1.\" 2.\" Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved. 3.\" 4.\" This code is derived from software contributed to The DragonFly Project 5.\" by Hiten Pandya <hmp@backplane.com>. 6.\" 7.\" Redistribution and use in source and binary forms, with or without 8.\" modification, are permitted provided that the following conditions 9.\" are met: 10.\" 11.\" 1. Redistributions of source code must retain the above copyright 12.\" notice, this list of conditions and the following disclaimer. 13.\" 2. Redistributions in binary form must reproduce the above copyright 14.\" notice, this list of conditions and the following disclaimer in 15.\" the documentation and/or other materials provided with the 16.\" distribution. 17.\" 3. Neither the name of The DragonFly Project nor the names of its 18.\" contributors may be used to endorse or promote products derived 19.\" from this software without specific, prior written permission. 20.\" 21.\" THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22.\" ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23.\" LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24.\" FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25.\" COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26.\" INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27.\" BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28.\" LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29.\" AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30.\" OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31.\" OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32.\" SUCH DAMAGE. 33.\" 34.\" Copyright (c) 1996, 1997, 1998, 2001 The NetBSD Foundation, Inc. 35.\" All rights reserved. 36.\" 37.\" This code is derived from software contributed to The NetBSD Foundation 38.\" by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 39.\" NASA Ames Research Center. 40.\" 41.\" Redistribution and use in source and binary forms, with or without 42.\" modification, are permitted provided that the following conditions 43.\" are met: 44.\" 1. Redistributions of source code must retain the above copyright 45.\" notice, this list of conditions and the following disclaimer. 46.\" 2. Redistributions in binary form must reproduce the above copyright 47.\" notice, this list of conditions and the following disclaimer in the 48.\" documentation and/or other materials provided with the distribution. 49.\" 3. All advertising materials mentioning features or use of this software 50.\" must display the following acknowledgment: 51.\" This product includes software developed by the NetBSD 52.\" Foundation, Inc. and its contributors. 53.\" 4. Neither the name of The NetBSD Foundation nor the names of its 54.\" contributors may be used to endorse or promote products derived 55.\" from this software without specific prior written permission. 56.\" 57.\" THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 58.\" ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 59.\" TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 60.\" PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 61.\" BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 62.\" CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 63.\" SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 64.\" INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 65.\" CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 66.\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 67.\" POSSIBILITY OF SUCH DAMAGE. 68.\" 69.\" $FreeBSD: /repoman/r/ncvs/src/share/man/man9/bus_dma.9,v 1.7 2003/07/27 14:05:29 mux Exp $ 70.\" $NetBSD: bus_dma.9,v 1.25 2002/10/14 13:43:16 wiz Exp $ 71.\" $DragonFly: src/share/man/man9/bus_dma.9,v 1.8 2007/04/09 21:20:37 swildner Exp $ 72.