDR
661The data is a single bus address range.
662.It CAM_DATA_SG
663The data is a scatter/gather list of KVA buffers.
664.It CAM_DATA_SG_PADDR
665The data is a scatter/gather list of bus address ranges.
666.It CAM_DATA_BIO
667The data is contained in a
668.Vt struct bio
669attached to the CCB.
670.El
671.Pp
672.Fn bus_dmamap_load_ccb
673supports the following CCB XPT function codes:
674.Pp
675.Bl -item -offset indent -compact
676.It
677XPT_ATA_IO
678.It
679XPT_CONT_TARGET_IO
680.It
681XPT_SCSI_IO
682.El
683.It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
684"flags"
685This is a variation of
686.Fn bus_dmamap_load
687which maps mbuf chains
688for DMA transfers.
689A
690.Vt bus_size_t
691argument is also passed to the callback routine, which
692contains the mbuf chain's packet header length.
693The
694.Dv BUS_DMA_NOWAIT
695flag is implied, thus no callback deferral will happen.
696.Pp
697Mbuf chains are assumed to be in kernel virtual address space.
698.Pp
699Beside the error values listed for
700.Fn bus_dmamap_load ,
701.Er EINVAL
702will be returned if the size of the mbuf chain exceeds the maximum limit of the
703DMA tag.
704.It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
705This is just like
706.Fn bus_dmamap_load_mbuf
707except that it returns immediately without calling a callback function.
708It is provided for efficiency.
709The scatter/gather segment array
710.Va segs
711is provided by the caller and filled in directly by the function.
712The
713.Va nsegs
714argument is returned with the number of segments filled in.
715Returns the same errors as
716.Fn bus_dmamap_load_mbuf .
717.It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
718This is a variation of
719.Fn bus_dmamap_load
720which maps buffers pointed to by
721.Fa uio
722for DMA transfers.
723A
724.Vt bus_size_t
725argument is also passed to the callback routine, which contains the size of
726.Fa uio ,
727i.e.
728.Fa uio->uio_resid .
729The
730.Dv BUS_DMA_NOWAIT
731flag is implied, thus no callback deferral will happen.
732Returns the same errors as
733.Fn bus_dmamap_load .
734.Pp
735If
736.Fa uio->uio_segflg
737is
738.Dv UIO_USERSPACE ,
739then it is assumed that the buffer,
740.Fa uio
741is in
742.Fa "uio->uio_td->td_proc" Ns 's
743address space.
744User space memory must be in-core and wired prior to attempting a map
745load operation.
746Pages may be locked using
747.Xr vslock 9 .
748.It Fn bus_dmamap_unload "dmat" "map"
749Unloads a DMA map.
750Arguments are as follows:
751.Bl -tag -width ".Fa dmam"
752.It Fa dmat
753DMA tag used to allocate
754.Fa map .
755.It Fa map
756The DMA map that is to be unloaded.
757.El
758.Pp
759.Fn bus_dmamap_unload
760will not perform any implicit synchronization of DMA buffers.
761This must be done explicitly by a call to
762.Fn bus_dmamap_sync
763prior to unloading the map.
764.It Fn bus_dmamap_sync "dmat" "map" "op"
765Performs synchronization of a device visible mapping with the CPU visible
766memory referenced by that mapping.
767Arguments are as follows:
768.Bl -tag -width ".Fa dmat"
769.It Fa dmat
770DMA tag used to allocate
771.Fa map .
772.It Fa map
773The DMA mapping to be synchronized.
774.It Fa op
775Type of synchronization operation to perform.
776See the definition of
777.Vt bus_dmasync_op_t
778for a description of the acceptable values for
779.Fa op .
780.El
781.Pp
782The
783.Fn bus_dmamap_sync
784function
785is the method used to ensure that CPU's and device's direct
786memory access (DMA) to shared
787memory is coherent.
788For example, the CPU might be used to set up the contents of a buffer
789that is to be made available to a device.
790To ensure that the data are visible via the device's mapping of that
791memory, the buffer must be loaded and a DMA sync operation of
792.Dv BUS_DMASYNC_PREWRITE
793must be performed after the CPU has updated the buffer and before the device
794access is initiated.
