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.
41 .\" THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
42 .\" ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
43 .\" TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
44 .\" PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
45 .\" BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
46 .\" CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
47 .\" SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
48 .\" INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
49 .\" CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
50 .\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
51 .\" POSSIBILITY OF SUCH DAMAGE.
54 .\" $NetBSD: bus_dma.9,v 1.25 2002/10/14 13:43:16 wiz Exp $
61 .Nm bus_dma_tag_create ,
62 .Nm bus_dma_tag_destroy ,
63 .Nm bus_dmamap_create ,
64 .Nm bus_dmamap_destroy ,
66 .Nm bus_dmamap_load_bio ,
67 .Nm bus_dmamap_load_ccb ,
68 .Nm bus_dmamap_load_mbuf ,
69 .Nm bus_dmamap_load_mbuf_sg ,
70 .Nm bus_dmamap_load_uio ,
71 .Nm bus_dmamap_unload ,
73 .Nm bus_dmamem_alloc ,
75 .Nd Bus and Machine Independent DMA Mapping Interface
79 .Fn bus_dma_tag_create "bus_dma_tag_t parent" "bus_size_t alignment" \
80 "bus_addr_t boundary" "bus_addr_t lowaddr" "bus_addr_t highaddr" \
81 "bus_dma_filter_t *filtfunc" "void *filtfuncarg" "bus_size_t maxsize" \
82 "int nsegments" "bus_size_t maxsegsz" "int flags" "bus_dma_lock_t *lockfunc" \
83 "void *lockfuncarg" "bus_dma_tag_t *dmat"
85 .Fn bus_dma_tag_destroy "bus_dma_tag_t dmat"
87 .Fn bus_dmamap_create "bus_dma_tag_t dmat" "int flags" "bus_dmamap_t *mapp"
89 .Fn bus_dmamap_destroy "bus_dma_tag_t dmat" "bus_dmamap_t map"
91 .Fn bus_dmamap_load "bus_dma_tag_t dmat" "bus_dmamap_t map" "void *buf" \
92 "bus_size_t buflen" "bus_dmamap_callback_t *callback" "void *callback_arg" \
95 .Fn bus_dmamap_load_bio "bus_dma_tag_t dmat" "bus_dmamap_t map" \
96 "struct bio *bio" "bus_dmamap_callback_t *callback" "void *callback_arg" \
99 .Fn bus_dmamap_load_ccb "bus_dma_tag_t dmat" "bus_dmamap_t map" \
100 "union ccb *ccb" "bus_dmamap_callback_t *callback" "void *callback_arg" \
103 .Fn bus_dmamap_load_mbuf "bus_dma_tag_t dmat" "bus_dmamap_t map" \
104 "struct mbuf *mbuf" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
107 .Fn bus_dmamap_load_mbuf_sg "bus_dma_tag_t dmat" "bus_dmamap_t map" \
108 "struct mbuf *mbuf" "bus_dma_segment_t *segs" "int *nsegs" "int flags"
110 .Fn bus_dmamap_load_uio "bus_dma_tag_t dmat" "bus_dmamap_t map" \
111 "struct uio *uio" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
114 .Fn bus_dmamap_unload "bus_dma_tag_t dmat" "bus_dmamap_t map"
116 .Fn bus_dmamap_sync "bus_dma_tag_t dmat" "bus_dmamap_t map" \
119 .Fn bus_dmamem_alloc "bus_dma_tag_t dmat" "void **vaddr" \
120 "int flags" "bus_dmamap_t *mapp"
122 .Fn bus_dmamem_free "bus_dma_tag_t dmat" "void *vaddr" \
125 Direct Memory Access (DMA) is a method of transferring data
126 without involving the CPU, thus providing higher performance.
127 A DMA transaction can be achieved between device to memory,
128 device to device, or memory to memory.
