BUS_DMA(9) BSD Kernel Manual BUS_DMA(9)NAME
bus_dma - bus and machine independent DMA mapping interface
SYNOPSIS
#include <machine/bus.h>
DESCRIPTION
The bus_dma interface provides a bus and machine independent mechanism
for managing DMA data transfers to and from devices.
The basic abstraction is the bus_dmamap_t, a pointer to a structure which
contains an array of bus_dma_segment_t's (dm_segs) and a count of how
many are currently valid (dm_nsegs).
Each segment in the array describes a single physical area of memory
which can be DMA'd, with a starting address (ds_addr) and a length
(ds_len). These are the values that must be communicated to the DMA dev-
ice. Taken together the segments exactly and completely describe the
buffer being used to transfer data.
bus_dma_tag_t's are an opaque type, received from higher software layers
and are never created, changed, deleted or even examined in this inter-
face.
The basic cycle to transfer data to/from a DMA device is:
bus_dmamap_create(); /* get a dmamap to load/unload */
for each DMA xfer {
bus_dmamem_alloc(); /* allocate some DMA'able memory */
bus_dmamem_map(); /* map it into the kernel address space */
/*
* Fill the allocated DMA'able memory with whatever data
* is to be sent out, using the pointer obtained with
* bus_dmamem_map().
*/
bus_dmamap_load(); /* initialize the segments of dmamap */
bus_dmamap_sync(); /* synchronize/flush any DMA cache */
for (i=0; i<dm_nsegs; i++) {
/*
* Tell the DMA device the physical address
* (dmamap->dm_segs[i].ds_addr) and the length
* (dmamap->dm_segs[i].ds_len) of the memory to xfer.
*
* Start the DMA, wait until it's done
*/
}
bus_dmamap_sync(); /* synchronize/flush any DMA cache */
bus_dmamap_unload(); /* prepare dmamap for reuse */
/*
* Copy any data desired from the DMA'able memory using the
* pointer created by bus_dmamem_map().
*/
bus_dmamem_unmap(); /* free kernel virtual address space */
bus_dmamem_free(); /* free DMA'able memory */
}
bus_dmamap_destroy(); /* release any resources used by dmamap */
DATA TYPES
Individual implementations may name these structures whatever they wish,
providing that the external representations are:
bus_addr_t A device bus address to be used for CPU access or DMA.
bus_size_t The size of a bus address range.
bus_dma_tag_t A machine-dependent opaque type describing the imple-
mentation of DMA for a given host/bus. Machine-
dependent code is responsible for passing these struc-
tures to a bus's autoconfiguration machinery, which in
turn passes it down to the device drivers.
bus_dma_segment_t A structure describing an individual DMA segment. The
structure may have machine-dependent members and arbi-
trary layout, but has at least the following members:
bus_addr_t ds_addr;
bus_size_t ds_len;
The values in ds_addr and ds_len are suitable for pro-
gramming into a DMA controller's address and length
registers.
bus_dmamap_t A pointer to a structure describing an individual DMA
mapping. The structure may have machine-dependent
members and arbitrary layout, but has at least the
following members:
int dm_nsegs;
bus_dma_segment_t *dm_segs;
The dm_segs member may be an array of segments or a
pointer to an array of segments. The dm_nsegs member
indicates the number of segments in dm_segs.
DMA MAPS
int
bus_dmamap_create(bus_dma_tag_t tag, bus_size_t size, int nsegments,
bus_size_t maxsegsz, bus_size_t boundary, int flags,
bus_dmamap_t *dmamp);
void
bus_dmamap_destroy(bus_dma_tag_t tag, bus_dmamap_t dmam);
The bus_dmamap_create() function allocates a DMA handle and initializes
it according to the parameters provided. This function returns 0 on suc-
cess, an error code otherwise.