\" 73.Dd March 17, 2004 74.Dt BUS_DMA 9 75.Os 76.Sh NAME 77.Nm bus_dma , 78.Nm bus_dma_tag_create , 79.Nm bus_dma_tag_destroy , 80.Nm bus_dmamap_create , 81.Nm bus_dmamap_destroy , 82.Nm bus_dmamap_load , 83.Nm bus_dmamap_load_mbuf , 84.Nm bus_dmamap_load_mbuf_segment , 85.Nm bus_dmamap_load_mbuf_defrag , 86.Nm bus_dmamap_load_uio , 87.Nm bus_dmamap_unload , 88.Nm bus_dmamap_sync , 89.Nm bus_dmamem_alloc , 90.Nm bus_dmamem_coherent , 91.Nm bus_dmamem_coherent_any , 92.Nm bus_dmamem_free 93.Nd Bus and Machine Independent DMA Mapping Interface 94.Sh SYNOPSIS 95.In machine/bus.h 96.Ft int 97.Fn bus_dma_tag_create "bus_dma_tag_t parent" "bus_size_t alignment" \ 98"bus_size_t boundary" "bus_addr_t lowaddr" "bus_addr_t highaddr" \ 99"bus_dma_filter_t *filtfunc" "void *filtfuncarg" "bus_size_t maxsize" \ 100"int nsegments" "bus_size_t maxsegsz" "int flags" "bus_dma_tag_t *dmat" 101.Ft int 102.Fn bus_dma_tag_destroy "bus_dma_tag_t dmat" 103.Ft int 104.Fn bus_dmamap_create "bus_dma_tag_t dmat" "int flags" "bus_dmamap_t *mapp" 105.Ft int 106.Fn bus_dmamap_destroy "bus_dma_tag_t dmat" "bus_dmamap_t map" 107.Ft int 108.Fn bus_dmamap_load "bus_dma_tag_t dmat" "bus_dmamap_t map" "void *buf" \ 109"bus_size_t buflen" "bus_dmamap_callback_t *callback" "void *callback_arg" \ 110"int flags" 111.Ft int 112.Fn bus_dmamap_load_mbuf "bus_dma_tag_t dmat" "bus_dmamap_t map" \ 113"struct mbuf *mbuf" "bus_dmamap_callback2_t *callback" "void *callback_arg" \ 114"int flags" 115.Ft int 116.Fn bus_dmamap_load_mbuf_segment "bus_dma_tag_t dmat" "bus_dmamap_t map" \ 117"struct mbuf *mbuf" "bus_dma_segment_t *segs" "int maxsegs" "int *nsegs" \ 118"int flags" 119.Ft int 120.Fn bus_dmamap_load_mbuf_defrag "bus_dma_tag_t dmat" "bus_dmamap_t map" \ 121"struct mbuf **mbuf" "bus_dma_segment_t *segs" "int maxsegs" "int *nsegs" \ 122"int flags" 123.Ft int 124.Fn bus_dmamap_load_uio "bus_dma_tag_t dmat" "bus_dmamap_t map" \ 125"struct uio *uio" "bus_dmamap_callback2_t *callback" "void *callback_arg" \ 126"int flags" 127.Ft int 128.Fn bus_dmamem_alloc "bus_dma_tag_t dmat" "void **vaddr" \ 129"int flags" "bus_dmamap_t *mapp" 130.Ft int 131.Fn bus_dmamem_coherent "bus_dma_tag_t parent" "bus_size_t alignment" \ 132"bus_size_t boundary" "bus_addr_t lowaddr" "bus_addr_t highaddr" \ 133"bus_size_t maxsize" "int flags" "bus_dmamem_t *dmem" 134.Ft void * 135.Fn bus_dmamem_coherent_any "bus_dma_tag_t parent" "bus_size_t alignment" \ 136"bus_size_t maxsize" "int flags" "bus_dma_tag_t *dtag" "bus_dmamap_t *dmap" \ 137"bus_addr_t *busaddr" 138.Ft void 139.Fn bus_dmamap_unload "bus_dma_tag_t dmat" "bus_dmamap_t map" 140.Ft void 141.Fn bus_dmamap_sync "bus_dma_tag_t dmat" "bus_dmamap_t map" \ 142"bus_dmasync_op_t op" 143.Ft void 144.Fn bus_dmamem_free "bus_dma_tag_t dmat" "void *vaddr" \ 145"bus_dmamap_t map" 146.Sh DESCRIPTION 147Direct Memory Access (DMA) is a method of transferring data 148without involving the CPU, thus providing higher performance. 149A DMA transaction can be achieved between device to memory, 150device to device, or memory to memory. 151.Pp 152The 153.Nm 154API is a bus, device, and machine-independent (MI) interface to 155DMA mechanisms. 156It provides the client with flexibility and simplicity by 157abstracting machine dependent issues like setting up 158DMA mappings, handling cache issues, bus specific features 159and limitations. 160.Sh STRUCTURES AND TYPES 161.Bl -tag -width compact 162.It Vt bus_dma_tag_t 163A machine-dependent (MD) opaque type that describes the 164characteristics of DMA transactions. 165DMA tags are organized into a hierarchy, with each child 166tag inheriting the restrictions of its parent. 167This allows all devices along the path of DMA transactions 168to contribute to the constraints of those transactions. 169.It Vt bus_dma_filter_t 170Client specified address filter having the format: 171.Bl -tag -width compact 172.It Ft int 173.Fn "client_filter" "void *filtarg" "bus_addr_t testaddr" 174.El 175.sp 176Address filters can be specified during tag creation to allow 177for devices who's DMA address restrictions cannot be specified 178by a single window. 