132 API is a bus, device, and machine-independent (MI) interface to
134 It provides the client with flexibility and simplicity by
135 abstracting machine dependent issues like setting up
136 DMA mappings, handling cache issues, bus specific features
140 .Vt ( bus_dma_tag_t )
141 is used to describe the properties of a group of related DMA
143 One way to view this is that a tag describes the limitations of a DMA engine.
144 For example, if a DMA engine in a device is limited to 32-bit addresses,
145 that limitation is specified by a parameter when creating the tag
147 Similarly, a tag can be marked as requiring buffers whose addresses are
148 aligned to a specific boundary.
150 Some devices may require multiple tags to describe DMA
151 transactions with differing properties.
152 For example, a device might require 16-byte alignment of its descriptor ring
153 while permitting arbitrary alignment of I/O buffers.
155 the driver must create one tag for the descriptor ring and a separate tag for
157 If a device has restrictions that are common to all DMA transactions
158 in addition to restrictions that differ between unrelated groups of
160 the driver can first create a
162 tag that decribes the common restrictions.
163 The per-group tags can then inherit these restrictions from this
165 tag rather than having to list them explicitly when creating the per-group tags.
169 represents a mapping of a memory region for DMA.
170 On systems with I/O MMUs,
171 the mapping structure tracks any I/O MMU entries used by a request.
172 For DMA requests that require bounce pages,
173 the mapping tracks the bounce pages used.
175 To prepare for one or more DMA transactions,
176 a mapping must be bound to a memory region by calling one of the
179 These functions configure the mapping which can include programming entries
180 in an I/O MMU and/or allocating bounce pages.
181 An output of these functions
182 (either directly or indirectly by invoking a callback routine)
183 is the list of scatter/gather address ranges a consumer can pass to a DMA
184 engine to access the memory region.
185 When a mapping is no longer needed,
186 the mapping must be unloaded via
187 .Fn bus_dmamap_unload .
189 Before and after each DMA transaction,
191 must be used to ensure that the correct data is used by the DMA engine and
193 If a mapping uses bounce pages,
194 the sync operations copy data between the bounce pages and the memory region
195 bound to the mapping.
196 Sync operations also handle architecture-specific details such as CPU cache
197 flushing and CPU memory operation ordering.
198 .Sh STATIC VS DYNAMIC
200 handles two types of DMA transactions: static and dynamic.
201 Static transactions are used with a long-lived memory region that is reused
202 for many transactions such as a descriptor ring.
203 Dynamic transactions are used for transfers to or from transient buffers
204 such as I/O buffers holding a network packet or disk block.
205 Each transaction type uses a different subset of the
208 .Ss Static Transactions
209 Static transactions use memory regions allocated by
211 Each static memory region is allocated by calling
212 .Fn bus_dmamem_alloc .
213 This function requires a valid tag describing the properties of the
214 DMA transactions to this region such as alignment or address restrictions.
215 Multiple regions can share a single tag if they share the same restrictions.
218 allocates a memory region along with a mapping object.
219 The associated tag, memory region, and mapping object must then be passed to
221 to bind the mapping to the allocated region and obtain the
226 will attempt to allocate memory requiring less expensive sync operations
227 (for example, implementations should not allocate regions requiring bounce
229 but sync operations should still be used.
230 For example, a driver should use
232 in an interrupt handler before reading descriptor ring entries written by the
233 device prior to the interrupt.
235 When a consumer is finished with a memory region,
236 it should unload the mapping via
237 .Fn bus_dmamap_unload
238 and then release the memory region and mapping object via
239 .Fn bus_dmamem_free .
240 .Ss Dynamic Transactions
241 Dynamic transactions map memory regions provided by other parts of the system.
242 A tag must be created via
243 .Fn bus_dma_tag_create
244 to describe the DMA transactions to and from these memory regions,
245 and a pool of mapping objects must be allocated via
246 .Fn bus_dmamap_create
247 to track the mappings of any in-flight transactions.