The bus_dmamap_create() arguments are as follows:
tag This is the bus_dma_tag_t passed down from the parent driver
via <bus>_attach_args.
size This is the maximum DMA transfer that can be mapped by the
handle.
nsegments Number of segments the device can support in a single DMA
transaction. This may be the number of scatter-gather descrip-
tors supported by the device.
maxsegsz The maximum number of bytes that may be transferred by any
given DMA segment.
boundary Some DMA controllers are not able to transfer data that
crosses a particular boundary. This argument allows this boun-
dary to be specified. The boundary lines begin at 0, and occur
every boundary bytes. Mappings may begin on a boundary line
but may not end on or cross a boundary line. If no boundary
condition needs to be observed, a boundary argument of 0
should be used.
flags Flags are defined as follows:
BUS_DMA_WAITOK It is safe to wait (sleep) for resources
during this call.
BUS_DMA_NOWAIT It is not safe to wait (sleep) for resources
during this call.
BUS_DMA_ALLOCNOW Perform any resource allocation this handle
may need now. If this is not specified, the
allocation may be deferred to
bus_dmamap_load(). If this flag is speci-
fied, bus_dmamap_load() will not block on
resource allocation.
BUS_DMA_BUS[1-4] These flags are placeholders, and may be
used by busses to provide bus-dependent
functionality.
dmamp This is a pointer to a bus_dmamap_t. A DMA map will be allo-
cated and pointed to by dmamp upon successful completion of
this routine.
The bus_dmamap_destroy() function frees all resources associated with a
given DMA handle. This function always succeeds if given valid arguments.
The bus_dmamap_destroy() arguments are as follows:
tag This is the bus_dma_tag_t passed down from the parent driver via
<bus>_attach_args.
dmam The DMA handle to destroy.
In the event that the DMA handle contains a valid mapping, the mapping
will be unloaded via the same mechanism used by bus_dmamap_unload().
DMA MAP SEGMENTS
int
bus_dmamap_load(bus_dma_tag_t tag, bus_dmamap_t dmam, void *buf,
bus_size_t buflen, struct proc *p, int flags);
int
bus_dmamap_load_mbuf(bus_dma_tag_t tag, bus_dmamap_t dmam,
struct mbuf *chain, int flags);
int
bus_dmamap_load_uio(bus_dma_tag_t tag, bus_dmamap_t dmam,
struct uio *uio, int flags);
int
bus_dmamap_load_raw(bus_dma_tag_t tag, bus_dmamap_t dmam,
bus_dma_segment_t *segs, int nsegs, bus_size_t size, int flags);
void
bus_dmamap_unload(bus_dma_tag_t tag, bus_dmamap_t dmam);
The bus_dmamap_load() function loads a DMA handle with mappings for a DMA
transfer. It assumes that all pages involved in a DMA transfer are wired.
This function returns 0 on success, an error code otherwise.
The bus_dmamap_load() arguments are as follows:
tag This is the bus_dma_tag_t passed down from the parent driver via
<bus>_attach_args.
dmam The DMA handle with which to map the transfer.
buf The buffer to be used for the DMA transfer.
buflen The size of the buffer.
p Used to indicate the address space in which the buffer is locat-
ed. If NULL, the buffer is assumed to be in kernel space. Other-
wise, the buffer is assumed to be in process p's address space.
flags Flags are defined as follows:
BUS_DMA_WAITOK It is safe to wait (sleep) for resources dur-
ing this call.
BUS_DMA_NOWAIT It is not safe to wait (sleep) for resources
during this call.
BUS_DMA_BUS[1-4] These flags are placeholders, and may be used
by busses to provide bus-dependent functional-
ity.
BUS_DMA_STREAMING By default, the bus_dma API assumes that there
is coherency between memory and the device
performing the DMA transaction. Some plat-
forms, however, have special hardware, such as
an "I/O cache", which may improve performance
of some types of DMA transactions, but which
break the assumption that there is coherency
between memory and the device performing the
DMA transaction. This flag allows the use of
this special hardware, provided that the dev-
ice is doing sequential, unidirectional
transfers which conform to certain alignment
and size constraints defined by the platform.
If the platform does not support the feature,
or if the buffer being loaded into the DMA map
does not conform to the constraints required
for use of the feature, then this flag will be
silently ignored. Also refer to the use of
this flag with the bus_dmamem_alloc() func-
tion.