179The 180.Fa filtarg 181is client specified during tag creation to be passed to all 182invocations of the callback. 183The 184.Fa testaddr 185argument contains a potential starting address of a DMA mapping. 186The filter function operates on the set of addresses from 187.Fa testaddr 188to 189.Ql trunc_page(testaddr) + PAGE_SIZE - 1 , 190inclusive. 191The filter function should return zero for any mapping in this range 192that can be accommodated by the device and non-zero otherwise. 193.It Vt bus_dma_segment_t 194A machine-dependent type that describes individual 195DMA segments. 196.Bd -literal 197 bus_addr_t ds_addr; 198 bus_size_t ds_len; 199.Ed 200.sp 201The 202.Fa ds_addr 203field contains the device visible address of the DMA segment, and 204.Fa ds_len 205contains the length of the DMA segment. 206Although the DMA segments returned by a mapping call will adhere to 207all restrictions necessary for a successful DMA operation, some conversion 208(e.g. a conversion from host byte order to the device's byte order) is 209almost always required when presenting segment information to the device. 210.It Vt bus_dmamap_t 211A machine-dependent opaque type describing an individual mapping. 212Multiple DMA maps can be associated with one DMA tag. 213.It Vt bus_dmamem_t 214A machine-dependent type that describes DMA memory created by 215.Fn bus_dmamem_coherent . 216.Bd -literal 217 bus_dma_tag_t dmem_tag; 218 bus_dmamap_t dmem_map; 219 void *dmem_addr; 220 bus_addr_t dmem_busaddr; 221.Ed 222.sp 223The 224.Fa dmem_tag 225field contains the DMA tag of the DMA memory and 226.Fa dmem_map 227field contains the DMA map of the DMA memory. 228The 229.Fa dmem_addr 230field points to the allocated DMA memory in kernel virtual address space. 231The 232.Fa dmem_busaddr 233field contains the device visible address of the DMA memory. 234.It Vt bus_dmamap_callback_t 235Client specified callback for receiving mapping information resulting from 236the load of a 237.Vt bus_dmamap_t 238via 239.Fn bus_dmamap_load . 240Callbacks are of the format: 241.Bl -tag -width compact 242.It Ft void 243.Fn "client_callback" "void *callback_arg" "bus_dma_segment_t *segs" \ 244"int nseg" "int error" 245.El 246.sp 247The 248.Fa callback_arg 249is the callback argument passed to dmamap load functions. 250The 251.Fa segs 252and 253.Fa nseg 254parameters describe an array of 255.Vt bus_dma_segment_t 256structures that represent the mapping. 257This array is only valid within the scope of the callback function. 258The success or failure of the mapping is indicated by the 259.Fa error 260parameter. 261More information on the use of callbacks can be found in the 262description of the individual dmamap load functions. 263.It Vt bus_dmamap_callback2_t 264Client specified callback for receiving mapping information resulting from 265the load of a 266.Vt bus_dmamap_t 267via 268.Fn bus_dmamap_load_uio 269or 270.Fn bus_dmamap_load_mbuf . 271.sp 272Callback2s are of the format: 273.Bl -tag -width compact 274.It Ft void 275.Fn "client_callback2" "void *callback_arg" "bus_dma_segment_t *segs" \ 276"int nseg" "bus_size_t mapsize" "int error" 277.El 278.sp 279Callback2's behavior is the same as 280.Vt bus_dmamap_callback_t 281with the addition that the length of the data mapped is provided via 282.Fa mapsize . 283.It Vt bus_dmasync_op_t 284Memory synchronization operation specifier. 285Bus DMA requires explicit synchronization of memory with it's device 286visible mapping in order to guarantee memory coherency. 287The 288.Vt bus_dmasync_op_t 289allows the type of DMA operation that will be or has been performed 290to be communicated to the system so that the correct coherency measures 291are taken. 