249 When a consumer wishes to schedule a transaction for a memory region,
250 the consumer must first obtain an unused mapping object from its pool
252 The memory region must be bound to the mapping object via one of the
255 Before scheduling the transaction,
256 the consumer should sync the memory region via
258 with one or more of the
261 After the transaction has completed,
262 the consumer should sync the memory region via
264 with one or more of the
267 The mapping can then be unloaded via
268 .Fn bus_dmamap_unload ,
269 and the mapping object can be returned to the pool of unused mapping objects.
271 When a consumer is no longer scheduling DMA transactions,
272 the mapping objects should be freed via
273 .Fn bus_dmamap_destroy ,
274 and the tag should be freed via
275 .Fn bus_dma_tag_destroy .
276 .Sh STRUCTURES AND TYPES
277 .Bl -tag -width indent
279 A machine-dependent (MD) opaque type that describes the
280 characteristics of a group of DMA transactions.
281 DMA tags are organized into a hierarchy, with each child
282 tag inheriting the restrictions of its parent.
283 This allows all devices along the path of DMA transactions
284 to contribute to the constraints of those transactions.
285 .It Vt bus_dma_filter_t
286 Client specified address filter having the format:
287 .Bl -tag -width indent
289 .Fn "client_filter" "void *filtarg" "bus_addr_t testaddr"
292 Address filters can be specified during tag creation to allow
293 for devices whose DMA address restrictions cannot be specified
297 argument is specified by the client during tag creation to be passed to all
298 invocations of the callback.
301 argument contains a potential starting address of a DMA mapping.
302 The filter function operates on the set of addresses from
305 .Ql trunc_page(testaddr) + PAGE_SIZE - 1 ,
307 The filter function should return zero if any mapping in this range
308 can be accommodated by the device and non-zero otherwise.
309 .It Vt bus_dma_segment_t
310 A machine-dependent type that describes individual
312 It contains the following fields:
320 field contains the device visible address of the DMA segment, and
322 contains the length of the DMA segment.
323 Although the DMA segments returned by a mapping call will adhere to
324 all restrictions necessary for a successful DMA operation, some conversion
325 (e.g.\& a conversion from host byte order to the device's byte order) is
326 almost always required when presenting segment information to the device.
328 A machine-dependent opaque type describing an individual mapping.
329 One map is used for each memory allocation that will be loaded.
330 Maps can be reused once they have been unloaded.
331 Multiple maps can be associated with one DMA tag.
332 While the value of the map may evaluate to
334 on some platforms under certain conditions,
335 it should never be assumed that it will be
338 .It Vt bus_dmamap_callback_t
339 Client specified callback for receiving mapping information resulting from
343 .Fn bus_dmamap_load ,
344 .Fn bus_dmamap_load_bio
346 .Fn bus_dmamap_load_ccb .
347 Callbacks are of the format:
348 .Bl -tag -width indent
350 .Fn "client_callback" "void *callback_arg" "bus_dma_segment_t *segs" \
351 "int nseg" "int error"
356 is the callback argument passed to dmamap load functions.
361 arguments describe an array of
362 .Vt bus_dma_segment_t
363 structures that represent the mapping.
364 This array is only valid within the scope of the callback function.
365 The success or failure of the mapping is indicated by the
368 More information on the use of callbacks can be found in the
369 description of the individual dmamap load functions.
370 .It Vt bus_dmamap_callback2_t
371 Client specified callback for receiving mapping information resulting from
375 .Fn bus_dmamap_load_uio
377 .Fn bus_dmamap_load_mbuf .
379 Callback2s are of the format:
380 .Bl -tag -width indent
382 .Fn "client_callback2" "void *callback_arg" "bus_dma_segment_t *segs" \
383 "int nseg" "bus_size_t mapsize" "int error"
386 Callback2's behavior is the same as
387 .Vt bus_dmamap_callback_t
388 with the addition that the length of the data mapped is provided via
390 .It Vt bus_dmasync_op_t
391 Memory synchronization operation specifier.
392 Bus DMA requires explicit synchronization of memory with its device
393 visible mapping in order to guarantee memory coherency.
