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_dma_template_init ,
64 .Nm bus_dma_template_tag ,
65 .Nm bus_dma_template_clone ,
66 .Nm bus_dmamap_create ,
67 .Nm bus_dmamap_destroy ,
69 .Nm bus_dmamap_load_bio ,
70 .Nm bus_dmamap_load_ccb ,
71 .Nm bus_dmamap_load_crp ,
72 .Nm bus_dmamap_load_crp_buffer ,
73 .Nm bus_dmamap_load_mbuf ,
74 .Nm bus_dmamap_load_mbuf_sg ,
75 .Nm bus_dmamap_load_uio ,
76 .Nm bus_dmamap_unload ,
78 .Nm bus_dmamem_alloc ,
80 .Nd Bus and Machine Independent DMA Mapping Interface
84 .Fn bus_dma_tag_create "bus_dma_tag_t parent" "bus_size_t alignment" \
85 "bus_addr_t boundary" "bus_addr_t lowaddr" "bus_addr_t highaddr" \
86 "bus_dma_filter_t *filtfunc" "void *filtfuncarg" "bus_size_t maxsize" \
87 "int nsegments" "bus_size_t maxsegsz" "int flags" "bus_dma_lock_t *lockfunc" \
88 "void *lockfuncarg" "bus_dma_tag_t *dmat"
90 .Fn bus_dma_tag_destroy "bus_dma_tag_t dmat"
92 .Fo bus_dma_template_init
93 .Fa "bus_dma_template_t template"
94 .Fa "bus_dma_tag_t parent"
97 .Fo bus_dma_template_tag
98 .Fa "bus_dma_template_t template"
99 .Fa "bus_dma_tag_t *dmat"
102 .Fo bus_dma_template_clone
103 .Fa "bus_dma_template_t template"
104 .Fa "bus_dma_tag_t dmat"
107 .Fn bus_dmamap_create "bus_dma_tag_t dmat" "int flags" "bus_dmamap_t *mapp"
109 .Fn bus_dmamap_destroy "bus_dma_tag_t dmat" "bus_dmamap_t map"
111 .Fn bus_dmamap_load "bus_dma_tag_t dmat" "bus_dmamap_t map" "void *buf" \
112 "bus_size_t buflen" "bus_dmamap_callback_t *callback" "void *callback_arg" \
115 .Fn bus_dmamap_load_bio "bus_dma_tag_t dmat" "bus_dmamap_t map" \
116 "struct bio *bio" "bus_dmamap_callback_t *callback" "void *callback_arg" \
119 .Fn bus_dmamap_load_ccb "bus_dma_tag_t dmat" "bus_dmamap_t map" \
120 "union ccb *ccb" "bus_dmamap_callback_t *callback" "void *callback_arg" \
123 .Fn bus_dmamap_load_crp "bus_dma_tag_t dmat" "bus_dmamap_t map" \
124 "struct crypto *crp" "bus_dmamap_callback_t *callback" "void *callback_arg" \
127 .Fn bus_dmamap_load_crp_buffer "bus_dma_tag_t dmat" "bus_dmamap_t map" \
128 "struct crypto_buffer *cb" "bus_dmamap_callback_t *callback" \
129 "void *callback_arg" "int flags"
131 .Fn bus_dmamap_load_mbuf "bus_dma_tag_t dmat" "bus_dmamap_t map" \
132 "struct mbuf *mbuf" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
135 .Fn bus_dmamap_load_mbuf_sg "bus_dma_tag_t dmat" "bus_dmamap_t map" \
136 "struct mbuf *mbuf" "bus_dma_segment_t *segs" "int *nsegs" "int flags"
138 .Fn bus_dmamap_load_uio "bus_dma_tag_t dmat" "bus_dmamap_t map" \
139 "struct uio *uio" "bus_dmamap_callback2_t *callback" "void *callback_arg" \
142 .Fn bus_dmamap_unload "bus_dma_tag_t dmat" "bus_dmamap_t map"
144 .Fn bus_dmamap_sync "bus_dma_tag_t dmat" "bus_dmamap_t map" \
147 .Fn bus_dmamem_alloc "bus_dma_tag_t dmat" "void **vaddr" \
148 "int flags" "bus_dmamap_t *mapp"
150 .Fn bus_dmamem_free "bus_dma_tag_t dmat" "void *vaddr" \
153 Direct Memory Access (DMA) is a method of transferring data
154 without involving the CPU, thus providing higher performance.