292All operations specified below are performed from the DMA engine's 293point of view: 294.Bl -tag -width BUS_DMASYNC_POSTWRITE 295.It Dv BUS_DMASYNC_PREREAD 296Perform any synchronization required after an update of memory by the CPU 297but prior to DMA read operations. 298.It Dv BUS_DMASYNC_PREWRITE 299Perform any synchronization required after an update of memory by the CPU 300but prior to DMA write operations. 301.It Dv BUS_DMASYNC_POSTREAD 302Perform any synchronization required after DMA read operations, but prior 303to CPU access of the memory. 304.It Dv BUS_DMASYNC_POSTWRITE 305Perform any synchronization required after DMA write operations, but prior 306to CPU access of the memory. 307.El 308.El 309.sp 310.Sh FUNCTIONS 311.Bl -tag -width compact 312.It Fn bus_dma_tag_create "parent" "alignment" "boundary" "lowaddr" \ 313"highaddr" "*filtfunc" "*filtfuncarg" "maxsize" "nsegments" "maxsegsz" \ 314"flags" "*dmat" 315Allocates a device specific DMA tag, and initializes it according to 316the arguments provided: 317.Bl -tag -width *filtfuncarg -compact 318.It Fa parent 319Indicates restrictions between the parent bridge, CPU memory, and the 320device. 321May be NULL, if no DMA restrictions are to be inherited. 322.It Fa alignment 323Alignment constraint, in bytes, of any mappings created using this tag. 324The alignment must be a power of 2. 325Hardware that can DMA starting at any address would specify 326.Em 1 327for byte alignment. 328Hardware requiring DMA transfers to start on a multiple of 4K 329would specify 330.Em 4096 . 331.It Fa boundary 332Boundary constraint, in bytes, of the target DMA memory region. 333The boundary indicates the set of addresses, all multiples of the 334boundary argument, that cannot be crossed by a single 335.Vt bus_dma_segment_t . 336The boundary must be either a power of 2 or 0. 337.Ql 0 338indicates that there are no boundary restrictions. 339.It Fa lowaddr 340.It Fa highaddr 341Bounds of the window of bus address space that 342.Em cannot 343be directly accessed by the device. 344The window contains all address greater than lowaddr and 345less than or equal to highaddr. 346For example, a device incapable of DMA above 4GB, would specify 347a highaddr of 348.Dv BUS_SPACE_MAXADDR 349and a lowaddr of 350.Dv BUS_SPACE_MAXADDR_32BIT . 351Similarly a device that can only dma to addresses bellow 16MB would 352specify a highaddr of 353.Dv BUS_SPACE_MAXADDR 354and a lowaddr of 355.Dv BUS_SPACE_MAXADDR_24BIT . 356Some implementations requires that some region of device visible 357address space, overlapping available host memory, be outside the 358window. 359This area of 360.Ql safe memory 361is used to bounce requests that would otherwise conflict with 362the exclusion window. 363.It Fa filtfunc 364Optional filter function (may be NULL) to be called for any attempt to 365map memory into the window described by 366.Fa lowaddr 367and 368.Fa highaddr . 369A filter function is only required when the single window described 370by 371.Fa lowaddr 372and 373.Fa highaddr 374cannot adequately describe the constraints of the device. 375The filter function will be called for every machine page 376that overlaps the exclusion window. 377.It Fa filtfuncarg 378Argument passed to all calls to the filter function for this tag. 379May be NULL. 380.It Fa maxsize 381Maximum size, in bytes, of the sum of all segment lengths in a given 382DMA mapping associated with this tag. 383.It Fa nsegments 384Number of discontinuities (scatter/gather segments) allowed 385in a DMA mapped region. 386If there is no restriction, 387.Dv BUS_SPACE_UNRESTRICTED 388may be specified. 389.It Fa maxsegsz 390Maximum size, in bytes, of a segment in any DMA mapped region associated 391with 392.Fa dmat . 393.It Fa flags 394Are as follows: 395.Bl -tag -width ".Dv BUS_DMA_ALLOCNOW" -compact 396.It Dv BUS_DMA_ALLOCNOW 397Allocate the minimum resources necessary to guarantee that all map load 398operations associated with this tag may not block. 