396 allows the type of DMA operation that will be or has been performed
397 to be communicated to the system so that the correct coherency measures
399 The operations are represented as bitfield flags that can be combined together,
400 though it only makes sense to combine PRE flags or POST flags, not both.
403 description below for more details on how to use these operations.
405 All operations specified below are performed from the host memory point of view,
406 where a read implies data coming from the device to the host memory, and a write
407 implies data going from the host memory to the device.
408 Alternatively, the operations can be thought of in terms of driver operations,
409 where reading a network packet or storage sector corresponds to a read operation
412 .Bl -tag -width ".Dv BUS_DMASYNC_POSTWRITE"
413 .It Dv BUS_DMASYNC_PREREAD
414 Perform any synchronization required prior to an update of host memory by the
416 .It Dv BUS_DMASYNC_PREWRITE
417 Perform any synchronization required after an update of host memory by the CPU
418 and prior to device access to host memory.
419 .It Dv BUS_DMASYNC_POSTREAD
420 Perform any synchronization required after an update of host memory by the
421 device and prior to CPU access to host memory.
422 .It Dv BUS_DMASYNC_POSTWRITE
423 Perform any synchronization required after device access to host memory.
425 .It Vt bus_dma_lock_t
426 Client specified lock/mutex manipulation method.
427 This will be called from
428 within busdma whenever a client lock needs to be manipulated.
429 In its current form, the function will be called immediately before
430 the callback for a DMA load operation that has been deferred with
432 and immediately after with
434 If the load operation does not need to be deferred, then it
435 will not be called since the function loading the map should
436 be holding the appropriate locks.
437 This method is of the format:
438 .Bl -tag -width indent
440 .Fn "lockfunc" "void *lockfunc_arg" "bus_dma_lock_op_t op"
445 argument is specified by the client during tag creation to be passed to all
446 invocations of the callback.
449 argument specifies the lock operation to perform.
453 implementations are provided for convenience.
454 .Fn busdma_lock_mutex
455 performs standard mutex operations on the sleep mutex provided via
458 will generate a system panic if it is called.
459 It is substituted into the tag when
464 .Fn bus_dma_tag_create
465 and is useful for tags that should not be used with deferred load operations.
466 .It Vt bus_dma_lock_op_t
467 Operations to be performed by the client-specified
469 .Bl -tag -width ".Dv BUS_DMA_UNLOCK"
471 Acquires and/or locks the client locking primitive.
472 .It Dv BUS_DMA_UNLOCK
473 Releases and/or unlocks the client locking primitive.
477 .Bl -tag -width indent
478 .It Fn bus_dma_tag_create "parent" "alignment" "boundary" "lowaddr" \
479 "highaddr" "*filtfunc" "*filtfuncarg" "maxsize" "nsegments" "maxsegsz" \
480 "flags" "lockfunc" "lockfuncarg" "*dmat"
481 Allocates a DMA tag, and initializes it according to
482 the arguments provided:
483 .Bl -tag -width ".Fa filtfuncarg"
485 A parent tag from which to inherit restrictions.
486 The restrictions passed in other arguments can only further tighten the
487 restrictions inherited from the parent tag.
489 All tags created by a device driver must inherit from the tag returned by
491 to honor restrictions between the parent bridge, CPU memory, and the
494 Alignment constraint, in bytes, of any mappings created using this tag.
495 The alignment must be a power of 2.
496 Hardware that can DMA starting at any address would specify
499 Hardware requiring DMA transfers to start on a multiple of 4K
503 Boundary constraint, in bytes, of the target DMA memory region.
504 The boundary indicates the set of addresses, all multiples of the
505 boundary argument, that cannot be crossed by a single
506 .Vt bus_dma_segment_t .
507 The boundary must be a power of 2 and must be no smaller than the
508 maximum segment size.
510 indicates that there are no boundary restrictions.