155 A DMA transaction can be achieved between device to memory,
156 device to device, or memory to memory.
160 API is a bus, device, and machine-independent (MI) interface to
162 It provides the client with flexibility and simplicity by
163 abstracting machine dependent issues like setting up
164 DMA mappings, handling cache issues, bus specific features
168 .Vt ( bus_dma_tag_t )
169 is used to describe the properties of a group of related DMA
171 One way to view this is that a tag describes the limitations of a DMA engine.
172 For example, if a DMA engine in a device is limited to 32-bit addresses,
173 that limitation is specified by a parameter when creating the tag
175 Similarly, a tag can be marked as requiring buffers whose addresses are
176 aligned to a specific boundary.
178 Some devices may require multiple tags to describe DMA
179 transactions with differing properties.
180 For example, a device might require 16-byte alignment of its descriptor ring
181 while permitting arbitrary alignment of I/O buffers.
183 the driver must create one tag for the descriptor ring and a separate tag for
185 If a device has restrictions that are common to all DMA transactions
186 in addition to restrictions that differ between unrelated groups of
188 the driver can first create a
190 tag that decribes the common restrictions.
191 The per-group tags can then inherit these restrictions from this
193 tag rather than having to list them explicitly when creating the per-group tags.
197 represents a mapping of a memory region for DMA.
198 On systems with I/O MMUs,
199 the mapping structure tracks any I/O MMU entries used by a request.
200 For DMA requests that require bounce pages,
201 the mapping tracks the bounce pages used.
203 To prepare for one or more DMA transactions,
204 a mapping must be bound to a memory region by calling one of the
207 These functions configure the mapping which can include programming entries
208 in an I/O MMU and/or allocating bounce pages.
209 An output of these functions
210 (either directly or indirectly by invoking a callback routine)
211 is the list of scatter/gather address ranges a consumer can pass to a DMA
212 engine to access the memory region.
213 When a mapping is no longer needed,
214 the mapping must be unloaded via
215 .Fn bus_dmamap_unload .
217 Before and after each DMA transaction,
219 must be used to ensure that the correct data is used by the DMA engine and
221 If a mapping uses bounce pages,
222 the sync operations copy data between the bounce pages and the memory region
223 bound to the mapping.
224 Sync operations also handle architecture-specific details such as CPU cache
225 flushing and CPU memory operation ordering.
226 .Sh STATIC VS DYNAMIC
228 handles two types of DMA transactions: static and dynamic.
229 Static transactions are used with a long-lived memory region that is reused
230 for many transactions such as a descriptor ring.
231 Dynamic transactions are used for transfers to or from transient buffers
232 such as I/O buffers holding a network packet or disk block.
233 Each transaction type uses a different subset of the
236 .Ss Static Transactions
237 Static transactions use memory regions allocated by
239 Each static memory region is allocated by calling
240 .Fn bus_dmamem_alloc .
241 This function requires a valid tag describing the properties of the
242 DMA transactions to this region such as alignment or address restrictions.
243 Multiple regions can share a single tag if they share the same restrictions.
246 allocates a memory region along with a mapping object.
247 The associated tag, memory region, and mapping object must then be passed to
249 to bind the mapping to the allocated region and obtain the
254 will attempt to allocate memory requiring less expensive sync operations
255 (for example, implementations should not allocate regions requiring bounce
257 but sync operations should still be used.
258 For example, a driver should use
260 in an interrupt handler before reading descriptor ring entries written by the
261 device prior to the interrupt.
263 When a consumer is finished with a memory region,
264 it should unload the mapping via
265 .Fn bus_dmamap_unload
266 and then release the memory region and mapping object via
267 .Fn bus_dmamem_free .
268 .Ss Dynamic Transactions
269 Dynamic transactions map memory regions provided by other parts of the system.
270 A tag must be created via
271 .Fn bus_dma_tag_create
272 to describe the DMA transactions to and from these memory regions,
273 and a pool of mapping objects must be allocated via
274 .Fn bus_dmamap_create
275 to track the mappings of any in-flight transactions.
