man/man9/bus_space.9

.\"	$OpenBSD: bus_space.9,v 1.20 2009/07/26 18:48:53 miod Exp $
.\" $NetBSD: bus_space.9,v 1.15 2000/08/09 03:11:00 tv Exp $
.\"
.\" Copyright (c) 1997 The NetBSD Foundation, Inc.
.\" All rights reserved.
.\"
.\" This code is derived from software contributed to The NetBSD Foundation
.\" by Christopher G. Demetriou.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\"    notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\"    notice, this list of conditions and the following disclaimer in the
.\"    documentation and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
.\" ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
.\" TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
.\" PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
.\" BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
.\" CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
.\" SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
.\" INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
.\" CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
.\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
.\" POSSIBILITY OF SUCH DAMAGE.
.\"
.Dd $Mdocdate: July 26 2009 $
.Dt BUS_SPACE 9
.Os
.Sh NAME
.Nm bus_space ,
.Nm bus_space_alloc ,
.Nm bus_space_barrier ,
.Nm bus_space_copy_1 ,
.Nm bus_space_copy_2 ,
.Nm bus_space_copy_4 ,
.Nm bus_space_copy_8 ,
.Nm bus_space_free ,
.Nm bus_space_map ,
.Nm bus_space_read_1 ,
.Nm bus_space_read_2 ,
.Nm bus_space_read_4 ,
.Nm bus_space_read_8 ,
.Nm bus_space_read_multi_1 ,
.Nm bus_space_read_multi_2 ,
.Nm bus_space_read_multi_4 ,
.Nm bus_space_read_multi_8 ,
.\".Nm bus_space_read_raw_multi_1 ,
.Nm bus_space_read_raw_multi_2 ,
.Nm bus_space_read_raw_multi_4 ,
.Nm bus_space_read_raw_multi_8 ,
.Nm bus_space_read_region_1 ,
.Nm bus_space_read_region_2 ,
.Nm bus_space_read_region_4 ,
.Nm bus_space_read_region_8 ,
.\".Nm bus_space_read_raw_region_1 ,
.Nm bus_space_read_raw_region_2 ,
.Nm bus_space_read_raw_region_4 ,
.Nm bus_space_read_raw_region_8 ,
.\".Nm bus_space_read_stream_1 ,
.\".Nm bus_space_read_stream_2 ,
.\".Nm bus_space_read_stream_4 ,
.\".Nm bus_space_read_stream_8 ,
.Nm bus_space_set_multi_1 ,
.Nm bus_space_set_multi_2 ,
.Nm bus_space_set_multi_4 ,
.Nm bus_space_set_multi_8 ,
.Nm bus_space_set_region_1 ,
.Nm bus_space_set_region_2 ,
.Nm bus_space_set_region_4 ,
.Nm bus_space_set_region_8 ,
.Nm bus_space_subregion ,
.Nm bus_space_unmap ,
.Nm bus_space_vaddr ,
.Nm bus_space_write_1 ,
.Nm bus_space_write_2 ,
.Nm bus_space_write_4 ,
.Nm bus_space_write_8 ,
.Nm bus_space_write_multi_1 ,
.Nm bus_space_write_multi_2 ,
.Nm bus_space_write_multi_4 ,
.Nm bus_space_write_multi_8 ,
.\".Nm bus_space_write_raw_multi_1 ,
.Nm bus_space_write_raw_multi_2 ,
.Nm bus_space_write_raw_multi_4 ,
.Nm bus_space_write_raw_multi_8 ,
.Nm bus_space_write_region_1 ,
.Nm bus_space_write_region_2 ,
.Nm bus_space_write_region_4 ,
.Nm bus_space_write_region_8 ,
.\".Nm bus_space_write_raw_region_1 ,
.Nm bus_space_write_raw_region_2 ,
.Nm bus_space_write_raw_region_4 ,
.Nm bus_space_write_raw_region_8 ,
.\".Nm bus_space_write_stream_1 ,
.\".Nm bus_space_write_stream_2 ,
.\".Nm bus_space_write_stream_4 ,
.\".Nm bus_space_write_stream_8 ,
.Nd bus space manipulation functions
.Sh SYNOPSIS
.Fd #include <machine/bus.h>
.Ft int
.Fn bus_space_map "bus_space_tag_t space" "bus_addr_t address" \
"bus_size_t size" "int flags" "bus_space_handle_t *handlep"
.Ft void
.Fn bus_space_unmap "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t size"
.Ft int
.Fn bus_space_subregion "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset" "bus_size_t size" "bus_space_handle_t *nhandlep"
.Ft int
.Fo bus_space_alloc
.Fa "bus_space_tag_t space" "bus_addr_t reg_start" "bus_addr_t reg_end"
.Fa "bus_size_t size" "bus_size_t alignment" "bus_size_t boundary"
.Fa "int flags" "bus_addr_t *addrp" "bus_space_handle_t *handlep"
.Fc
.Ft void
.Fn bus_space_free "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t size"
.Ft void *
.Fn bus_space_vaddr "bus_space_tag_t space" "bus_space_handle_t handle"
.Ft u_int8_t
.Fn bus_space_read_1 "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset"
.Ft u_int16_t
.Fn bus_space_read_2 "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset"
.Ft u_int32_t
.Fn bus_space_read_4 "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset"
