1 .\" Copyright (c) 2002, 2003 Hiten M. Pandya.
2 .\" All rights reserved.
4 .\" Redistribution and use in source and binary forms, with or without
5 .\" modification, are permitted provided that the following conditions
7 .\" 1. Redistributions of source code must retain the above copyright
8 .\" notice, this list of conditions, and the following disclaimer,
9 .\" without modification, immediately at the beginning of the file.
10 .\" 2. The name of the author may not be used to endorse or promote products
11 .\" derived from this software without specific prior written permission.
13 .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR, CONTRIBUTORS OR THE
17 .\" VOICES IN HITEN PANDYA'S HEAD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
18 .\" SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
19 .\" TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
20 .\" PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
21 .\" LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
22 .\" NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
23 .\" SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 .\" Copyright (c) 1996, 1997, 1998, 2001 The NetBSD Foundation, Inc.
26 .\" All rights reserved.
28 .\" This code is derived from software contributed to The NetBSD Foundation
29 .\" by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
30 .\" NASA Ames Research Center.
32 .\" Redistribution and use in source and binary forms, with or without
33 .\" modification, are permitted provided that the following conditions
35 .\" 1. Redistributions of source code must retain the above copyright
36 .\" notice, this list of conditions and the following disclaimer.
37 .\" 2. Redistributions in binary form must reproduce the above copyright
38 .\" notice, this list of conditions and the following disclaimer in the
39 .\" documentation and/or other materials provided with the distribution.
40 .\" 3. All advertising materials mentioning features or use of this software
41 .\" must display the following acknowledgment:
42 .\" This product includes software developed by the NetBSD
43 .\" Foundation, Inc. and its contributors.
44 .\" 4. Neither the name of The NetBSD Foundation nor the names of its
45 .\" contributors may be used to endorse or promote products derived
46 .\" from this software without specific prior written permission.
48 .\" THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
49 .\" ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
50 .\" TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
51 .\" PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
52 .\" BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
53 .\" CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
54 .\" SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
55 .\" INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
56 .\" CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
57 .\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
58 .\" POSSIBILITY OF SUCH DAMAGE.
61 .\" $NetBSD: bus_dma.9,v 1.25 2002/10/14 13:43:16 wiz Exp $
68 .Nm bus_dma_tag_create ,
69 .Nm bus_dma_tag_destroy ,
70 .Nm bus_dmamap_create ,
71 .Nm bus_dmamap_destroy ,
73 .Nm bus_dmamap_load_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 ,
80 .Nm bus_dmamem_alloc ,
82 .Nd Bus and Machine Independent DMA Mapping Interface
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"
92 .Fn bus_dma_tag_destroy "bus_dma_tag_t dmat"
94 .Fn bus_dmamap_create "bus_dma_tag_t dmat" "int flags" "bus_dmamap_t *mapp"
96 .Fn bus_dmamap_destroy "bus_dma_tag_t dmat" "bus_dmamap_t map"
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" \
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" \
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" \
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" \
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"
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" \
121 .Fn bus_dmamap_unload "bus_dma_tag_t dmat" "bus_dmamap_t map"
123 .Fn bus_dmamap_sync "bus_dma_tag_t dmat" "bus_dmamap_t map" \
126 .Fn bus_dmamem_alloc "bus_dma_tag_t dmat" "void **vaddr" \
127 "int flags" "bus_dmamap_t *mapp"
129 .Fn bus_dmamem_free "bus_dma_tag_t dmat" "void *vaddr" \
132 Direct Memory Access (DMA) is a method of transferring data
133 without involving the CPU, thus providing higher performance.
134 A DMA transaction can be achieved between device to memory,
135 device to device, or memory to memory.
139 API is a bus, device, and machine-independent (MI) interface to
141 It provides the client with flexibility and simplicity by
142 abstracting machine dependent issues like setting up
143 DMA mappings, handling cache issues, bus specific features
145 .Sh STRUCTURES AND TYPES
146 .Bl -tag -width indent
148 A machine-dependent (MD) opaque type that describes the
149 characteristics of DMA transactions.
150 DMA tags are organized into a hierarchy, with each child
151 tag inheriting the restrictions of its parent.
152 This allows all devices along the path of DMA transactions
153 to contribute to the constraints of those transactions.