277 When a consumer wishes to schedule a transaction for a memory region,
278 the consumer must first obtain an unused mapping object from its pool
280 The memory region must be bound to the mapping object via one of the
283 Before scheduling the transaction,
284 the consumer should sync the memory region via
286 with one or more of the
289 After the transaction has completed,
290 the consumer should sync the memory region via
292 with one or more of the
295 The mapping can then be unloaded via
296 .Fn bus_dmamap_unload ,
297 and the mapping object can be returned to the pool of unused mapping objects.
299 When a consumer is no longer scheduling DMA transactions,
300 the mapping objects should be freed via
301 .Fn bus_dmamap_destroy ,
302 and the tag should be freed via
303 .Fn bus_dma_tag_destroy .
304 .Sh STRUCTURES AND TYPES
305 .Bl -tag -width indent
307 A machine-dependent (MD) opaque type that describes the
308 characteristics of a group of DMA transactions.
309 DMA tags are organized into a hierarchy, with each child
310 tag inheriting the restrictions of its parent.
311 This allows all devices along the path of DMA transactions
312 to contribute to the constraints of those transactions.
313 .It Vt bus_dma_template_t
314 A template structure for creating a
316 from a set of defaults.
317 Once initialized with
318 .Fn bus_dma_template_init ,
319 a driver can over-ride individual fields to suit its needs.
320 The following fields have the indicated values:
324 lowaddr BUS_SPACE_MAXADDR
325 highaddr BUS_SPACE_MAXADDR
326 maxsize BUS_SPACE_MAXSIZE
327 nsegments BUS_SPACE_UNRESTRICTED
328 maxsegsize BUS_SPACE_MAXSIZE
334 Descriptions of each field are documented with
335 .Fn bus_dma_tag_create .
340 attributes of the DMA tag are not supported with templates.
341 .It Vt bus_dma_filter_t
342 Client specified address filter having the format:
343 .Bl -tag -width indent
345 .Fn "client_filter" "void *filtarg" "bus_addr_t testaddr"
348 Address filters can be specified during tag creation to allow
349 for devices whose DMA address restrictions cannot be specified
353 argument is specified by the client during tag creation to be passed to all
354 invocations of the callback.
357 argument contains a potential starting address of a DMA mapping.
358 The filter function operates on the set of addresses from
361 .Ql trunc_page(testaddr) + PAGE_SIZE - 1 ,
363 The filter function should return zero if any mapping in this range
364 can be accommodated by the device and non-zero otherwise.
365 .It Vt bus_dma_segment_t
366 A machine-dependent type that describes individual
368 It contains the following fields:
376 field contains the device visible address of the DMA segment, and
378 contains the length of the DMA segment.
379 Although the DMA segments returned by a mapping call will adhere to
380 all restrictions necessary for a successful DMA operation, some conversion
381 (e.g.\& a conversion from host byte order to the device's byte order) is
382 almost always required when presenting segment information to the device.
384 A machine-dependent opaque type describing an individual mapping.
385 One map is used for each memory allocation that will be loaded.
386 Maps can be reused once they have been unloaded.
387 Multiple maps can be associated with one DMA tag.
388 While the value of the map may evaluate to
390 on some platforms under certain conditions,
391 it should never be assumed that it will be
394 .It Vt bus_dmamap_callback_t
395 Client specified callback for receiving mapping information resulting from
399 .Fn bus_dmamap_load ,
400 .Fn bus_dmamap_load_bio ,
401 .Fn bus_dmamap_load_ccb ,
402 .Fn bus_dmamap_load_crp ,
404 .Fn bus_dmamap_load_crp_buffer .
405 Callbacks are of the format:
406 .Bl -tag -width indent
408 .Fn "client_callback" "void *callback_arg" "bus_dma_segment_t *segs" \
409 "int nseg" "int error"
414 is the callback argument passed to dmamap load functions.
419 arguments describe an array of
420 .Vt bus_dma_segment_t
421 structures that represent the mapping.
422 This array is only valid within the scope of the callback function.
423 The success or failure of the mapping is indicated by the
426 More information on the use of callbacks can be found in the
427 description of the individual dmamap load functions.
428 .It Vt bus_dmamap_callback2_t
429 Client specified callback for receiving mapping information resulting from
433 .Fn bus_dmamap_load_uio
435 .Fn bus_dmamap_load_mbuf .