511 .It Fa lowaddr , highaddr
512 Bounds of the window of bus address space that
514 be directly accessed by the device.
515 The window contains all addresses greater than
517 and less than or equal to
519 For example, a device incapable of DMA above 4GB, would specify a
522 .Dv BUS_SPACE_MAXADDR
526 .Dv BUS_SPACE_MAXADDR_32BIT .
527 Similarly a device that can only perform DMA to addresses below
531 .Dv BUS_SPACE_MAXADDR
535 .Dv BUS_SPACE_MAXADDR_24BIT .
536 Some implementations require that some region of device visible
537 address space, overlapping available host memory, be outside the
541 is used to bounce requests that would otherwise conflict with
542 the exclusion window.
544 Optional filter function (may be
546 to be called for any attempt to
547 map memory into the window described by
551 A filter function is only required when the single window described
556 cannot adequately describe the constraints of the device.
557 The filter function will be called for every machine page
558 that overlaps the exclusion window.
560 Argument passed to all calls to the filter function for this tag.
564 Maximum size, in bytes, of the sum of all segment lengths in a given
565 DMA mapping associated with this tag.
567 Number of discontinuities (scatter/gather segments) allowed
568 in a DMA mapped region.
569 If there is no restriction,
570 .Dv BUS_SPACE_UNRESTRICTED
573 Maximum size, in bytes, of a segment in any DMA mapped region associated
578 .Bl -tag -width ".Dv BUS_DMA_ALLOCNOW"
579 .It Dv BUS_DMA_ALLOCNOW
580 Pre-allocate enough resources to handle at least one map load operation on
582 If sufficient resources are not available,
585 This should not be used for tags that only describe buffers that will be
587 .Fn bus_dmamem_alloc .
588 Also, due to resource sharing with other tags, this flag does not guarantee
589 that resources will be allocated or reserved exclusively for this tag.
590 It should be treated only as a minor optimization.
593 Optional lock manipulation function (may be
595 to be called when busdma
596 needs to manipulate a lock on behalf of the client.
603 Optional argument to be passed to the function specified by
606 Pointer to a bus_dma_tag_t where the resulting DMA tag will
612 if sufficient memory is not available for tag creation
613 or allocating mapping resources.
614 .It Fn bus_dma_tag_destroy "dmat"
615 Deallocate the DMA tag
618 .Fn bus_dma_tag_create .
622 if any DMA maps remain associated with
627 .It Fn bus_dmamap_create "dmat" "flags" "*mapp"
628 Allocates and initializes a DMA map.
629 Arguments are as follows:
630 .Bl -tag -width ".Fa nsegments"
635 .Bl -tag -width ".Dv BUS_DMA_COHERENT"
636 .It Dv BUS_DMA_COHERENT
637 Attempt to map the memory loaded with this map such that cache sync
638 operations are as cheap as possible.
639 This flag is typically set on maps when the memory loaded with these will
640 be accessed by both a CPU and a DMA engine, frequently such as control data
641 and as opposed to streamable data such as receive and transmit buffers.
642 Use of this flag does not remove the requirement of using
643 .Fn bus_dmamap_sync ,
644 but it may reduce the cost of performing these operations.
646 .Fn bus_dmamap_create ,
649 flag is currently implemented on sparc64.
654 where the resulting DMA map will be stored.
659 if sufficient memory is not available for creating the
660 map or allocating mapping resources.
661 .It Fn bus_dmamap_destroy "dmat" "map"
662 Frees all resources associated with a given DMA map.
663 Arguments are as follows:
664 .Bl -tag -width ".Fa dmat"
666 DMA tag used to allocate
669 The DMA map to destroy.
674 if a mapping is still active for
676 .It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
677 "callback_arg" "flags"
678 Creates a mapping in device visible address space of
682 associated with the DMA map
684 This call will always return immediately and will not block for any reason.
685 Arguments are as follows:
686 .Bl -tag -width ".Fa buflen"
688 DMA tag used to allocate
691 A DMA map without a currently active mapping.
693 A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
694 mapped into device visible address space.