154 .It Vt bus_dma_filter_t
155 Client specified address filter having the format:
156 .Bl -tag -width indent
158 .Fn "client_filter" "void *filtarg" "bus_addr_t testaddr"
161 Address filters can be specified during tag creation to allow
162 for devices whose DMA address restrictions cannot be specified
166 argument is specified by the client during tag creation to be passed to all
167 invocations of the callback.
170 argument contains a potential starting address of a DMA mapping.
171 The filter function operates on the set of addresses from
174 .Ql trunc_page(testaddr) + PAGE_SIZE - 1 ,
176 The filter function should return zero if any mapping in this range
177 can be accommodated by the device and non-zero otherwise.
178 .It Vt bus_dma_segment_t
179 A machine-dependent type that describes individual
181 It contains the following fields:
189 field contains the device visible address of the DMA segment, and
191 contains the length of the DMA segment.
192 Although the DMA segments returned by a mapping call will adhere to
193 all 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
195 almost always required when presenting segment information to the device.
197 A machine-dependent opaque type describing an individual mapping.
198 One map is used for each memory allocation that will be loaded.
199 Maps can be reused once they have been unloaded.
200 Multiple maps can be associated with one DMA tag.
201 While the value of the map may evaluate to
203 on some platforms under certain conditions,
204 it should never be assumed that it will be
207 .It Vt bus_dmamap_callback_t
208 Client specified callback for receiving mapping information resulting from
212 .Fn bus_dmamap_load ,
213 .Fn bus_dmamap_load_bio
215 .Fn bus_dmamap_load_ccb .
216 Callbacks are of the format:
217 .Bl -tag -width indent
219 .Fn "client_callback" "void *callback_arg" "bus_dma_segment_t *segs" \
220 "int nseg" "int error"
225 is the callback argument passed to dmamap load functions.
230 arguments describe an array of
231 .Vt bus_dma_segment_t
232 structures that represent the mapping.
233 This array is only valid within the scope of the callback function.
234 The success or failure of the mapping is indicated by the
237 More information on the use of callbacks can be found in the
238 description of the individual dmamap load functions.
239 .It Vt bus_dmamap_callback2_t
240 Client specified callback for receiving mapping information resulting from
244 .Fn bus_dmamap_load_uio
246 .Fn bus_dmamap_load_mbuf .
248 Callback2s are of the format:
249 .Bl -tag -width indent
251 .Fn "client_callback2" "void *callback_arg" "bus_dma_segment_t *segs" \
252 "int nseg" "bus_size_t mapsize" "int error"
255 Callback2's behavior is the same as
256 .Vt bus_dmamap_callback_t
257 with the addition that the length of the data mapped is provided via
259 .It Vt bus_dmasync_op_t
260 Memory synchronization operation specifier.
261 Bus DMA requires explicit synchronization of memory with its device
262 visible mapping in order to guarantee memory coherency.
265 allows the type of DMA operation that will be or has been performed
266 to be communicated to the system so that the correct coherency measures
268 The operations are represented as bitfield flags that can be combined together,
269 though it only makes sense to combine PRE flags or POST flags, not both.
272 description below for more details on how to use these operations.
274 All operations specified below are performed from the host memory point of view,
275 where a read implies data coming from the device to the host memory, and a write
276 implies data going from the host memory to the device.
277 Alternatively, the operations can be thought of in terms of driver operations,
278 where reading a network packet or storage sector corresponds to a read operation
281 .Bl -tag -width ".Dv BUS_DMASYNC_POSTWRITE"
282 .It Dv BUS_DMASYNC_PREREAD
283 Perform any synchronization required prior to an update of host memory by the
285 .It Dv BUS_DMASYNC_PREWRITE
286 Perform any synchronization required after an update of host memory by the CPU
287 and prior to device access to host memory.
288 .It Dv BUS_DMASYNC_POSTREAD
289 Perform any synchronization required after an update of host memory by the
290 device and prior to CPU access to host memory.
291 .It Dv BUS_DMASYNC_POSTWRITE
292 Perform any synchronization required after device access to host memory.
294 .It Vt bus_dma_lock_t
295 Client specified lock/mutex manipulation method.
296 This will be called from
297 within busdma whenever a client lock needs to be manipulated.