437 Callback2s are of the format:
438 .Bl -tag -width indent
440 .Fn "client_callback2" "void *callback_arg" "bus_dma_segment_t *segs" \
441 "int nseg" "bus_size_t mapsize" "int error"
444 Callback2's behavior is the same as
445 .Vt bus_dmamap_callback_t
446 with the addition that the length of the data mapped is provided via
448 .It Vt bus_dmasync_op_t
449 Memory synchronization operation specifier.
450 Bus DMA requires explicit synchronization of memory with its device
451 visible mapping in order to guarantee memory coherency.
454 allows the type of DMA operation that will be or has been performed
455 to be communicated to the system so that the correct coherency measures
457 The operations are represented as bitfield flags that can be combined together,
458 though it only makes sense to combine PRE flags or POST flags, not both.
461 description below for more details on how to use these operations.
463 All operations specified below are performed from the host memory point of view,
464 where a read implies data coming from the device to the host memory, and a write
465 implies data going from the host memory to the device.
466 Alternatively, the operations can be thought of in terms of driver operations,
467 where reading a network packet or storage sector corresponds to a read operation
470 .Bl -tag -width ".Dv BUS_DMASYNC_POSTWRITE"
471 .It Dv BUS_DMASYNC_PREREAD
472 Perform any synchronization required prior to an update of host memory by the
474 .It Dv BUS_DMASYNC_PREWRITE
475 Perform any synchronization required after an update of host memory by the CPU
476 and prior to device access to host memory.
477 .It Dv BUS_DMASYNC_POSTREAD
478 Perform any synchronization required after an update of host memory by the
479 device and prior to CPU access to host memory.
480 .It Dv BUS_DMASYNC_POSTWRITE
481 Perform any synchronization required after device access to host memory.
483 .It Vt bus_dma_lock_t
484 Client specified lock/mutex manipulation method.
485 This will be called from
486 within busdma whenever a client lock needs to be manipulated.
487 In its current form, the function will be called immediately before
488 the callback for a DMA load operation that has been deferred with
490 and immediately after with
492 If the load operation does not need to be deferred, then it
493 will not be called since the function loading the map should
494 be holding the appropriate locks.
495 This method is of the format:
496 .Bl -tag -width indent
498 .Fn "lockfunc" "void *lockfunc_arg" "bus_dma_lock_op_t op"
503 argument is specified by the client during tag creation to be passed to all
504 invocations of the callback.
507 argument specifies the lock operation to perform.
511 implementations are provided for convenience.
512 .Fn busdma_lock_mutex
513 performs standard mutex operations on the sleep mutex provided via
516 will generate a system panic if it is called.
517 It is substituted into the tag when
522 .Fn bus_dma_tag_create
523 and is useful for tags that should not be used with deferred load operations.
524 .It Vt bus_dma_lock_op_t
525 Operations to be performed by the client-specified
527 .Bl -tag -width ".Dv BUS_DMA_UNLOCK"
529 Acquires and/or locks the client locking primitive.
530 .It Dv BUS_DMA_UNLOCK
531 Releases and/or unlocks the client locking primitive.
535 .Bl -tag -width indent
536 .It Fn bus_dma_tag_create "parent" "alignment" "boundary" "lowaddr" \
537 "highaddr" "*filtfunc" "*filtfuncarg" "maxsize" "nsegments" "maxsegsz" \
538 "flags" "lockfunc" "lockfuncarg" "*dmat"
539 Allocates a DMA tag, and initializes it according to
540 the arguments provided:
541 .Bl -tag -width ".Fa filtfuncarg"
543 A parent tag from which to inherit restrictions.
544 The restrictions passed in other arguments can only further tighten the
545 restrictions inherited from the parent tag.
547 All tags created by a device driver must inherit from the tag returned by
549 to honor restrictions between the parent bridge, CPU memory, and the
552 Alignment constraint, in bytes, of any mappings created using this tag.
553 The alignment must be a power of 2.
554 Hardware that can DMA starting at any address would specify
557 Hardware requiring DMA transfers to start on a multiple of 4K
561 Boundary constraint, in bytes, of the target DMA memory region.
562 The boundary indicates the set of addresses, all multiples of the
563 boundary argument, that cannot be crossed by a single
564 .Vt bus_dma_segment_t .
565 The boundary must be a power of 2 and must be no smaller than the
566 maximum segment size.
568 indicates that there are no boundary restrictions.
569 .It Fa lowaddr , highaddr
570 Bounds of the window of bus address space that
572 be directly accessed by the device.