696 The size of the buffer.
697 .It Fa callback Fa callback_arg
698 The callback function, and its argument.
699 This function is called once sufficient mapping resources are available for
701 If resources are temporarily unavailable, this function will be deferred until
702 later, but the load operation will still return immediately to the caller.
703 Thus, callers should not assume that the callback will be called before the
704 load returns, and code should be structured appropriately to handle this.
705 See below for specific flags and error codes that control this behavior.
708 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
709 .It Dv BUS_DMA_NOWAIT
710 The load should not be deferred in case of insufficient mapping resources,
711 and instead should return immediately with an appropriate error.
712 .It Dv BUS_DMA_NOCACHE
713 The generated transactions to and from the virtual page are non-cacheable.
715 .Fn bus_dmamap_load ,
718 flag is currently implemented on sparc64.
722 Return values to the caller are as follows:
723 .Bl -tag -width ".Er EINPROGRESS"
725 The callback has been called and completed.
726 The status of the mapping has been delivered to the callback.
728 The mapping has been deferred for lack of resources.
729 The callback will be called as soon as resources are available.
730 Callbacks are serviced in FIFO order.
732 Note that subsequent load operations for the same tag that do not require
733 extra resources will still succeed.
734 This may result in out-of-order processing of requests.
735 If the caller requires the order of requests to be preserved,
736 then the caller is required to stall subsequent requests until a pending
737 request's callback is invoked.
739 The load request has failed due to insufficient resources, and the caller
740 specifically used the
744 The load request was invalid.
745 The callback has been called and has been provided the same error.
746 This error value may indicate that
756 argument used to create the dma tag
760 When the callback is called, it is presented with an error value
761 indicating the disposition of the mapping.
762 Error may be one of the following:
763 .Bl -tag -width ".Er EINPROGRESS"
765 The mapping was successful and the
767 callback argument contains an array of
768 .Vt bus_dma_segment_t
769 elements describing the mapping.
770 This array is only valid during the scope of the callback function.
772 A mapping could not be achieved within the segment constraints provided
773 in the tag even though the requested allocation size was less than maxsize.
775 .It Fn bus_dmamap_load_bio "dmat" "map" "bio" "callback" "callback_arg" "flags"
776 This is a variation of
778 which maps buffers pointed to by
782 may point to either a mapped or unmapped buffer.
783 .It Fn bus_dmamap_load_ccb "dmat" "map" "ccb" "callback" "callback_arg" "flags"
784 This is a variation of
786 which maps data pointed to by
791 may be any of the following types:
792 .Bl -tag -width ".Er CAM_DATA_SG_PADDR"
794 The data is a single KVA buffer.
796 The data is a single bus address range.
798 The data is a scatter/gather list of KVA buffers.
799 .It CAM_DATA_SG_PADDR
800 The data is a scatter/gather list of bus address ranges.
802 The data is contained in a
807 .Fn bus_dmamap_load_ccb
808 supports the following CCB XPT function codes:
810 .Bl -item -offset indent -compact
818 .It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
820 This is a variation of
822 which maps mbuf chains
826 argument is also passed to the callback routine, which
827 contains the mbuf chain's packet header length.
830 flag is implied, thus no callback deferral will happen.
832 Mbuf chains are assumed to be in kernel virtual address space.
834 Beside the error values listed for
835 .Fn bus_dmamap_load ,
837 will be returned if the size of the mbuf chain exceeds the maximum limit of the
839 .It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
841 .Fn bus_dmamap_load_mbuf
842 except that it returns immediately without calling a callback function.
843 It is provided for efficiency.
844 The scatter/gather segment array
846 is provided by the caller and filled in directly by the function.
849 argument is returned with the number of segments filled in.
850 Returns the same errors as
851 .Fn bus_dmamap_load_mbuf .