298 In its current form, the function will be called immediately before
299 the callback for a DMA load operation that has been deferred with
301 and immediately after with
303 If the load operation does not need to be deferred, then it
304 will not be called since the function loading the map should
305 be holding the appropriate locks.
306 This method is of the format:
307 .Bl -tag -width indent
309 .Fn "lockfunc" "void *lockfunc_arg" "bus_dma_lock_op_t op"
314 argument is specified by the client during tag creation to be passed to all
315 invocations of the callback.
318 argument specifies the lock operation to perform.
322 implementations are provided for convenience.
323 .Fn busdma_lock_mutex
324 performs standard mutex operations on the sleep mutex provided via
327 will generate a system panic if it is called.
328 It is substituted into the tag when
333 .Fn bus_dma_tag_create
334 and is useful for tags that should not be used with deferred load operations.
335 .It Vt bus_dma_lock_op_t
336 Operations to be performed by the client-specified
338 .Bl -tag -width ".Dv BUS_DMA_UNLOCK"
340 Acquires and/or locks the client locking primitive.
341 .It Dv BUS_DMA_UNLOCK
342 Releases and/or unlocks the client locking primitive.
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"
350 Allocates a device specific DMA tag, and initializes it according to
351 the arguments provided:
352 .Bl -tag -width ".Fa filtfuncarg"
354 Indicates restrictions between the parent bridge, CPU memory, and the
356 Each device must use a master parent tag by calling
357 .Fn bus_get_dma_tag .
359 Alignment constraint, in bytes, of any mappings created using this tag.
360 The alignment must be a power of 2.
361 Hardware that can DMA starting at any address would specify
364 Hardware requiring DMA transfers to start on a multiple of 4K
368 Boundary constraint, in bytes, of the target DMA memory region.
369 The boundary indicates the set of addresses, all multiples of the
370 boundary argument, that cannot be crossed by a single
371 .Vt bus_dma_segment_t .
372 The boundary must be a power of 2 and must be no smaller than the
373 maximum segment size.
375 indicates that there are no boundary restrictions.
376 .It Fa lowaddr , highaddr
377 Bounds of the window of bus address space that
379 be directly accessed by the device.
380 The window contains all addresses greater than
382 and less than or equal to
384 For example, a device incapable of DMA above 4GB, would specify a
387 .Dv BUS_SPACE_MAXADDR
391 .Dv BUS_SPACE_MAXADDR_32BIT .
392 Similarly a device that can only perform DMA to addresses below
396 .Dv BUS_SPACE_MAXADDR
400 .Dv BUS_SPACE_MAXADDR_24BIT .
401 Some implementations requires that some region of device visible
402 address space, overlapping available host memory, be outside the
406 is used to bounce requests that would otherwise conflict with
407 the exclusion window.
409 Optional filter function (may be
411 to be called for any attempt to
412 map memory into the window described by
416 A filter function is only required when the single window described
421 cannot adequately describe the constraints of the device.
422 The filter function will be called for every machine page
423 that overlaps the exclusion window.
425 Argument passed to all calls to the filter function for this tag.
429 Maximum size, in bytes, of the sum of all segment lengths in a given
430 DMA mapping associated with this tag.
432 Number of discontinuities (scatter/gather segments) allowed
433 in a DMA mapped region.
434 If there is no restriction,
435 .Dv BUS_SPACE_UNRESTRICTED
438 Maximum size, in bytes, of a segment in any DMA mapped region associated
443 .Bl -tag -width ".Dv BUS_DMA_ALLOCNOW"
444 .It Dv BUS_DMA_ALLOCNOW
445 Pre-allocate enough resources to handle at least one map load operation on
447 If sufficient resources are not available,
450 This should not be used for tags that only describe buffers that will be
452 .Fn bus_dmamem_alloc .
453 Also, due to resource sharing with other tags, this flag does not guarantee
454 that resources will be allocated or reserved exclusively for this tag.
455 It should be treated only as a minor optimization.
458 Optional lock manipulation function (may be
460 to be called when busdma
461 needs to manipulate a lock on behalf of the client.
468 Optional argument to be passed to the function specified by
471 Pointer to a bus_dma_tag_t where the resulting DMA tag will
477 if sufficient memory is not available for tag creation
478 or allocating mapping resources.
479 .It Fn bus_dma_tag_destroy "dmat"
480 Deallocate the DMA tag
483 .Fn bus_dma_tag_create .