573 The window contains all addresses greater than
575 and less than or equal to
577 For example, a device incapable of DMA above 4GB, would specify a
580 .Dv BUS_SPACE_MAXADDR
584 .Dv BUS_SPACE_MAXADDR_32BIT .
585 Similarly a device that can only perform DMA to addresses below
589 .Dv BUS_SPACE_MAXADDR
593 .Dv BUS_SPACE_MAXADDR_24BIT .
594 Some implementations require that some region of device visible
595 address space, overlapping available host memory, be outside the
599 is used to bounce requests that would otherwise conflict with
600 the exclusion window.
602 Optional filter function (may be
604 to be called for any attempt to
605 map memory into the window described by
609 A filter function is only required when the single window described
614 cannot adequately describe the constraints of the device.
615 The filter function will be called for every machine page
616 that overlaps the exclusion window.
618 Argument passed to all calls to the filter function for this tag.
622 Maximum size, in bytes, of the sum of all segment lengths in a given
623 DMA mapping associated with this tag.
625 Number of discontinuities (scatter/gather segments) allowed
626 in a DMA mapped region.
628 Maximum size, in bytes, of a segment in any DMA mapped region associated
633 .Bl -tag -width ".Dv BUS_DMA_ALLOCNOW"
634 .It Dv BUS_DMA_ALLOCNOW
635 Pre-allocate enough resources to handle at least one map load operation on
637 If sufficient resources are not available,
640 This should not be used for tags that only describe buffers that will be
642 .Fn bus_dmamem_alloc .
643 Also, due to resource sharing with other tags, this flag does not guarantee
644 that resources will be allocated or reserved exclusively for this tag.
645 It should be treated only as a minor optimization.
646 .It Dv BUS_DMA_COHERENT
647 Indicate that the DMA engine and CPU are cache-coherent.
648 Cached memory may be used to back allocations created by
649 .Fn bus_dmamem_alloc .
651 .Fn bus_dma_tag_create ,
654 flag is currently implemented on arm64.
657 Optional lock manipulation function (may be
659 to be called when busdma
660 needs to manipulate a lock on behalf of the client.
667 Optional argument to be passed to the function specified by
670 Pointer to a bus_dma_tag_t where the resulting DMA tag will
676 if sufficient memory is not available for tag creation
677 or allocating mapping resources.
678 .It Fn bus_dma_tag_destroy "dmat"
679 Deallocate the DMA tag
682 .Fn bus_dma_tag_create .
686 if any DMA maps remain associated with
691 .It Fn bus_dma_template_init "*template" "parent"
693 .Fa bus_dma_template_t
694 structure and associates it with an optional
698 argument may be NULL.
699 .It Fn bus_dma_template_tag "*template" "*dmat"
700 Unpacks a template into a tag, and returns the tag via the
702 All return values are identical to
703 .Fn bus_dma_tag_create .
704 .It Fn bus_dma_template_clone "*template" "dmat"
705 Clones the fields from a tag to a template.
706 This is useful for cloning tags when paired with
707 .Fn bus_dma_template_tag .
708 A template that is filled in as a clone does not need to be initialized
710 .It Fn bus_dmamap_create "dmat" "flags" "*mapp"
711 Allocates and initializes a DMA map.
712 Arguments are as follows:
713 .Bl -tag -width ".Fa nsegments"
718 .Bl -tag -width ".Dv BUS_DMA_COHERENT"
719 .It Dv BUS_DMA_COHERENT
720 Attempt to map the memory loaded with this map such that cache sync
721 operations are as cheap as possible.
722 This flag is typically set on maps when the memory loaded with these will
723 be accessed by both a CPU and a DMA engine, frequently such as control data
724 and as opposed to streamable data such as receive and transmit buffers.
725 Use of this flag does not remove the requirement of using
726 .Fn bus_dmamap_sync ,
727 but it may reduce the cost of performing these operations.
732 where the resulting DMA map will be stored.
737 if sufficient memory is not available for creating the
738 map or allocating mapping resources.