.Ft u_int64_t
.Fn bus_space_read_8 "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset"
.Ft void
.Fn bus_space_write_1 "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset" "u_int8_t value"
.Ft void
.Fn bus_space_write_2 "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset" "u_int16_t value"
.Ft void
.Fn bus_space_write_4 "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset" "u_int32_t value"
.Ft void
.Fn bus_space_write_8 "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset" "u_int64_t value"
.Ft void
.Fn bus_space_barrier "bus_space_tag_t space" "bus_space_handle_t handle" \
"bus_size_t offset" "bus_size_t length" "int flags"
.Ft void
.Fn bus_space_read_region_1 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_read_region_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int16_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_read_region_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int32_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_read_region_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int64_t *datap" \
"bus_size_t count"
.\".Ft void
.\".Fn bus_space_read_raw_region_1 "bus_space_tag_t space" \
.\""bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
.\""bus_size_t count"
.Ft void
.Fn bus_space_read_raw_region_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_read_raw_region_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_read_raw_region_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_write_region_1 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_write_region_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int16_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_write_region_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int32_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_write_region_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int64_t *datap" \
"bus_size_t count"
.\".Ft void
.\".Fn bus_space_write_raw_region_1 "bus_space_tag_t space" \
.\""bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
.\""bus_size_t count"
.Ft void
.Fn bus_space_write_raw_region_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_write_raw_region_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_write_raw_region_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_copy_1 "bus_space_tag_t space" \
"bus_space_handle_t srchandle" "bus_size_t srcoffset" \
"bus_space_handle_t dsthandle" "bus_size_t dstoffset" "bus_size_t count"
.Ft void
.Fn bus_space_copy_2 "bus_space_tag_t space" \
"bus_space_handle_t srchandle" "bus_size_t srcoffset" \
"bus_space_handle_t dsthandle" "bus_size_t dstoffset" "bus_size_t count"
.Ft void
.Fn bus_space_copy_4 "bus_space_tag_t space" \
"bus_space_handle_t srchandle" "bus_size_t srcoffset" \
"bus_space_handle_t dsthandle" "bus_size_t dstoffset" "bus_size_t count"
.Ft void
.Fn bus_space_copy_8 "bus_space_tag_t space" \
"bus_space_handle_t srchandle" "bus_size_t srcoffset" \
"bus_space_handle_t dsthandle" "bus_size_t dstoffset" "bus_size_t count"
.Ft void
.Fn bus_space_set_multi_1 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t value" \
"bus_size_t count"
.Ft void
.Fn bus_space_set_multi_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int16_t value" \
"bus_size_t count"
.Ft void
.Fn bus_space_set_multi_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int32_t value" \
"bus_size_t count"
.Ft void
.Fn bus_space_set_multi_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int64_t value" \
"bus_size_t count"
.Ft void
.Fn bus_space_set_region_1 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t value" \
"bus_size_t count"
.Ft void
.Fn bus_space_set_region_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int16_t value" \
"bus_size_t count"
.Ft void
.Fn bus_space_set_region_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int32_t value" \
"bus_size_t count"
.Ft void
.Fn bus_space_set_region_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int64_t value" \
"bus_size_t count"
.Ft void
.Fn bus_space_read_multi_1 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_read_multi_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int16_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_read_multi_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int32_t *datap" \
"bus_size_t count"
.Ft void