399If sufficient resources are not available, 400.Er ENOMEM 401is returned. 402.It Dv BUS_DMA_WAITOK 403Indicates that it is OK to wait for resources. 404However, 405unlike 406.Xr kmalloc 9 , 407it is not guaranteed that the resource allocation will succeed. 408This flag is the default one, 409if 410.Dv BUS_DMA_NOWAIT 411is not supplied. 412.It Dv BUS_DMA_NOWAIT 413If the resource allocation request cannot be immediately fulfilled, 414.Er ENOMEM 415is returned. 416.It Dv BUS_DMA_ONEBPAGE 417Allocte one bounce page at most, 418even if the 419.Fa maxsize 420indicates that multiple bounce pages are needed. 421.It Dv BUS_DMA_ALIGNED 422Indicates that all memory to be loaded into the DMA maps associated 423with this DMA tag is properly aligned according to 424.Fa alignment 425constraint. 426No resources, 427e.g. bounce pages, 428will be allocated due to the 429.Fa alignment 430constraint. 431If unaligned memory was loaded into the DMA maps associated with this DMA tag, 432system will panic. 433.El 434.It Fa dmat 435Pointer to a bus_dma_tag_t where the resulting DMA tag will 436be stored. 437.El 438.Pp 439Returns 440.Er ENOMEM 441if sufficient memory is not available for tag creation 442or allocating mapping resources. 443.It Fn bus_dma_tag_destroy "dmat" 444Deallocate the DMA tag 445.Fa dmat 446that was created by 447.Fn bus_dma_tag_create . 448.Pp 449Returns 450.Er EBUSY 451if any DMA maps remain associated with 452.Fa dmat 453or 454.Ql 0 455on success. 456.It Fn bus_dmamap_create "dmat" "flags" "*mapp" 457Allocates and initializes a DMA map. 458Arguments are as follows: 459.Bl -tag -width nsegments -compact 460.It Fa dmat 461DMA tag. 462.It Fa flags 463Are as follows: 464.Bl -tag -width ".Dv BUS_DMA_ONEBPAGE" -compact 465.It Dv BUS_DMA_WAITOK 466Indicates that it is OK to wait for resources. 467However, 468unlike 469.Xr kmalloc 9 , 470it is not guaranteed that the resource allocation will succeed. 471This flag is the default one, 472if 473.Dv BUS_DMA_NOWAIT 474is not supplied. 475.It Dv BUS_DMA_NOWAIT 476If the resource allocation request cannot be immediately fulfilled, 477.Er ENOMEM 478is returned. 479.It Dv BUS_DMA_ONEBPAGE 480Allocte one bounce page at most, 481even if the 482.Fa maxsize 483used to create the 484.Fa dmat 485indicates that multiple bounce pages are needed. 486.El 487.It Fa mapp 488Pointer to a 489.Vt bus_dmamap_t 490where the resulting DMA map will be stored. 491.El 492.Pp 493Returns 494.Er ENOMEM 495if sufficient memory is not available for creating the 496map or allocating mapping resources. 497.It Fn bus_dmamap_destroy "dmat" "map" 498Frees all resources associated with a given DMA map. 499Arguments are as follows: 500.Bl -tag -width dmat -compact 501.It Fa dmat 502DMA tag used to allocate 503.Fa map . 504.It Fa map 505The DMA map to destroy. 506.El 507.Pp 508Returns 509.Er EBUSY 510if a mapping is still active for 511.Fa map . 512.It Fn bus_dmamap_load "dmat" "map" "buf" "buflen" "*callback" "..." 513Creates a mapping in device visible address space of 514.Fa buflen 515bytes of 516.Fa buf , 517associated with the DMA map 518.Fa map . 519Arguments are as follows: 520.Bl -tag -width buflen -compact 521.It Fa dmat 522DMA tag used to allocate 523.Fa map . 524.It Fa map 525A DMA map without a currently active mapping. 526.It Fa buf 527A kernel virtual address pointer to a contiguous (in KVA) buffer, to be 528mapped into device visible address space. 529.It Fa buflen 530The size of the buffer. 531.It Fa callback Fa callback_arg 532The callback function, and its argument. 533.It Fa flags 534The value of this argument is currently undefined, and should be 535specified as 536.Ql 0 . 537.El 538.Pp 539Return values to the caller are as follows: 540.Bl -tag -width ".Er EINPROGRESS" -compact 541.It 0 542The callback has been called and completed. 