852 .It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
853 This is a variation of
855 which maps buffers pointed to by
860 argument is also passed to the callback routine, which contains the size of
866 flag is implied, thus no callback deferral will happen.
867 Returns the same errors as
868 .Fn bus_dmamap_load .
874 then it is assumed that the buffer,
877 .Fa "uio->uio_td->td_proc" Ns 's
879 User space memory must be in-core and wired prior to attempting a map
881 Pages may be locked using
883 .It Fn bus_dmamap_unload "dmat" "map"
885 Arguments are as follows:
886 .Bl -tag -width ".Fa dmam"
888 DMA tag used to allocate
891 The DMA map that is to be unloaded.
894 .Fn bus_dmamap_unload
895 will not perform any implicit synchronization of DMA buffers.
896 This must be done explicitly by a call to
898 prior to unloading the map.
899 .It Fn bus_dmamap_sync "dmat" "map" "op"
900 Performs synchronization of a device visible mapping with the CPU visible
901 memory referenced by that mapping.
902 Arguments are as follows:
903 .Bl -tag -width ".Fa dmat"
905 DMA tag used to allocate
908 The DMA mapping to be synchronized.
910 Type of synchronization operation to perform.
911 See the definition of
913 for a description of the acceptable values for
920 is the method used to ensure that CPU's and device's direct
921 memory access (DMA) to shared
923 For example, the CPU might be used to set up the contents of a buffer
924 that is to be made available to a device.
925 To ensure that the data are visible via the device's mapping of that
926 memory, the buffer must be loaded and a DMA sync operation of
927 .Dv BUS_DMASYNC_PREWRITE
928 must be performed after the CPU has updated the buffer and before the device
930 If the CPU modifies this buffer again later, another
931 .Dv BUS_DMASYNC_PREWRITE
932 sync operation must be performed before an additional device
934 Conversely, suppose a device updates memory that is to be read by a CPU.
935 In this case, the buffer must be loaded, and a DMA sync operation of
936 .Dv BUS_DMASYNC_PREREAD
937 must be performed before the device access is initiated.
938 The CPU will only be able to see the results of this memory update
939 once the DMA operation has completed and a
940 .Dv BUS_DMASYNC_POSTREAD
941 sync operation has been performed.
943 If read and write operations are not preceded and followed by the
944 appropriate synchronization operations, behavior is undefined.
945 .It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
946 Allocates memory that is mapped into KVA at the address returned
949 and that is permanently loaded into the newly created
953 Arguments are as follows:
954 .Bl -tag -width ".Fa alignment"
956 DMA tag describing the constraints of the DMA mapping.
958 Pointer to a pointer that will hold the returned KVA mapping of
959 the allocated region.
961 Flags are defined as follows:
962 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
963 .It Dv BUS_DMA_WAITOK
964 The routine can safely wait (sleep) for resources.
965 .It Dv BUS_DMA_NOWAIT
966 The routine is not allowed to wait for resources.
967 If resources are not available,
970 .It Dv BUS_DMA_COHERENT
971 Attempt to map this memory in a coherent fashion.
973 .Fn bus_dmamap_create
974 above for a description of this flag.
976 .Fn bus_dmamem_alloc ,
979 flag is currently implemented on arm and sparc64.
981 Causes the allocated memory to be set to all zeros.
982 .It Dv BUS_DMA_NOCACHE
983 The allocated memory will not be cached in the processor caches.
984 All memory accesses appear on the bus and are executed
987 .Fn bus_dmamem_alloc ,
990 flag is currently implemented on amd64 and i386 where it results in the
991 Strong Uncacheable PAT to be set for the allocated virtual address range.
996 where the resulting DMA map will be stored.
999 The size of memory to be allocated is
1001 as specified in the call to
1002 .Fn bus_dma_tag_create
1006 The current implementation of
1007 .Fn bus_dmamem_alloc
1008 will allocate all requests as a single segment.
1010 An initial load operation is required to obtain the bus address of the allocated
1011 memory, and an unload operation is required before freeing the memory, as
1013 .Fn bus_dmamem_free .