.Fn bus_space_read_multi_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int64_t *datap" \
"bus_size_t count"
.\".Ft void
.\".Fn bus_space_read_raw_multi_1 "bus_space_tag_t space" \
.\""bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
.\""bus_size_t size"
.Ft void
.Fn bus_space_read_raw_multi_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
"bus_size_t size"
.Ft void
.Fn bus_space_read_raw_multi_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
"bus_size_t size"
.Ft void
.Fn bus_space_read_raw_multi_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "u_int8_t *datap" \
"bus_size_t size"
.Ft void
.Fn bus_space_write_multi_1 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
"bus_size_t size"
.Ft void
.Fn bus_space_write_multi_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int16_t *datap" \
"bus_size_t size"
.Ft void
.Fn bus_space_write_multi_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int32_t *datap" \
"bus_size_t size"
.Ft void
.Fn bus_space_write_multi_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int64_t *datap" \
"bus_size_t size"
.\".Ft void
.\".Fn bus_space_write_raw_multi_1 "bus_space_tag_t space" \
.\""bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
.\""bus_size_t size"
.Ft void
.Fn bus_space_write_raw_multi_2 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
"bus_size_t size"
.Ft void
.Fn bus_space_write_raw_multi_4 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
"bus_size_t size"
.Ft void
.Fn bus_space_write_raw_multi_8 "bus_space_tag_t space" \
"bus_space_handle_t handle" "bus_size_t offset" "const u_int8_t *datap" \
"bus_size_t size"
.Sh DESCRIPTION
The
.Nm
functions exist to allow device drivers machine-independent access to
bus memory and register areas.
All of the functions and types described in this document can be used by
including the
.Aq Pa machine/bus.h
header file.
.Pp
Many common devices are used on multiple architectures, but are accessed
differently on each because of architectural constraints.
For instance, a device which is mapped in one system's I/O space may be
mapped in memory space on a second system.
On a third system, architectural limitations might change the way
registers need to be accessed (e.g. creating a non-linear register
space).
In some cases, a single driver may need to access the same type of
device in multiple ways in a single system or architecture.
The goal of the
.Nm
functions is to allow a single driver source file to manipulate a set of
devices on different system architectures, and to allow a single driver
object file to manipulate a set of devices on multiple bus types on a
single architecture.
.Pp
Not all buses have to implement all functions described in this
document, though that is encouraged if the operations are logically
supported by the bus.
Unimplemented functions should cause compile-time errors if possible.
.Pp
All of the interface definitions described in this document are shown as
function prototypes and discussed as if they were required to be
functions.
Implementations are encouraged to implement prototyped (type-checked)
versions of these interfaces, but may implement them as macros if
appropriate.
Machine-dependent types, variables, and functions should be marked
clearly in
.Aq Pa machine/bus.h
to avoid confusion with the machine-independent types and functions,
and, if possible, should be given names which make the
machine-dependence clear.
.Sh CONCEPTS AND GUIDELINES
Bus spaces are described by bus space tags, which can be created only by
machine-dependent code.
A given machine may have several different types of bus space (e.g.\&
memory space and I/O space), and thus may provide multiple different bus
space tags.
Individual buses or devices on a machine may use more than one bus
space tag.
For instance, ISA devices are given an ISA memory space tag and an ISA
I/O space tag.
Architectures may have several different tags which represent the same
type of space, for instance because of multiple different host bus
interface chipsets.
.Pp
A range in bus space is described by a bus address and a bus size.
The bus address describes the start of the range in bus space.
The bus size describes the size of the range in bytes.
Buses which are not byte addressable may require use of bus space
ranges with appropriately aligned addresses and properly rounded sizes.