543The status of the mapping has been delivered to the callback. 544.It Er EINPROGRESS 545The mapping has been deferred for lack of resources. 546The callback will be called as soon as resources are available. 547Callbacks are serviced in FIFO order. 548DMA maps created from DMA tags that are allocated with 549the 550.Dv BUS_DMA_ALLOCNOW 551flag will never return this status for a load operation. 552.It Er EINVAL 553The load request was invalid. 554The callback has not, and will not be called. 555This error value may indicate that 556.Fa dmat , 557.Fa map , 558.Fa buf , 559or 560.Fa callback 561were invalid, or 562.Fa buslen 563was larger than the 564.Fa maxsize 565argument used to create the dma tag 566.Fa dmat . 567.El 568.Pp 569When the callback is called, it is presented with an error value 570indicating the disposition of the mapping. 571Error may be one of the following: 572.Bl -tag -width ".Er EINPROGRESS" -compact 573.It 0 574The mapping was successful and the 575.Fa dm_segs 576callback argument contains an array of 577.Vt bus_dma_segment_t 578elements describing the mapping. 579This array is only valid during the scope of the callback function. 580.It Er EFBIG 581A mapping could not be achieved within the segment constraints provided 582in the tag even though the requested allocation size was less than maxsize. 583.El 584.It Fn bus_dmamap_load_mbuf "dmat" "map" "mbuf" "callback2" "callback_arg" \ 585"flags" 586This is a variation of 587.Fn bus_dmamap_load 588which maps mbuf chains 589for DMA transfers. 590A 591.Vt bus_size_t 592argument is also passed to the callback routine, which 593contains the mbuf chain's packet header length. 594.Pp 595Mbuf chains are assumed to be in kernel virtual address space. 596.Pp 597Returns 598.Er EINVAL 599if the size of the mbuf chain exceeds the maximum limit of the 600DMA tag. 601.It Fn bus_dmamap_load_mbuf_segment "dmat" "map" "mbuf" "*segs" "maxsegs" \ 602"*nsegs" "flags" 603It is like 604.Fn bus_dmamap_load_mbuf 605without callback. 606Segmentation information are saved in the 607.Fa segs 608and 609.Fa nsegs 610if the loading is successful. 611The 612.Fa maxsegs , 613which indicates the number of elements in the 614.Fa segs , 615must be set by the caller and must be at least 1 but less than the 616.Fa nsegments 617used to create the 618.Fa dmat . 619The 620.Fa flags 621must have 622.Dv BUS_DMA_NOWAIT 623turned on. 624.Pp 625This function will not block. 626When system is short of DMA resources, 627this function will return 628.Er ENOMEM , 629instead of 630.Er EINPROGRESS . 631.It Fn bus_dmamap_load_mbuf_defrag "dmat" "map" "*mbuf" "*segs" "maxsegs" \ 632"*nsegs" "flags" 633This function is like 634.Fn bus_dmamap_load_mbuf_segment , 635but it will call 636.Fn m_defrag 637on the 638.Fa *mbuf 639and try reloading, 640if low level code indicates too many fragments in the 641.Fa *mbuf ; 642the 643.Fa mbuf 644will be updated under this situation. 645However, 646.Fa *mbuf 647would not be freed by this function, 648even if 649.Fn m_defrag 650failed. 651.Pp 652Return 653.Er ENOBUFS , 654if the calling of 655.Fn m_defrag 656failed. 657.It Fn bus_dmamap_load_uio "dmat" "map" "uio" "callback2" "callback_arg" "flags" 658This is a variation of 659.Fn bus_dmamap_load 660which maps buffers pointed to by 661.Fa uio 662for DMA transfers. 663A 664.Vt bus_size_t 665argument is also passed to the callback routine, which contains the size of 666.Fa uio , 667i.e. 668.Fa uio->uio_resid . 669.Pp 670If 671.Fa uio->uio_segflg 672is 673.Dv UIO_USERSPACE , 674then it is assumed that the buffer, 675.Fa uio 676is in 677.Fa "uio->uio_td->td_proc" Ns 's 678address space. 679User space memory must be in-core and wired prior to attempting a map 680load operation. 681.It Fn bus_dmamap_unload "dmat" "map" 682Unloads a DMA map. 683Arguments are as follows: 684.Bl -tag -width dmam -compact 685.It Fa dmat 686DMA tag used to allocate 687.Fa map . 