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.
591 .It Dv BUS_DMA_COHERENT
592 Indicate that the DMA engine and CPU are cache-coherent.
593 Cached memory may be used to back allocations created by
594 .Fn bus_dmamem_alloc .
596 .Fn bus_dma_tag_create ,
599 flag is currently implemented on arm64.
602 Optional lock manipulation function (may be
604 to be called when busdma
605 needs to manipulate a lock on behalf of the client.
612 Optional argument to be passed to the function specified by
615 Pointer to a bus_dma_tag_t where the resulting DMA tag will
621 if sufficient memory is not available for tag creation
622 or allocating mapping resources.
623 .It Fn bus_dma_tag_destroy "dmat"
624 Deallocate the DMA tag
627 .Fn bus_dma_tag_create .
631 if any DMA maps remain associated with
636 .It Fn bus_dmamap_create "dmat" "flags" "*mapp"
637 Allocates and initializes a DMA map.
638 Arguments are as follows:
639 .Bl -tag -width ".Fa nsegments"
644 .Bl -tag -width ".Dv BUS_DMA_COHERENT"
645 .It Dv BUS_DMA_COHERENT
646 Attempt to map the memory loaded with this map such that cache sync
647 operations are as cheap as possible.
648 This flag is typically set on maps when the memory loaded with these will
649 be accessed by both a CPU and a DMA engine, frequently such as control data
650 and as opposed to streamable data such as receive and transmit buffers.
651 Use of this flag does not remove the requirement of using
652 .Fn bus_dmamap_sync ,
653 but it may reduce the cost of performing these operations.
655 .Fn bus_dmamap_create ,
658 flag is currently implemented on sparc64.
663 where the resulting DMA map will be stored.
668 if sufficient memory is not available for creating the
669 map or allocating mapping resources.
670 .It Fn bus_dmamap_destroy "dmat" "map"
671 Frees all resources associated with a given DMA map.
672 Arguments are as follows:
673 .Bl -tag -width ".Fa dmat"
675 DMA tag used to allocate
678 The DMA map to destroy.
683 if a mapping is still active for
685 .It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
686 "callback_arg" "flags"
687 Creates a mapping in device visible address space of
691 associated with the DMA map
693 This call will always return immediately and will not block for any reason.
694 Arguments are as follows:
695 .Bl -tag -width ".Fa buflen"
697 DMA tag used to allocate
700 A DMA map without a currently active mapping.
702 A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
703 mapped into device visible address space.
705 The size of the buffer.
706 .It Fa callback Fa callback_arg
707 The callback function, and its argument.
708 This function is called once sufficient mapping resources are available for
710 If resources are temporarily unavailable, this function will be deferred until
711 later, but the load operation will still return immediately to the caller.
712 Thus, callers should not assume that the callback will be called before the
713 load returns, and code should be structured appropriately to handle this.
714 See below for specific flags and error codes that control this behavior.
717 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
718 .It Dv BUS_DMA_NOWAIT
719 The load should not be deferred in case of insufficient mapping resources,
720 and instead should return immediately with an appropriate error.
721 .It Dv BUS_DMA_NOCACHE
722 The generated transactions to and from the virtual page are non-cacheable.
724 .Fn bus_dmamap_load ,
727 flag is currently implemented on sparc64.
731 Return values to the caller are as follows:
732 .Bl -tag -width ".Er EINPROGRESS"
734 The callback has been called and completed.
735 The status of the mapping has been delivered to the callback.
737 The mapping has been deferred for lack of resources.
738 The callback will be called as soon as resources are available.
739 Callbacks are serviced in FIFO order.
741 Note that subsequent load operations for the same tag that do not require
742 extra resources will still succeed.
743 This may result in out-of-order processing of requests.
744 If the caller requires the order of requests to be preserved,
745 then the caller is required to stall subsequent requests until a pending
746 request's callback is invoked.
748 The load request has failed due to insufficient resources, and the caller
749 specifically used the
753 The load request was invalid.
754 The callback has been called and has been provided the same error.
755 This error value may indicate that
765 argument used to create the dma tag
769 When the callback is called, it is presented with an error value
770 indicating the disposition of the mapping.
771 Error may be one of the following:
772 .Bl -tag -width ".Er EINPROGRESS"
774 The mapping was successful and the
776 callback argument contains an array of
777 .Vt bus_dma_segment_t
778 elements describing the mapping.
779 This array is only valid during the scope of the callback function.
781 A mapping could not be achieved within the segment constraints provided
782 in the tag even though the requested allocation size was less than maxsize.
784 .It Fn bus_dmamap_load_bio "dmat" "map" "bio" "callback" "callback_arg" "flags"
785 This is a variation of
787 which maps buffers pointed to by
791 may point to either a mapped or unmapped buffer.