739 .It Fn bus_dmamap_destroy "dmat" "map"
740 Frees all resources associated with a given DMA map.
741 Arguments are as follows:
742 .Bl -tag -width ".Fa dmat"
744 DMA tag used to allocate
747 The DMA map to destroy.
752 if a mapping is still active for
754 .It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" \
755 "callback_arg" "flags"
756 Creates a mapping in device visible address space of
760 associated with the DMA map
762 This call will always return immediately and will not block for any reason.
763 Arguments are as follows:
764 .Bl -tag -width ".Fa buflen"
766 DMA tag used to allocate
769 A DMA map without a currently active mapping.
771 A kernel virtual address pointer to a contiguous (in KVA) buffer, to be
772 mapped into device visible address space.
774 The size of the buffer.
775 .It Fa callback Fa callback_arg
776 The callback function, and its argument.
777 This function is called once sufficient mapping resources are available for
779 If resources are temporarily unavailable, this function will be deferred until
780 later, but the load operation will still return immediately to the caller.
781 Thus, callers should not assume that the callback will be called before the
782 load returns, and code should be structured appropriately to handle this.
783 See below for specific flags and error codes that control this behavior.
786 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
787 .It Dv BUS_DMA_NOWAIT
788 The load should not be deferred in case of insufficient mapping resources,
789 and instead should return immediately with an appropriate error.
790 .It Dv BUS_DMA_NOCACHE
791 The generated transactions to and from the virtual page are non-cacheable.
795 Return values to the caller are as follows:
796 .Bl -tag -width ".Er EINPROGRESS"
798 The callback has been called and completed.
799 The status of the mapping has been delivered to the callback.
801 The mapping has been deferred for lack of resources.
802 The callback will be called as soon as resources are available.
803 Callbacks are serviced in FIFO order.
805 Note that subsequent load operations for the same tag that do not require
806 extra resources will still succeed.
807 This may result in out-of-order processing of requests.
808 If the caller requires the order of requests to be preserved,
809 then the caller is required to stall subsequent requests until a pending
810 request's callback is invoked.
812 The load request has failed due to insufficient resources, and the caller
813 specifically used the
817 The load request was invalid.
818 The callback has been called and has been provided the same error.
819 This error value may indicate that
829 argument used to create the dma tag
833 When the callback is called, it is presented with an error value
834 indicating the disposition of the mapping.
835 Error may be one of the following:
836 .Bl -tag -width ".Er EINPROGRESS"
838 The mapping was successful and the
840 callback argument contains an array of
841 .Vt bus_dma_segment_t
842 elements describing the mapping.
843 This array is only valid during the scope of the callback function.
845 A mapping could not be achieved within the segment constraints provided
846 in the tag even though the requested allocation size was less than maxsize.
848 .It Fn bus_dmamap_load_bio "dmat" "map" "bio" "callback" "callback_arg" "flags"
849 This is a variation of
851 which maps buffers pointed to by
855 may point to either a mapped or unmapped buffer.
856 .It Fn bus_dmamap_load_ccb "dmat" "map" "ccb" "callback" "callback_arg" "flags"
857 This is a variation of
859 which maps data pointed to by
864 may be any of the following types:
865 .Bl -tag -width ".Er CAM_DATA_SG_PADDR"
867 The data is a single KVA buffer.
869 The data is a single bus address range.
871 The data is a scatter/gather list of KVA buffers.
872 .It CAM_DATA_SG_PADDR
873 The data is a scatter/gather list of bus address ranges.
875 The data is contained in a
880 .Fn bus_dmamap_load_ccb
881 supports the following CCB XPT function codes:
883 .Bl -item -offset indent -compact
891 .It Fn bus_dmamap_load_crp "dmat" "map" "crp" "callback" "callback_arg" "flags"
892 This is a variation of
894 which maps the input buffer pointed to by
899 flag is implied, thus no callback deferral will happen.
900 .It Fn bus_dmamap_load_crp_buffer "dmat" "map" "cb" "callback" "callback_arg" \
902 This is a variation of
904 which maps the crypto data buffer pointed to by
909 flag is implied, thus no callback deferral will happen.
910 .It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \
912 This is a variation of
914 which maps mbuf chains
918 argument is also passed to the callback routine, which
919 contains the mbuf chain's packet header length.
922 flag is implied, thus no callback deferral will happen.
924 Mbuf chains are assumed to be in kernel virtual address space.
926 Beside the error values listed for
927 .Fn bus_dmamap_load ,
929 will be returned if the size of the mbuf chain exceeds the maximum limit of the
931 .It Fn bus_dmamap_load_mbuf_sg "dmat" "map" "mbuf" "segs" "nsegs" "flags"
933 .Fn bus_dmamap_load_mbuf
934 except that it returns immediately without calling a callback function.