688.It Fa map 689The DMA map that is to be unloaded. 690.El 691.Pp 692.Fn bus_dmamap_unload 693will not perform any implicit synchronization of DMA buffers. 694This must be done explicitly by a call to 695.Fn bus_dmamap_sync 696prior to unloading the map. 697.It Fn bus_dmamap_sync "dmat" "map" "op" 698Performs synchronization of a device visible mapping with the CPU visible 699memory referenced by that mapping. 700Arguments are as follows: 701.Bl -tag -width dmat -compact 702.It Fa dmat 703DMA tag used to allocate 704.Fa map . 705.It Fa map 706The DMA mapping to be synchronized. 707.It Fa op 708Type of synchronization operation to perform. 709See the definition of 710.Vt bus_dmasync_op_t 711for a description of the acceptable values for 712.Fa op . 713.El 714.Pp 715.Fn bus_dmamap_sync 716is the method used to ensure that CPU and device DMA access to shared 717memory is coherent. 718For example, the CPU might be used to setup the contents of a buffer 719that is to be DMA'ed into a device. 720To ensure that the data are visible via the device's mapping of that 721memory, the buffer must be loaded and a dma sync operation of 722.Dv BUS_DMASYNC_PREREAD 723must be performed. 724Additional sync operations must be performed after every CPU write 725to this memory if additional DMA reads are to be performed. 726Conversely, for the DMA write case, the buffer must be loaded, 727and a dma sync operation of 728.Dv BUS_DMASYNC_PREWRITE 729must be performed. 730The CPU will only be able to see the results of this DMA write 731once the DMA has completed and a 732.Dv BUS_DMASYNC_POSTWRITE 733operation has been performed. 734.Pp 735If DMA read and write operations are not preceded and followed by the 736appropriate synchronization operations, behavior is undefined. 737.It Fn bus_dmamem_alloc "dmat" "**vaddr" "flags" "mapp" 738Allocates memory that is mapped into KVA at the address returned 739in 740.Fa vaddr 741that is permanently loaded into the newly created 742.Vt bus_dmamap_t 743returned via 744.Fa mapp . 745Arguments are as follows: 746.Bl -tag -width alignment -compact 747.It Fa dmat 748DMA tag describing the constraints of the DMA mapping. 749.It Fa vaddr 750Pointer to a pointer that will hold the returned KVA mapping of 751the allocated region. 752.It Fa flags 753Flags are defined as follows: 754.Bl -tag -width ".Dv BUS_DMA_NOWAIT" -compact 755.It Dv BUS_DMA_WAITOK 756The routine can safely wait (sleep) for resources. 757.It Dv BUS_DMA_NOWAIT 758The routine is not allowed to wait for resources. 759If resources are not available, 760.Er ENOMEM 761is returned. 762.It Dv BUS_DMA_COHERENT 763Attempt to map this memory such that cache sync operations are 764as cheap as possible. 765This flag is typically set on memory that will be accessed by both 766a CPU and a DMA engine, frequently. 767Use of this flag does not remove the requirement of using 768bus_dmamap_sync, but it may reduce the cost of performing 769these operations. 770.It Dv BUS_DMA_ZERO 771Causes the allocated memory to be set to all zeros. 772.El 773.It Fa mapp 774Pointer to storage for the returned DMA map. 775.El 776.Pp 777The size of memory to be allocated is 778.Fa maxsize 779as specified in 780.Fa dmat . 781.Pp 782The current implementation of 783.Fn bus_dmamem_alloc 784will allocate all requests as a single segment. 785.Pp 786Although no explicit loading is required to access the memory 787referenced by the returned map, the synchronization requirements 788as described in the 789.Fn bus_dmamap_sync 790section still apply. 791.Pp 792Returns 793.Er ENOMEM 794if sufficient memory is not available for completing 795the operation. 796.It Fn bus_dmamem_coherent "parent" "alignment" "boundary" "lowaddr" \ 797"highaddr" "maxsize" "flags" "*dmem" 798This is a convenient function to create one segment of DMA memory. 