1014 Maps are automatically handled by this function and should not be explicitly
1015 allocated or destroyed.
1017 Although an explicit load is not required for each access to the memory
1018 referenced by the returned map, the synchronization requirements
1021 section still apply and should be used to achieve portability on architectures
1022 without coherent buses.
1026 if sufficient memory is not available for completing
1028 .It Fn bus_dmamem_free "dmat" "*vaddr" "map"
1029 Frees memory previously allocated by
1030 .Fn bus_dmamem_alloc .
1032 will be invalidated.
1033 Arguments are as follows:
1034 .Bl -tag -width ".Fa vaddr"
1038 Kernel virtual address of the memory.
1040 DMA map to be invalidated.
1044 Behavior is undefined if invalid arguments are passed to
1045 any of the above functions.
1046 If sufficient resources cannot be allocated for a given
1051 routines that are not of type
1053 will return 0 on success or an error
1054 code on failure as discussed above.
1058 routines will succeed if provided with valid arguments.
1060 Two locking protocols are used by
1062 The first is a private global lock that is used to synchronize access to the
1063 bounce buffer pool on the architectures that make use of them.
1064 This lock is strictly a leaf lock that is only used internally to
1066 and is not exposed to clients of the API.
1068 The second protocol involves protecting various resources stored in the tag.
1071 operations are done through requests from the driver that created the tag,
1072 the most efficient way to protect the tag resources is through the lock that
1076 acts on its own without being called by the driver, the lock primitive
1077 specified in the tag is acquired and released automatically.
1078 An example of this is when the
1080 callback function is called from a deferred context instead of the driver
1082 This means that certain
1084 functions must always be called with the same lock held that is specified in the
1086 These functions include:
1088 .Bl -item -offset indent -compact
1092 .Fn bus_dmamap_load_bio
1094 .Fn bus_dmamap_load_ccb
1096 .Fn bus_dmamap_load_mbuf
1098 .Fn bus_dmamap_load_mbuf_sg
1100 .Fn bus_dmamap_load_uio
1102 .Fn bus_dmamap_unload
1107 There is one exception to this rule.
1108 It is common practice to call some of these functions during driver start-up
1109 without any locks held.
1110 So long as there is a guarantee of no possible concurrent use of the tag by
1111 different threads during this operation, it is safe to not hold a lock for
1116 operations should not be called with the driver lock held, either because
1117 they are already protected by an internal lock, or because they might sleep
1118 due to memory or resource allocation.
1119 The following functions must not be
1120 called with any non-sleepable locks held:
1122 .Bl -item -offset indent -compact
1124 .Fn bus_dma_tag_create
1126 .Fn bus_dmamap_create
1128 .Fn bus_dmamem_alloc
1131 All other functions do not have a locking protocol and can thus be
1132 called with or without any system or driver locks held.
1141 .%A "Jason R. Thorpe"
1142 .%T "A Machine-Independent DMA Framework for NetBSD"
1143 .%J "Proceedings of the Summer 1998 USENIX Technical Conference"
1144 .%Q "USENIX Association"
1150 interface first appeared in
1155 API was adopted from
1157 for use in the CAM SCSI subsystem.
1158 The alterations to the original API were aimed to remove the need for
1160 .Vt bus_dma_segment_t
1161 array stored in each
1163 while allowing callers to queue up on scarce resources.
1167 interface was designed and implemented by
1169 of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
1170 Additional input on the
1172 design was provided by
1174 .An Chris Demetriou ,
1175 .An Charles Hannum ,
1178 .An Jonathan Stone ,
1186 benefits from the contributions of
1187 .An Justin T. Gibbs ,
1190 .An Matthew N. Dodd ,
1192 .An Maxime Henrion ,
1193 .An Jake Burkholder ,
1194 .An Takahashi Yoshihiro ,
1198 This manual page was written by
1201 .An Justin T. Gibbs .