935 It is provided for efficiency.
936 The scatter/gather segment array
938 is provided by the caller and filled in directly by the function.
941 argument is returned with the number of segments filled in.
942 Returns the same errors as
943 .Fn bus_dmamap_load_mbuf .
944 .It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags"
945 This is a variation of
947 which maps buffers pointed to by
952 argument is also passed to the callback routine, which contains the size of
958 flag is implied, thus no callback deferral will happen.
959 Returns the same errors as
960 .Fn bus_dmamap_load .
966 then it is assumed that the buffer,
969 .Fa "uio->uio_td->td_proc" Ns 's
971 User space memory must be in-core and wired prior to attempting a map
973 Pages may be locked using
975 .It Fn bus_dmamap_unload "dmat" "map"
977 Arguments are as follows:
978 .Bl -tag -width ".Fa dmam"
980 DMA tag used to allocate
983 The DMA map that is to be unloaded.
986 .Fn bus_dmamap_unload
987 will not perform any implicit synchronization of DMA buffers.
988 This must be done explicitly by a call to
990 prior to unloading the map.
991 .It Fn bus_dmamap_sync "dmat" "map" "op"
992 Performs synchronization of a device visible mapping with the CPU visible
993 memory referenced by that mapping.
994 Arguments are as follows:
995 .Bl -tag -width ".Fa dmat"
997 DMA tag used to allocate
1000 The DMA mapping to be synchronized.
1002 Type of synchronization operation to perform.
1003 See the definition of
1004 .Vt bus_dmasync_op_t
1005 for a description of the acceptable values for
1012 is the method used to ensure that CPU's and device's direct
1013 memory access (DMA) to shared
1015 For example, the CPU might be used to set up the contents of a buffer
1016 that is to be made available to a device.
1017 To ensure that the data are visible via the device's mapping of that
1018 memory, the buffer must be loaded and a DMA sync operation of
1019 .Dv BUS_DMASYNC_PREWRITE
1020 must be performed after the CPU has updated the buffer and before the device
1021 access is initiated.
1022 If the CPU modifies this buffer again later, another
1023 .Dv BUS_DMASYNC_PREWRITE
1024 sync operation must be performed before an additional device
1026 Conversely, suppose a device updates memory that is to be read by a CPU.
1027 In this case, the buffer must be loaded, and a DMA sync operation of
1028 .Dv BUS_DMASYNC_PREREAD
1029 must be performed before the device access is initiated.
1030 The CPU will only be able to see the results of this memory update
1031 once the DMA operation has completed and a
1032 .Dv BUS_DMASYNC_POSTREAD
1033 sync operation has been performed.
1035 If read and write operations are not preceded and followed by the
1036 appropriate synchronization operations, behavior is undefined.
1037 .It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "*mapp"
1038 Allocates memory that is mapped into KVA at the address returned
1041 and that is permanently loaded into the newly created
1045 Arguments are as follows:
1046 .Bl -tag -width ".Fa alignment"
1048 DMA tag describing the constraints of the DMA mapping.
1050 Pointer to a pointer that will hold the returned KVA mapping of
1051 the allocated region.
1053 Flags are defined as follows:
1054 .Bl -tag -width ".Dv BUS_DMA_NOWAIT"
1055 .It Dv BUS_DMA_WAITOK
1056 The routine can safely wait (sleep) for resources.
1057 .It Dv BUS_DMA_NOWAIT
1058 The routine is not allowed to wait for resources.
1059 If resources are not available,
1062 .It Dv BUS_DMA_COHERENT
1063 Attempt to map this memory in a coherent fashion.
1065 .Fn bus_dmamap_create
1066 above for a description of this flag.
1068 .Fn bus_dmamem_alloc ,
1070 .Dv BUS_DMA_COHERENT
1071 flag is currently implemented on arm and arm64.
1073 Causes the allocated memory to be set to all zeros.
1074 .It Dv BUS_DMA_NOCACHE
1075 The allocated memory will not be cached in the processor caches.
1076 All memory accesses appear on the bus and are executed
1079 .Fn bus_dmamem_alloc ,
1082 flag is currently implemented on amd64 and i386 where it results in the
1083 Strong Uncacheable PAT to be set for the allocated virtual address range.
1088 where the resulting DMA map will be stored.