.Pp
Access to regions of bus space is facilitated by use of bus space
handles, which are usually created by mapping a specific range of a bus
space.
Handles may also be created by allocating and mapping a range of bus
space, the actual location of which is picked by the implementation
within bounds specified by the caller of the allocation function.
.Pp
All of the bus space access functions require one bus space tag
argument, at least one handle argument, and at least one offset argument
(a bus size).
The bus space tag specifies the space, each handle specifies a region in
the space, and each offset specifies the offset into the region of the
actual location(s) to be accessed.
Offsets are given in bytes, though buses may impose alignment constraints.
The offset used to access data relative to a given handle must be such
that all of the data being accessed is in the mapped region that the
handle describes.
Trying to access data outside that region is an error.
.Pp
Because some architectures' memory systems use buffering to improve
memory and device access performance, there is a mechanism which can be
used to create
.Dq barriers
in the bus space read and write stream.
There are three types of barriers: read, write, and read/write.
All reads started to the region before a read barrier must complete
before any reads after the read barrier are started.
The analogous requirement is true for write barriers.
Read/write barriers force all reads and writes started before the
barrier to complete before any reads or writes after the barrier are
started.
Correctly-written drivers will include all appropriate barriers, and
assume only the read/write ordering imposed by the barrier operations.
.Pp
People trying to write portable drivers with the
.Nm
functions should try to make minimal assumptions about what the system
allows.
In particular, they should expect that the system requires bus space
addresses being accessed to be naturally aligned (i.e. base address of
handle added to offset is a multiple of the access size), and that the
system does alignment checking on pointers (i.e. pointer to objects
being read and written must point to properly-aligned data).
.Pp
The descriptions of the
.Nm
functions given below all assume that they are called with proper
arguments.
If called with invalid arguments or arguments that are out of range
(e.g. trying to access data outside of the region mapped when a given
handle was created), undefined behaviour results.
In that case, they may cause the system to halt, either intentionally
(via panic) or unintentionally (by causing a fatal trap or by some other
means) or may cause improper operation which is not immediately fatal.
Functions which return void or which return data read from bus space
(i.e. functions which don't obviously return an error code) do not
fail.
They could only fail if given invalid arguments, and in that case their
behaviour is undefined.
Functions which take a count of bytes have undefined results if the
specified
.Fa count
is zero.
.Sh TYPES
Several types are defined in
.Aq Pa machine/bus.h
to facilitate use of the
.Nm
functions by drivers.
.Pp
.Bl -ohang -compact
.It Fa bus_addr_t
.Pp
The
.Fa bus_addr_t
type is used to describe bus addresses.
It must be an unsigned integral type capable of holding the largest bus
address usable by the architecture.
This type is primarily used when mapping and unmapping bus space.
.Pp
.It Fa bus_size_t
.Pp
The
.Fa bus_size_t
type is used to describe sizes of ranges in bus space.
It must be an unsigned integral type capable of holding the size of the
largest bus address range usable on the architecture.
This type is used by virtually all of the
.Nm
functions, describing sizes when mapping regions and offsets into
regions when performing space access operations.
.Pp
.It Fa bus_space_tag_t
.Pp
The
.Fa bus_space_tag_t
type is used to describe a particular bus space on a machine.
Its contents are machine-dependent and should be considered opaque by
machine-independent code.
This type is used by all
.Nm
functions to name the space on which they're operating.
.Pp
.It Fa bus_space_handle_t
.Pp
The
.Fa bus_space_handle_t
type is used to describe a mapping of a range of bus space.
Its contents are machine-dependent and should be considered opaque by
machine-independent code.
This type is used when performing bus space access operations.
.El
.Sh MAPPING AND UNMAPPING BUS SPACE
Bus space must be mapped before it can be used, and should be unmapped
when it is no longer needed.
The
.Fn bus_space_map
and
.Fn bus_space_unmap
functions provide these capabilities.
.Pp
Some drivers need to be able to pass a subregion of already-mapped bus
space to another driver or module within a driver.
The
.Fn bus_space_subregion
function allows such subregions to be created.