792 .It Fn bus_dmamap_load_ccb "dmat" "map" "ccb" "callback" "callback_arg" "flags"
793 This is a variation of
795 which maps data pointed to by
800 may be any of the following types:
801 .Bl -tag -width ".Er CAM_DATA_SG_PADDR"
803 The data is a single KVA buffer.
805 The data is a single bus address range.
807 The data is a scatter/gather list of KVA buffers.
808 .It CAM_DATA_SG_PADDR
809 The data is a scatter/gather list of bus address ranges.
811 The data is contained in a
816 .Fn bus_dmamap_load_ccb
817 supports the following CCB XPT function codes:
819 .Bl -item -offset indent -compact
827 .It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
829 This is a variation of
831 which maps mbuf chains
835 argument is also passed to the callback routine, which
836 contains the mbuf chain's packet header length.
839 flag is implied, thus no callback deferral will happen.
841 Mbuf chains are assumed to be in kernel virtual address space.
843 Beside the error values listed for
844 .Fn bus_dmamap_load ,
846 will be returned if the size of the mbuf chain exceeds the maximum limit of the
848 .It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
850 .Fn bus_dmamap_load_mbuf
851 except that it returns immediately without calling a callback function.
852 It is provided for efficiency.
853 The scatter/gather segment array
855 is provided by the caller and filled in directly by the function.
858 argument is returned with the number of segments filled in.
859 Returns the same errors as
860 .Fn bus_dmamap_load_mbuf .
861 .It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
862 This is a variation of
864 which maps buffers pointed to by
869 argument is also passed to the callback routine, which contains the size of
875 flag is implied, thus no callback deferral will happen.
876 Returns the same errors as
877 .Fn bus_dmamap_load .
883 then it is assumed that the buffer,
886 .Fa "uio->uio_td->td_proc" Ns 's
888 User space memory must be in-core and wired prior to attempting a map
890 Pages may be locked using
892 .It Fn bus_dmamap_unload "dmat" "map"
894 Arguments are as follows:
895 .Bl -tag -width ".Fa dmam"
897 DMA tag used to allocate
900 The DMA map that is to be unloaded.
903 .Fn bus_dmamap_unload
904 will not perform any implicit synchronization of DMA buffers.
905 This must be done explicitly by a call to
907 prior to unloading the map.
908 .It Fn bus_dmamap_sync "dmat" "map" "op"
909 Performs synchronization of a device visible mapping with the CPU visible
910 memory referenced by that mapping.
911 Arguments are as follows:
912 .Bl -tag -width ".Fa dmat"
914 DMA tag used to allocate
917 The DMA mapping to be synchronized.
919 Type of synchronization operation to perform.
920 See the definition of
922 for a description of the acceptable values for
929 is the method used to ensure that CPU's and device's direct
930 memory access (DMA) to shared
932 For example, the CPU might be used to set up the contents of a buffer
933 that is to be made available to a device.
934 To ensure that the data are visible via the device's mapping of that
935 memory, the buffer must be loaded and a DMA sync operation of
936 .Dv BUS_DMASYNC_PREWRITE
937 must be performed after the CPU has updated the buffer and before the device
939 If the CPU modifies this buffer again later, another
940 .Dv BUS_DMASYNC_PREWRITE
941 sync operation must be performed before an additional device
943 Conversely, suppose a device updates memory that is to be read by a CPU.
944 In this case, the buffer must be loaded, and a DMA sync operation of
945 .Dv BUS_DMASYNC_PREREAD
946 must be performed before the device access is initiated.
947 The CPU will only be able to see the results of this memory update
948 once the DMA operation has completed and a
949 .Dv BUS_DMASYNC_POSTREAD
950 sync operation has been performed.
952 If read and write operations are not preceded and followed by the
953 appropriate synchronization operations, behavior is undefined.
954 .It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
955 Allocates memory that is mapped into KVA at the address returned
958 and that is permanently loaded into the newly created
962 Arguments are as follows:
963 .Bl -tag -width ".Fa alignment"
965 DMA tag describing the constraints of the DMA mapping.
967 Pointer to a pointer that will hold the returned KVA mapping of
968 the allocated region.
970 Flags are defined as follows:
971 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
972 .It Dv BUS_DMA_WAITOK
973 The routine can safely wait (sleep) for resources.
974 .It Dv BUS_DMA_NOWAIT
975 The routine is not allowed to wait for resources.
976 If resources are not available,
979 .It Dv BUS_DMA_COHERENT
980 Attempt to map this memory in a coherent fashion.
982 .Fn bus_dmamap_create
983 above for a description of this flag.
985 .Fn bus_dmamem_alloc ,
988 flag is currently implemented on arm, arm64 and sparc64.