487 if any DMA maps remain associated with
492 .It Fn bus_dmamap_create "dmat" "flags" "*mapp"
493 Allocates and initializes a DMA map.
494 Arguments are as follows:
495 .Bl -tag -width ".Fa nsegments"
500 .Bl -tag -width ".Dv BUS_DMA_COHERENT"
501 .It Dv BUS_DMA_COHERENT
502 Attempt to map the memory loaded with this map such that cache sync
503 operations are as cheap as possible.
504 This flag is typically set on maps when the memory loaded with these will
505 be accessed by both a CPU and a DMA engine, frequently such as control data
506 and as opposed to streamable data such as receive and transmit buffers.
507 Use of this flag does not remove the requirement of using
508 .Fn bus_dmamap_sync ,
509 but it may reduce the cost of performing these operations.
511 .Fn bus_dmamap_create ,
514 flag is currently implemented on sparc64.
519 where the resulting DMA map will be stored.
524 if sufficient memory is not available for creating the
525 map or allocating mapping resources.
526 .It Fn bus_dmamap_destroy "dmat" "map"
527 Frees all resources associated with a given DMA map.
528 Arguments are as follows:
529 .Bl -tag -width ".Fa dmat"
531 DMA tag used to allocate
534 The DMA map to destroy.
539 if a mapping is still active for
541 .It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
542 "callback_arg" "flags"
543 Creates a mapping in device visible address space of
547 associated with the DMA map
549 This call will always return immediately and will not block for any reason.
550 Arguments are as follows:
551 .Bl -tag -width ".Fa buflen"
553 DMA tag used to allocate
556 A DMA map without a currently active mapping.
558 A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
559 mapped into device visible address space.
561 The size of the buffer.
562 .It Fa callback Fa callback_arg
563 The callback function, and its argument.
564 This function is called once sufficient mapping resources are available for
566 If resources are temporarily unavailable, this function will be deferred until
567 later, but the load operation will still return immediately to the caller.
568 Thus, callers should not assume that the callback will be called before the
569 load returns, and code should be structured appropriately to handle this.
570 See below for specific flags and error codes that control this behavior.
573 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
574 .It Dv BUS_DMA_NOWAIT
575 The load should not be deferred in case of insufficient mapping resources,
576 and instead should return immediately with an appropriate error.
577 .It Dv BUS_DMA_NOCACHE
578 The generated transactions to and from the virtual page are non-cacheable.
580 .Fn bus_dmamap_load ,
583 flag is currently implemented on sparc64.
587 Return values to the caller are as follows:
588 .Bl -tag -width ".Er EINPROGRESS"
590 The callback has been called and completed.
591 The status of the mapping has been delivered to the callback.
593 The mapping has been deferred for lack of resources.
594 The callback will be called as soon as resources are available.
595 Callbacks are serviced in FIFO order.
597 Note that subsequent load operations for the same tag that do not require
598 extra resources will still succeed.
599 This may result in out-of-order processing of requests.
600 If the caller requires the order of requests to be preserved,
601 then the caller is required to stall subsequent requests until a pending
602 request's callback is invoked.
604 The load request has failed due to insufficient resources, and the caller
605 specifically used the
609 The load request was invalid.
610 The callback has been called and has been provided the same error.
611 This error value may indicate that
621 argument used to create the dma tag
625 When the callback is called, it is presented with an error value
626 indicating the disposition of the mapping.
627 Error may be one of the following:
628 .Bl -tag -width ".Er EINPROGRESS"
630 The mapping was successful and the
632 callback argument contains an array of
633 .Vt bus_dma_segment_t
634 elements describing the mapping.
635 This array is only valid during the scope of the callback function.
637 A mapping could not be achieved within the segment constraints provided
638 in the tag even though the requested allocation size was less than maxsize.
640 .It Fn bus_dmamap_load_bio "dmat" "map" "bio" "callback" "callback_arg" "flags"
641 This is a variation of
643 which maps buffers pointed to by
647 may point to either a mapped or unmapped buffer.