795If the CPU modifies this buffer again later, another
796.Dv BUS_DMASYNC_PREWRITE
797sync operation must be performed before an additional device
798access.
799Conversely, suppose a device updates memory that is to be read by a CPU.
800In this case, the buffer must be loaded, and a DMA sync operation of
801.Dv BUS_DMASYNC_PREREAD
802must be performed before the device access is initiated.
803The CPU will only be able to see the results of this memory update
804once the DMA operation has completed and a
805.Dv BUS_DMASYNC_POSTREAD
806sync operation has been performed.
807.Pp
808If read and write operations are not preceded and followed by the
809appropriate synchronization operations, behavior is undefined.
810.It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
811Allocates memory that is mapped into KVA at the address returned
812in
813.Fa vaddr
814and that is permanently loaded into the newly created
815.Vt bus_dmamap_t
816returned via
817.Fa mapp .
818Arguments are as follows:
819.Bl -tag -width ".Fa alignment"
820.It Fa dmat
821DMA tag describing the constraints of the DMA mapping.
822.It Fa vaddr
823Pointer to a pointer that will hold the returned KVA mapping of
824the allocated region.
825.It Fa flags
826Flags are defined as follows:
827.Bl -tag -width ".Dv BUS_DMA_NOWAIT"
828.It Dv BUS_DMA_WAITOK
829The routine can safely wait (sleep) for resources.
830.It Dv BUS_DMA_NOWAIT
831The routine is not allowed to wait for resources.
832If resources are not available,
833.Dv ENOMEM
834is returned.
835.It Dv BUS_DMA_COHERENT
836Attempt to map this memory in a coherent fashion.
837See
838.Fn bus_dmamap_create
839above for a description of this flag.
840For
841.Fn bus_dmamem_alloc ,
842the
843.Dv BUS_DMA_COHERENT
844flag is currently implemented on arm and sparc64.
845.It Dv BUS_DMA_ZERO
846Causes the allocated memory to be set to all zeros.
847.It Dv BUS_DMA_NOCACHE
848The allocated memory will not be cached in the processor caches.
849All memory accesses appear on the bus and are executed
850without reordering.
851For
852.Fn bus_dmamem_alloc ,
853the
854.Dv BUS_DMA_NOCACHE
855flag is currently implemented on amd64 and i386 where it results in the
856Strong Uncacheable PAT to be set for the allocated virtual address range.
857.El
858.It Fa mapp
859Pointer to a
860.Vt bus_dmamap_t
861where the resulting DMA map will be stored.
862.El
863.Pp
864The size of memory to be allocated is
865.Fa maxsize
866as specified in the call to
867.Fn bus_dma_tag_create
868for
869.Fa dmat .
870.Pp
871The current implementation of
872.Fn bus_dmamem_alloc
873will allocate all requests as a single segment.
874.Pp
875An initial load operation is required to obtain the bus address of the allocated
876memory, and an unload operation is required before freeing the memory, as
877described below in
878.Fn bus_dmamem_free .
879Maps are automatically handled by this function and should not be explicitly
880allocated or destroyed.
881.Pp
882Although an explicit load is not required for each access to the memory
883referenced by the returned map, the synchronization requirements
884as described in the
885.Fn bus_dmamap_sync
886section still apply and should be used to achieve portability on architectures
887without coherent buses.
888.Pp
889Returns
890.Er ENOMEM
891if sufficient memory is not available for completing
892the operation.
893.It Fn bus_dmamem_free "dmat" "*vaddr" "map"
894Frees memory previously allocated by
895.Fn bus_dmamem_alloc .
896Any mappings
897will be invalidated.
898Arguments are as follows:
899.Bl -tag -width ".Fa vaddr"
900.It Fa dmat
901DMA tag.
902.It Fa vaddr
903Kernel virtual address of the memory.
904.It Fa map
905DMA map to be invalidated.
906.El
907.El
908.Sh RETURN VALUES
909Behavior is undefined if invalid arguments are passed to
910any of the above functions.
911If sufficient resources cannot be allocated for a given
912transaction,
913.Er ENOMEM
914is returned.
915All
916routines that are not of type
917.Vt void
918will return 0 on success or an error
919code on failure as discussed above.
920.Pp
921All
922.Vt void
923routines will succeed if provided with valid arguments.