BUS_DMA_READ This is a hint to the machine-dependent back-
end that indicates the mapping will be used
only for a device -> memory transaction. The
back-end may perform optimizations based on
this information.
BUS_DMA_WRITE This is a hint to the machine-dependent back-
end that indicates the mapping will be used
only for a memory -> device transaction. The
back-end may perform optimizations based on
this information.
As noted above, if a DMA handle is created with BUS_DMA_ALLOCNOW,
bus_dmamap_load() will never block.
If a call to bus_dmamap_load() fails, the mapping in the DMA handle will
be invalid. It is the responsibility of the caller to clean up any incon-
sistent device state resulting from incomplete iteration through the uio.
The bus_dmamap_load_mbuf() function is a variation of bus_dmamap_load()
which maps mbuf chains for DMA transfers. Mbuf chains are assumed to be
in kernel virtual address space.
The bus_dmamap_load_uio() function is a variation of bus_dmamap_load()
which maps buffers pointed to by uio for DMA transfers. The value of
uio->uio_segflg will determine if the buffers are in user or kernel vir-
tual address space. If the buffers are in user address space, the buffers
are assumed to be in uio->uio_procp's address space.
The bus_dmamap_load_raw() function is a variation of bus_dmamap_load()
which maps buffers allocated by bus_dmamem_alloc() (see below). The segs
argument is an array of bus_dma_segment_t's filled in by
bus_dmamem_alloc(). The nsegs argument is the number of segments in the
array. The size argument is the size of the DMA transfer.
The bus_dmamap_unload() function deletes the mappings for a given DMA
handle. This function always succeeds if given valid arguments.
The bus_dmamap_unload() arguments are as follows:
tag This is the bus_dma_tag_t passed down from the parent driver via
<bus>_attach_args.
dmam The DMA handle containing the mappings which are to be deleted.
If the DMA handle was created with BUS_DMA_ALLOCNOW, bus_dmamap_unload()
will not free the corresponding resources which were allocated by
bus_dmamap_create(). This is to ensure that bus_dmamap_load() will never
block on resources if the handle was created with BUS_DMA_ALLOCNOW.
SYNCHRONIZATION
void
bus_dmamap_sync(bus_dma_tag_t tag, bus_dmamap_t dmam, bus_addr_t offset,
bus_size_t size, bus_dmasync_op_t ops);
The bus_dmamap_sync() function performs pre- and post-DMA operation cache
and/or buffer synchronization. This function always succeeds if given
valid arguments.
The bus_dmamap_sync() arguments are as follows:
tag This is the bus_dma_tag_t passed down from the parent driver via
<bus>_attach_args.
dmam The DMA mapping to be synchronized.
offset
Offset in the DMA mapping to be synchronized.
size The size of the region to be synchronized.
ops One or more synchronization operations to perform. The following
DMA synchronization operations are defined:
BUS_DMASYNC_PREREAD Perform any pre-read DMA cache and/or bounce
operations.
BUS_DMASYNC_POSTREAD Perform any post-read DMA cache and/or
bounce operations.
BUS_DMASYNC_PREWRITE Perform any pre-write DMA cache and/or
bounce operations.
BUS_DMASYNC_POSTWRITE Perform any post-write DMA cache and/or
bounce operations.
More than one operation may performed in a given synchronization
call. Mixing of PRE and POST operations is not allowed, and
behavior is undefined if this is attempted.
Synchronization operations are expressed from the perspective of the host
RAM, e.g., a device -> memory operation is a READ and a memory -> device
operation is a WRITE.
bus_dmamap_sync() may consult state kept within the DMA map to determine
if the memory is mapped in a DMA coherent fashion. If so,
bus_dmamap_sync() may elect to skip certain expensive operations, such as
flushing of the data cache. See bus_dmamem_map() for more information on
this subject.
On platforms which implement reordered stores, bus_dmamap_sync() will al-
ways cause the store buffer to be flushed.
This function exists so that multiple read and write transfers can be
performed with the same buffer, and so that drivers can explicitly inform
the bus_dma code when their data is "ready" in its DMA buffer.