990 Causes the allocated memory to be set to all zeros.
991 .It Dv BUS_DMA_NOCACHE
992 The allocated memory will not be cached in the processor caches.
993 All memory accesses appear on the bus and are executed
996 .Fn bus_dmamem_alloc ,
999 flag is currently implemented on amd64 and i386 where it results in the
1000 Strong Uncacheable PAT to be set for the allocated virtual address range.
1005 where the resulting DMA map will be stored.
1008 The size of memory to be allocated is
1010 as specified in the call to
1011 .Fn bus_dma_tag_create
1015 The current implementation of
1016 .Fn bus_dmamem_alloc
1017 will allocate all requests as a single segment.
1019 An initial load operation is required to obtain the bus address of the allocated
1020 memory, and an unload operation is required before freeing the memory, as
1022 .Fn bus_dmamem_free .
1023 Maps are automatically handled by this function and should not be explicitly
1024 allocated or destroyed.
1026 Although an explicit load is not required for each access to the memory
1027 referenced by the returned map, the synchronization requirements
1030 section still apply and should be used to achieve portability on architectures
1031 without coherent buses.
1035 if sufficient memory is not available for completing
1037 .It Fn bus_dmamem_free "dmat" "*vaddr" "map"
1038 Frees memory previously allocated by
1039 .Fn bus_dmamem_alloc .
1041 will be invalidated.
1042 Arguments are as follows:
1043 .Bl -tag -width ".Fa vaddr"
1047 Kernel virtual address of the memory.
1049 DMA map to be invalidated.
1053 Behavior is undefined if invalid arguments are passed to
1054 any of the above functions.
1055 If sufficient resources cannot be allocated for a given
1060 routines that are not of type
1062 will return 0 on success or an error
1063 code on failure as discussed above.
1067 routines will succeed if provided with valid arguments.
1069 Two locking protocols are used by
1071 The first is a private global lock that is used to synchronize access to the
1072 bounce buffer pool on the architectures that make use of them.
1073 This lock is strictly a leaf lock that is only used internally to
1075 and is not exposed to clients of the API.
1077 The second protocol involves protecting various resources stored in the tag.
1080 operations are done through requests from the driver that created the tag,
1081 the most efficient way to protect the tag resources is through the lock that
1085 acts on its own without being called by the driver, the lock primitive
1086 specified in the tag is acquired and released automatically.
1087 An example of this is when the
1089 callback function is called from a deferred context instead of the driver
1091 This means that certain
1093 functions must always be called with the same lock held that is specified in the
1095 These functions include:
1097 .Bl -item -offset indent -compact
1101 .Fn bus_dmamap_load_bio
1103 .Fn bus_dmamap_load_ccb
1105 .Fn bus_dmamap_load_mbuf
1107 .Fn bus_dmamap_load_mbuf_sg
1109 .Fn bus_dmamap_load_uio
1111 .Fn bus_dmamap_unload
1116 There is one exception to this rule.
1117 It is common practice to call some of these functions during driver start-up
1118 without any locks held.
1119 So long as there is a guarantee of no possible concurrent use of the tag by
1120 different threads during this operation, it is safe to not hold a lock for
1125 operations should not be called with the driver lock held, either because
1126 they are already protected by an internal lock, or because they might sleep
1127 due to memory or resource allocation.
1128 The following functions must not be
1129 called with any non-sleepable locks held:
1131 .Bl -item -offset indent -compact
1133 .Fn bus_dma_tag_create
1135 .Fn bus_dmamap_create
1137 .Fn bus_dmamem_alloc
1140 All other functions do not have a locking protocol and can thus be
1141 called with or without any system or driver locks held.
1150 .%A "Jason R. Thorpe"
1151 .%T "A Machine-Independent DMA Framework for NetBSD"
1152 .%J "Proceedings of the Summer 1998 USENIX Technical Conference"
1153 .%Q "USENIX Association"
1159 interface first appeared in
1164 API was adopted from
1166 for use in the CAM SCSI subsystem.
1167 The alterations to the original API were aimed to remove the need for
1169 .Vt bus_dma_segment_t
1170 array stored in each
1172 while allowing callers to queue up on scarce resources.
1176 interface was designed and implemented by
1178 of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
1179 Additional input on the
1181 design was provided by
1183 .An Chris Demetriou ,
1184 .An Charles Hannum ,
1187 .An Jonathan Stone ,
1195 benefits from the contributions of
1196 .An Justin T. Gibbs ,
1199 .An Matthew N. Dodd ,
1201 .An Maxime Henrion ,
1202 .An Jake Burkholder ,
1203 .An Takahashi Yoshihiro ,
1207 This manual page was written by
1210 .An Justin T. Gibbs .