648 .It Fn bus_dmamap_load_ccb "dmat" "map" "ccb" "callback" "callback_arg" "flags"
649 This is a variation of
651 which maps data pointed to by
656 may be any of the following types:
657 .Bl -tag -width ".Er CAM_DATA_SG_PADDR"
659 The data is a single KVA buffer.
661 The data is a single bus address range.
663 The data is a scatter/gather list of KVA buffers.
664 .It CAM_DATA_SG_PADDR
665 The data is a scatter/gather list of bus address ranges.
667 The data is contained in a
672 .Fn bus_dmamap_load_ccb
673 supports the following CCB XPT function codes:
675 .Bl -item -offset indent -compact
683 .It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
685 This is a variation of
687 which maps mbuf chains
691 argument is also passed to the callback routine, which
692 contains the mbuf chain's packet header length.
695 flag is implied, thus no callback deferral will happen.
697 Mbuf chains are assumed to be in kernel virtual address space.
699 Beside the error values listed for
700 .Fn bus_dmamap_load ,
702 will be returned if the size of the mbuf chain exceeds the maximum limit of the
704 .It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
706 .Fn bus_dmamap_load_mbuf
707 except that it returns immediately without calling a callback function.
708 It is provided for efficiency.
709 The scatter/gather segment array
711 is provided by the caller and filled in directly by the function.
714 argument is returned with the number of segments filled in.
715 Returns the same errors as
716 .Fn bus_dmamap_load_mbuf .
717 .It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
718 This is a variation of
720 which maps buffers pointed to by
725 argument is also passed to the callback routine, which contains the size of
731 flag is implied, thus no callback deferral will happen.
732 Returns the same errors as
733 .Fn bus_dmamap_load .
739 then it is assumed that the buffer,
742 .Fa "uio->uio_td->td_proc" Ns 's
744 User space memory must be in-core and wired prior to attempting a map
746 Pages may be locked using
748 .It Fn bus_dmamap_unload "dmat" "map"
750 Arguments are as follows:
751 .Bl -tag -width ".Fa dmam"
753 DMA tag used to allocate
756 The DMA map that is to be unloaded.
759 .Fn bus_dmamap_unload
760 will not perform any implicit synchronization of DMA buffers.
761 This must be done explicitly by a call to
763 prior to unloading the map.
764 .It Fn bus_dmamap_sync "dmat" "map" "op"
765 Performs synchronization of a device visible mapping with the CPU visible
766 memory referenced by that mapping.
767 Arguments are as follows:
768 .Bl -tag -width ".Fa dmat"
770 DMA tag used to allocate
773 The DMA mapping to be synchronized.
775 Type of synchronization operation to perform.
776 See the definition of
778 for a description of the acceptable values for
785 is the method used to ensure that CPU's and device's direct
786 memory access (DMA) to shared
788 For example, the CPU might be used to set up the contents of a buffer
789 that is to be made available to a device.
790 To ensure that the data are visible via the device's mapping of that
791 memory, the buffer must be loaded and a DMA sync operation of
792 .Dv BUS_DMASYNC_PREWRITE
793 must be performed after the CPU has updated the buffer and before the device
795 If the CPU modifies this buffer again later, another
796 .Dv BUS_DMASYNC_PREWRITE
797 sync operation must be performed before an additional device
799 Conversely, suppose a device updates memory that is to be read by a CPU.
800 In this case, the buffer must be loaded, and a DMA sync operation of
801 .Dv BUS_DMASYNC_PREREAD
802 must be performed before the device access is initiated.
803 The CPU will only be able to see the results of this memory update
804 once the DMA operation has completed and a
805 .Dv BUS_DMASYNC_POSTREAD
806 sync operation has been performed.
808 If read and write operations are not preceded and followed by the
809 appropriate synchronization operations, behavior is undefined.
810 .It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
811 Allocates memory that is mapped into KVA at the address returned
814 and that is permanently loaded into the newly created
818 Arguments are as follows:
819 .Bl -tag -width ".Fa alignment"
821 DMA tag describing the constraints of the DMA mapping.
823 Pointer to a pointer that will hold the returned KVA mapping of
824 the allocated region.
826 Flags are defined as follows:
827 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
828 .It Dv BUS_DMA_WAITOK
829 The routine can safely wait (sleep) for resources.
830 .It Dv BUS_DMA_NOWAIT
831 The routine is not allowed to wait for resources.
832 If resources are not available,
835 .It Dv BUS_DMA_COHERENT
836 Attempt to map this memory in a coherent fashion.
838 .Fn bus_dmamap_create
839 above for a description of this flag.