An example of multiple read-write use of a single mapping might look
like:
bus_dmamap_load(...);
while (not done) {
/* invalidate soon-to-be-stale cache blocks */
bus_dmamap_sync(..., BUS_DMASYNC_PREREAD);
[ do read DMA ]
/* copy from bounce */
bus_dmamap_sync(..., BUS_DMASYNC_POSTREAD);
/* read data now in driver-provided buffer */
[ computation ]
/* data to be written now in driver-provided buffer */
/* flush write buffers and writeback, copy to bounce */
bus_dmamap_sync(..., BUS_DMASYNC_PREWRITE);
[ do write DMA ]
/* probably a no-op, but provided for consistency */
bus_dmamap_sync(..., BUS_DMASYNC_POSTWRITE);
}
bus_dmamap_unload(...);
If DMA read and write operations are not preceded and followed by the ap-
propriate synchronization operations, behavior is undefined.
DMA-SAFE MEMORY
int
bus_dmamem_alloc(bus_dma_tag_t tag, bus_size_t size,
bus_size_t alignment, bus_size_t boundary,
bus_dma_segment_t *segs, int nsegs, int *rsegs, int flags);
void
bus_dmamem_free(bus_dma_tag_t tag, bus_dma_segment_t *segs, int nsegs);
The bus_dmamem_alloc() function allocates memory that is "DMA safe" for
the bus corresponding to the given tag. This function returns 0 on suc-
cess, or an error code indicating mode of failure.
The mapping of this memory is machine-dependent (or "opaque"); machine-
independent code should not assume that the addresses returned are valid
in kernel virtual address space, or that the addresses returned are sys-
tem physical addresses. The address value returned as part of segs can
thus not be used to program DMA controller address registers. Only the
values in the dm_segs array of a successfully loaded DMA map (using
bus_dmamap_load()) can be used for this purpose.
Allocations will always be rounded to the hardware page size. Callers may
wish to take advantage of this, and cluster allocation of small data
structures.
The bus_dmamem_alloc() arguments are as follows:
tag This is the bus_dma_tag_t passed down from the parent driver
via <bus>_attach_args.
size The amount of memory to allocate.
alignment Each segment in the allocated memory will be aligned to this
value. If the alignment is less than a hardware page size, it
will be rounded up to the hardware page size. This value must
be a power of two.
boundary Each segment in the allocated memory must not cross this boun-
dary (relative to zero). This value must be a power of two. A
boundary value less than the size of the allocation is in-
valid.
segs An array of bus_dma_segment_t's, filled in as memory is allo-
cated, representing the opaque addresses of the memory chunks.
nsegs Specifies the number of segments in segs, and this is the max-
imum number of segments that the allocated memory may contain.
rsegs Used to return the actual number of segments the memory con-
tains.
flags Flags are defined as follows:
BUS_DMA_WAITOK It is safe to wait (sleep) for resources
during this call.
BUS_DMA_NOWAIT It is not safe to wait (sleep) for
resources during this call.
BUS_DMA_STREAMING Adjusts, if necessary, the size, alignment,
and boundary constraints to conform to the
platform-dependent requirements for the use
of the BUS_DMA_STREAMING flag with the
bus_dmamap_load() function. If the platform
does not support the BUS_DMA_STREAMING
feature, or if the size, alignment, and
boundary constraints would already satisfy
the platform's requirements, this flag is
silently ignored. The BUS_DMA_STREAMING
flag will never relax the constraints
specified in the call.
BUS_DMA_BUS[1-4] These flags are placeholders, and may be
used by busses to provide bus-dependent
functionality.
All pages allocated by bus_dmamem_alloc() will be wired down until they
are freed by bus_dmamem_free().
The bus_dmamem_free() function frees memory previously allocated by
bus_dmamem_alloc(), invalidating any mapping. This function always
succeeds if given valid arguments.
The bus_dmamem_free() arguments are as follows:
tag This is the bus_dma_tag_t passed down from the parent driver via
<bus>_attach_args.
segs The array of bus_dma_segment_t's filled in by bus_dmamem_alloc().
nsegs The number of segments in segs.