924.Sh LOCKING
925Two locking protocols are used by
926.Nm .
927The first is a private global lock that is used to synchronize access to the
928bounce buffer pool on the architectures that make use of them.
929This lock is strictly a leaf lock that is only used internally to
930.Nm
931and is not exposed to clients of the API.
932.Pp
933The second protocol involves protecting various resources stored in the tag.
934Since almost all
935.Nm
936operations are done through requests from the driver that created the tag,
937the most efficient way to protect the tag resources is through the lock that
938the driver uses.
939In cases where
940.Nm
941acts on its own without being called by the driver, the lock primitive
942specified in the tag is acquired and released automatically.
943An example of this is when the
944.Fn bus_dmamap_load
945callback function is called from a deferred context instead of the driver
946context.
947This means that certain
948.Nm
949functions must always be called with the same lock held that is specified in the
950tag.
951These functions include:
952.Pp
953.Bl -item -offset indent -compact
954.It
955.Fn bus_dmamap_load
956.It
957.Fn bus_dmamap_load_bio
958.It
959.Fn bus_dmamap_load_ccb
960.It
961.Fn bus_dmamap_load_mbuf
962.It
963.Fn bus_dmamap_load_mbuf_sg
964.It
965.Fn bus_dmamap_load_uio
966.It
967.Fn bus_dmamap_unload
968.It
969.Fn bus_dmamap_sync
970.El
971.Pp
972There is one exception to this rule.
973It is common practice to call some of these functions during driver start-up
974without any locks held.
975So long as there is a guarantee of no possible concurrent use of the tag by
976different threads during this operation, it is safe to not hold a lock for
977these functions.
978.Pp
979Certain
980.Nm
981operations should not be called with the driver lock held, either because
982they are already protected by an internal lock, or because they might sleep
983due to memory or resource allocation.
984The following functions must not be
985called with any non-sleepable locks held:
986.Pp
987.Bl -item -offset indent -compact
988.It
989.Fn bus_dma_tag_create
990.It
991.Fn bus_dmamap_create
992.It
993.Fn bus_dmamem_alloc
994.El
995.Pp
996All other functions do not have a locking protocol and can thus be
997called with or without any system or driver locks held.
998.Sh SEE ALSO
999.Xr devclass 9 ,
1000.Xr device 9 ,
1001.Xr driver 9 ,
1002.Xr rman 9 ,
1003.Xr vslock 9
1004.Pp
1005.Rs
1006.%A "Jason R. Thorpe"
1007.%T "A Machine-Independent DMA Framework for NetBSD"
1008.%J "Proceedings of the Summer 1998 USENIX Technical Conference"
1009.%Q "USENIX Association"
1010.%D "June 1998"
1011.Re
1012.Sh HISTORY
1013The
1014.Nm
1015interface first appeared in
1016.Nx 1.3 .
1017.Pp
1018The
1019.Nm
1020API was adopted from
1021.Nx
1022for use in the CAM SCSI subsystem.
1023The alterations to the original API were aimed to remove the need for
1024a
1025.Vt bus_dma_segment_t
1026array stored in each
1027.Vt bus_dmamap_t
1028while allowing callers to queue up on scarce resources.
1029.Sh AUTHORS
1030The
1031.Nm
1032interface was designed and implemented by
1033.An Jason R. Thorpe
1034of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
1035Additional input on the
1036.Nm
1037design was provided by
1038.An -nosplit
1039.An Chris Demetriou ,
1040.An Charles Hannum ,
1041.An Ross Harvey ,
1042.An Matthew Jacob ,
1043.An Jonathan Stone ,
1044and
1045.An Matt Thomas .
1046.Pp
1047The
1048.Nm
1049interface in
1050.Fx
1051benefits from the contributions of
1052.An Justin T. Gibbs ,
1053.An Peter Wemm ,
1054.An Doug Rabson ,
1055.An Matthew N. Dodd ,
1056.An Sam Leffler ,
1057.An Maxime Henrion ,
1058.An Jake Burkholder ,
1059.An Takahashi Yoshihiro ,
1060.An Scott Long
1061and many others.
1062.Pp
1063This manual page was written by
1064.An Hiten M. Pandya
1065and
1066.An Justin T. Gibbs .
1067