841 .Fn bus_dmamem_alloc ,
844 flag is currently implemented on arm and sparc64.
846 Causes the allocated memory to be set to all zeros.
847 .It Dv BUS_DMA_NOCACHE
848 The allocated memory will not be cached in the processor caches.
849 All memory accesses appear on the bus and are executed
852 .Fn bus_dmamem_alloc ,
855 flag is currently implemented on amd64 and i386 where it results in the
856 Strong Uncacheable PAT to be set for the allocated virtual address range.
861 where the resulting DMA map will be stored.
864 The size of memory to be allocated is
866 as specified in the call to
867 .Fn bus_dma_tag_create
871 The current implementation of
873 will allocate all requests as a single segment.
875 An initial load operation is required to obtain the bus address of the allocated
876 memory, and an unload operation is required before freeing the memory, as
878 .Fn bus_dmamem_free .
879 Maps are automatically handled by this function and should not be explicitly
880 allocated or destroyed.
882 Although an explicit load is not required for each access to the memory
883 referenced by the returned map, the synchronization requirements
886 section still apply and should be used to achieve portability on architectures
887 without coherent buses.
891 if sufficient memory is not available for completing
893 .It Fn bus_dmamem_free "dmat" "*vaddr" "map"
894 Frees memory previously allocated by
895 .Fn bus_dmamem_alloc .
898 Arguments are as follows:
899 .Bl -tag -width ".Fa vaddr"
903 Kernel virtual address of the memory.
905 DMA map to be invalidated.
909 Behavior is undefined if invalid arguments are passed to
910 any of the above functions.
911 If sufficient resources cannot be allocated for a given
916 routines that are not of type
918 will return 0 on success or an error
919 code on failure as discussed above.
923 routines will succeed if provided with valid arguments.
925 Two locking protocols are used by
927 The first is a private global lock that is used to synchronize access to the
928 bounce buffer pool on the architectures that make use of them.
929 This lock is strictly a leaf lock that is only used internally to
931 and is not exposed to clients of the API.
933 The second protocol involves protecting various resources stored in the tag.
936 operations are done through requests from the driver that created the tag,
937 the most efficient way to protect the tag resources is through the lock that
941 acts on its own without being called by the driver, the lock primitive
942 specified in the tag is acquired and released automatically.
943 An example of this is when the
945 callback function is called from a deferred context instead of the driver
947 This means that certain
949 functions must always be called with the same lock held that is specified in the
951 These functions include:
953 .Bl -item -offset indent -compact
957 .Fn bus_dmamap_load_bio
959 .Fn bus_dmamap_load_ccb
961 .Fn bus_dmamap_load_mbuf
963 .Fn bus_dmamap_load_mbuf_sg
965 .Fn bus_dmamap_load_uio
967 .Fn bus_dmamap_unload
972 There is one exception to this rule.
973 It is common practice to call some of these functions during driver start-up
974 without any locks held.
975 So long as there is a guarantee of no possible concurrent use of the tag by
976 different threads during this operation, it is safe to not hold a lock for
981 operations should not be called with the driver lock held, either because
982 they are already protected by an internal lock, or because they might sleep
983 due to memory or resource allocation.
984 The following functions must not be
985 called with any non-sleepable locks held:
987 .Bl -item -offset indent -compact
989 .Fn bus_dma_tag_create
991 .Fn bus_dmamap_create
996 All other functions do not have a locking protocol and can thus be
997 called with or without any system or driver locks held.
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"
1015 interface first appeared in
1020 API was adopted from
1022 for use in the CAM SCSI subsystem.
1023 The alterations to the original API were aimed to remove the need for
1025 .Vt bus_dma_segment_t
1026 array stored in each
1028 while allowing callers to queue up on scarce resources.
1032 interface was designed and implemented by
1034 of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
1035 Additional input on the
1037 design was provided by
1039 .An Chris Demetriou ,
1040 .An Charles Hannum ,
1043 .An Jonathan Stone ,
1051 benefits from the contributions of
1052 .An Justin T. Gibbs ,
1055 .An Matthew N. Dodd ,
1057 .An Maxime Henrion ,
1058 .An Jake Burkholder ,
1059 .An Takahashi Yoshihiro ,
1063 This manual page was written by
1066 .An Justin T. Gibbs .