MAPPING DMA-SAFE MEMORY
int
bus_dmamem_map(bus_dma_tag_t tag, bus_dma_segment_t *segs, int nsegs,
size_t size, caddr_t *kvap, int flags);
void
bus_dmamem_unmap(bus_dma_tag_t tag, caddr_t kva, size_t size);
paddr_t
bus_dmamem_mmap(bus_dma_tag_t tag, bus_dma_segment_t *segs, int nsegs,
off_t off, int prot, int flags);
The bus_dmamem_map() function maps memory allocated with
bus_dmamem_alloc() into kernel virtual address space. This function re-
turns 0 on success, an error code otherwise, and must not be called from
an interrupt context.
The bus_dmamem_map() arguments are as follows:
tag This is the bus_dma_tag_t passed down from the parent driver via
<bus>_attach_args.
segs The array of bus_dma_segment_t's filled in by bus_dmamem_alloc(),
representing the memory regions to map.
nsegs The number of segments in segs.
size The size of the mapping.
kvap Filled in to specify the kernel virtual address where the memory
is mapped.
flags Flags are defined as follows:
BUS_DMA_WAITOK It is safe to wait (sleep) for resources during
this call.
BUS_DMA_NOWAIT It is not safe to wait (sleep) for resources
during this call.
BUS_DMA_BUS[1-4] These flags are placeholders, and may be used by
busses to provide bus-dependent functionality.
BUS_DMA_COHERENT This flag is a hint to machine-dependent code.
If possible, map the memory in such a way as it
will be DMA coherent. This may include mapping
the pages into uncached address space or setting
the cache-inhibit bits in page table entries. If
implementation of DMA coherent mappings is im-
possible, this is ignored.
Later, when this memory is loaded into a DMA
map, machine-dependent code will take whatever
steps are necessary to determine if the memory
was mapped in a DMA coherent fashion. This may
include checking if the kernel virtual address
lies within uncached address space or if the
cache-inhibit bits are set in page table en-
tries. If it is determined that the mapping is
DMA coherent, state may be placed into the DMA
map for use by later calls to bus_dmamap_sync().
The bus_dmamem_unmap() function unmaps memory previously mapped with
bus_dmamem_map(), freeing the kernel virtual address space used by the
mapping. This function always succeeds if given valid arguments, but must
not be called from an interrupt context.
bus_dmamem_unmap() arguments are as follows:
tag This is the bus_dma_tag_t passed down from the parent driver via
<bus>_attach_args.
kva The kernel virtual address of the mapped memory.
size The size of the mapping.
The bus_dmamem_mmap() function provides support for user mmap(2)'ing of
DMA-safe memory. bus_dmamem_mmap() is to be called by a device driver's
(*d_mmap)() entry point, which is called by the device pager for each
page to be mapped. This function returns an opaque value to be interpret-
ed by the device pager, or -1 on failure. bus_dmamem_mmap() arguments are
as follows:
tag This is the bus_dma_tag_t passed down from the parent driver via
<bus>_attach_args.
segs The array of bus_dma_segment_t's filled in by bus_dmamem_alloc(),
representing the memory to be mmap(2)'ed.
nsegs The number of elements in the segs array.
off The offset of the page in DMA memory which is to be mapped.
prot The protection codes for the mapping.
flags Flags are defined as follows:
BUS_DMA_WAITOK It is safe to wait (sleep) for resources during
this call.
BUS_DMA_NOWAIT It is not safe to wait (sleep) for resources
during this call.
BUS_DMA_BUS[1-4] These flags are placeholders, and may be used by
busses to provide bus-dependent functionality.
BUS_DMA_COHERENT See bus_dmamem_map() above for a description of
this flag.
SEE ALSObus_space(9)HISTORY
The bus_dma interface appeared in NetBSD 1.3.
AUTHORS
The bus_dma interface was designed and implemented by Jason R. Thorpe of
the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
Additional input on the bus_dma design was provided by Chris Demetriou,
Charles Hannum, Ross Harvey, Matthew Jacob, Jonathan Stone, and Matt Tho-
mas.
MirOS BSD #10-current November 23, 2000 7