799It combines following 800.Xr bus_dma 9 801function calls: 802.Bd -literal 803 bus_dma_tag_create(..., dtag); 804 bus_dmamem_alloc(*dtag, vaddr, ..., dmap); 805 bus_dmamap_load(*dtag, *dmap, *vaddr, ..., \\ 806 callback, busaddr, ...); 807.Ed 808.sp 809The final results of the above function calls are: 810DMA tag, 811DMA map, 812DMA memory's kernel virtual address and 813its device visible address. 814.Fn bus_dmamem_coherent 815saves the results in 816.Fa *dmem . 817.Pp 818The 819.Fa parent , 820.Fa alignment , 821.Fa boundary , 822.Fa lowaddr 823and 824.Fa highaddr 825will be passed to 826.Fn bus_dma_tag_create 827as they are. 828The 829.Fa maxsize 830will be passed to 831.Fn bus_dma_tag_create 832as its 833.Fa maxsize 834and 835.Fa maxsegsz 836and 837.Ql 1 838will be passed to 839.Fn bus_dma_tag_create 840as its 841.Fa nsegments . 842When 843.Fn bus_dmamem_alloc 844is called, 845.Fa flags 846will be first or'ed with 847.Dv BUS_DMA_COHERENT 848then passed to it. 849The final results of the above three functions, 850i.e. DMA tag, 851DMA map, 852DMA memory's kernel virtual address and 853its device visible address, 854are saved in 855.Fa *dmem . 856If any of the three functions failed, 857this function will return the error code and the 858.Fa *dmem 859should not be used. 860.It Fn bus_dmamem_coherent_any "parent" "alignment" "maxsize" "flags" \ 861"*dtag" "*dmap" "*busaddr" 862This function is a simplified version of 863.Fn bus_dmamem_coherent 864with 865its 866.Fa boundary 867set to 868.Ql 0 , 869.Fa lowaddr 870set to 871.Dv BUS_SPACE_MAXADDR 872and 873.Fa highaddr 874set to 875.Dv BUS_SPACE_MAXADDR . 876The 877.Fa parent 878usually should not be NULL. 879.Pp 880Return the DMA memory's kernel virtual address. 881The DMA tag, DMA map and device visible address are returned in 882.Fa *dtag , 883.Fa *dmap , 884and 885.Fa *busaddr . 886If this function failed, 887NULL will be returned; 888.Fa *dtag , 889.Fa *dmap , 890and 891.Fa *busaddr 892are left unchanged. 893.It Fn bus_dmamem_free "dmat" "*vaddr" "map" 894Frees memory previously allocated by 895.Fn bus_dmamem_alloc . 896Any mappings 897will be invalidated. 898Arguments are as follows: 899.Bl -tag -width vaddr -compact 900.It Fa dmat 901DMA tag. 902.It Fa vaddr 903Kernel virtual address of the memory. 904.It Fa map 905DMA map to be invalidated. 906.El 907.El 908.Sh RETURN VALUES 909Behavior is undefined if invalid arguments are passed to 910any of the above functions. 911If sufficient resources cannot be allocated for a given 912transaction, 913.Er ENOMEM 914is returned. 915All 916routines that are not of type, 917.Vt void , 918will return 0 on success or an error 919code, as discussed above. 920.Pp 921All 922.Vt void 923routines will succeed if provided with valid arguments. 924.Sh SEE ALSO 925.Xr devclass 9 , 926.Xr device 9 , 927.Xr driver 9 , 928.Xr rman 9 929.Pp 930.Rs 931.%A "Jason R. Thorpe" 932.%T "A Machine-Independent DMA Framework for NetBSD" 933.%J "Proceedings of the Summer 1998 USENIX Technical Conference" 934.%Q "USENIX Association" 935.%D "June 1998" 936.Re 937.Sh HISTORY 938The 939.Nm 940interface first appeared in 941.Nx 1.3 . 942.Pp 943The 944.Nm 945API was adopted from 946.Nx 947for use in the CAM SCSI subsystem. 948The alterations to the original API were aimed to remove the need for 949a 950.Vt bus_dma_segment_t 951array stored in each 952.Vt bus_dmamap_t 953while allowing callers to queue up on scarce resources. 954.Sh AUTHORS 955The 956.Nm 957interface was designed and implemented by 958.An Jason R. Thorpe 959of the Numerical Aerospace Simulation Facility, NASA Ames Research Center. 960Additional input on the 961.Nm 962design was provided by 963.An -nosplit 964.An Chris Demetriou , 965.An Charles Hannum , 966.An Ross Harvey , 967.An Matthew Jacob , 968.An Jonathan Stone , 969and 970.An Matt Thomas . 971.Pp 972This manual page was written by 973.An Hiten Pandya 974and 975.An Justin T. Gibbs . 976