1091 The size of memory to be allocated is
1093 as specified in the call to
1094 .Fn bus_dma_tag_create
1098 The current implementation of
1099 .Fn bus_dmamem_alloc
1100 will allocate all requests as a single segment.
1102 An initial load operation is required to obtain the bus address of the allocated
1103 memory, and an unload operation is required before freeing the memory, as
1105 .Fn bus_dmamem_free .
1106 Maps are automatically handled by this function and should not be explicitly
1107 allocated or destroyed.
1109 Although an explicit load is not required for each access to the memory
1110 referenced by the returned map, the synchronization requirements
1113 section still apply and should be used to achieve portability on architectures
1114 without coherent buses.
1118 if sufficient memory is not available for completing
1120 .It Fn bus_dmamem_free "dmat" "*vaddr" "map"
1121 Frees memory previously allocated by
1122 .Fn bus_dmamem_alloc .
1124 will be invalidated.
1125 Arguments are as follows:
1126 .Bl -tag -width ".Fa vaddr"
1130 Kernel virtual address of the memory.
1132 DMA map to be invalidated.
1136 Behavior is undefined if invalid arguments are passed to
1137 any of the above functions.
1138 If sufficient resources cannot be allocated for a given
1143 routines that are not of type
1145 will return 0 on success or an error
1146 code on failure as discussed above.
1150 routines will succeed if provided with valid arguments.
1152 Two locking protocols are used by
1154 The first is a private global lock that is used to synchronize access to the
1155 bounce buffer pool on the architectures that make use of them.
1156 This lock is strictly a leaf lock that is only used internally to
1158 and is not exposed to clients of the API.
1160 The second protocol involves protecting various resources stored in the tag.
1163 operations are done through requests from the driver that created the tag,
1164 the most efficient way to protect the tag resources is through the lock that
1168 acts on its own without being called by the driver, the lock primitive
1169 specified in the tag is acquired and released automatically.
1170 An example of this is when the
1172 callback function is called from a deferred context instead of the driver
1174 This means that certain
1176 functions must always be called with the same lock held that is specified in the
1178 These functions include:
1180 .Bl -item -offset indent -compact
1184 .Fn bus_dmamap_load_bio
1186 .Fn bus_dmamap_load_ccb
1188 .Fn bus_dmamap_load_mbuf
1190 .Fn bus_dmamap_load_mbuf_sg
1192 .Fn bus_dmamap_load_uio
1194 .Fn bus_dmamap_unload
1199 There is one exception to this rule.
1200 It is common practice to call some of these functions during driver start-up
1201 without any locks held.
1202 So long as there is a guarantee of no possible concurrent use of the tag by
1203 different threads during this operation, it is safe to not hold a lock for
1208 operations should not be called with the driver lock held, either because
1209 they are already protected by an internal lock, or because they might sleep
1210 due to memory or resource allocation.
1211 The following functions must not be
1212 called with any non-sleepable locks held:
1214 .Bl -item -offset indent -compact
1216 .Fn bus_dma_tag_create
1218 .Fn bus_dmamap_create
1220 .Fn bus_dmamem_alloc
1223 All other functions do not have a locking protocol and can thus be
1224 called with or without any system or driver locks held.
1233 .%A "Jason R. Thorpe"
1234 .%T "A Machine-Independent DMA Framework for NetBSD"
1235 .%J "Proceedings of the Summer 1998 USENIX Technical Conference"
1236 .%Q "USENIX Association"
1242 interface first appeared in
1247 API was adopted from
1249 for use in the CAM SCSI subsystem.
1250 The alterations to the original API were aimed to remove the need for
1252 .Vt bus_dma_segment_t
1253 array stored in each
1255 while allowing callers to queue up on scarce resources.
1259 interface was designed and implemented by
1261 of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
1262 Additional input on the
1264 design was provided by
1266 .An Chris Demetriou ,
1267 .An Charles Hannum ,
1270 .An Jonathan Stone ,
1278 benefits from the contributions of
1279 .An Justin T. Gibbs ,
1282 .An Matthew N. Dodd ,
1284 .An Maxime Henrion ,
1285 .An Jake Burkholder ,
1286 .An Takahashi Yoshihiro ,
1290 This manual page was written by
1293 .An Justin T. Gibbs .