.Pp
.Bl -ohang -compact
.It Fn bus_space_map "space" "address" "size" "flags" "handlep"
.Pp
The
.Fn bus_space_map
function maps the region of bus space named by the
.Fa space ,
.Fa address ,
and
.Fa size
arguments.
If successful, it returns zero and fills in the bus space handle pointed
to by
.Fa handlep
with the handle that can be used to access the mapped region.
If unsuccessful, it will return non-zero and leave the bus space handle
pointed to by
.Fa handlep
in an undefined state.
.Pp
The
.Fa flags
argument controls how the space is to be mapped.
Supported flags include:
.Bl -tag -width BUS_SPACE_MAP_CACHEABLE -offset indent
.It Dv BUS_SPACE_MAP_CACHEABLE
Try to map the space so that access can be cached by the system cache.
If this flag is not specified, the implementation should map the space
so that it will not be cached.
This mapping method will only be useful in very rare occasions.
.Pp
This flag must have a value of 1 on all implementations for backward
compatibility.
.It Dv BUS_SPACE_MAP_PREFETCHABLE
Try to map the space so that accesses can be prefetched by the system,
and writes can be buffered.
This means, accesses should be side effect free (idempotent).
The
.Fn bus_space_barrier
methods will flush the write buffer or force actual read accesses.
If this flag is not specified, the
implementation should map the space so that it will not be prefetched
or delayed.
.It Dv BUS_SPACE_MAP_LINEAR
Try to map the space so that its contents can be accessed linearly via
normal memory access methods (e.g. pointer dereferencing and structure
accesses).
The
.Fn bus_space_vaddr
method can be used to obtain the kernel virtual address of the mapped range.
This is useful when software wants to do direct access to a memory
device, e.g. a frame buffer.
If this flag is specified and linear mapping is not possible, the
.Fn bus_space_map
call should fail.
If this
flag is not specified, the system may map the space in whatever way is
most convenient.
Use of this mapping method is not encouraged for normal device access;
where linear access is not essential, use of the
.Fn bus_space_read/write
methods is strongly recommended.
.El
.Pp
.\"Not all combinations of flags make sense or are supported with all
.\"spaces.
.\"For instance,
.Dv BUS_SPACE_MAP_CACHEABLE
may be meaningless when used on many systems' I/O port spaces.
and on some systems
.Dv BUS_SPACE_MAP_LINEAR
without
.Dv BUS_SPACE_MAP_PREFETCHABLE
may never work.
When the system hardware or firmware provides hints as to how spaces should be
mapped (e.g. the PCI memory mapping registers' "prefetchable" bit), those
hints should be followed for maximum compatibility.
On some systems,
requesting a mapping that cannot be satisfied (e.g. requesting a
non-prefetchable mapping when the system can only provide a prefetchable one)
will cause the request to fail.
.Pp
Some implementations may keep track of use of bus space for some or all
bus spaces and refuse to allow duplicate allocations.
This is encouraged for bus spaces which have no notion of slot-specific
space addressing, such as ISA and VME, and for spaces which coexist with
those spaces (e.g. EISA and PCI memory and I/O spaces coexisting with
ISA memory and I/O spaces).
.Pp
Mapped regions may contain areas for which there is no device on the bus.
If space in those areas is accessed, the results are bus-dependent.
.Pp
.It Fn bus_space_unmap "space" "handle" "size"
.Pp
The
.Fn bus_space_unmap
function unmaps a region of bus space mapped with
.Fn bus_space_map .
When unmapping a region, the
.Fa size
specified should be the same as the size given to
.Fn bus_space_map
when mapping that region.
.Pp
After
.Fn bus_space_unmap
is called on a handle, that handle is no longer valid.
If copies were made of the handle they are no longer valid, either.
.Pp
This function will never fail.
If it would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case,
.Fn bus_space_unmap
will never return.
.Pp
.It Fn bus_space_subregion "space" "handle" "offset" "size" "nhandlep"
.Pp
The
.Fn bus_space_subregion
function is a convenience function which makes a new handle to some
subregion of an already-mapped region of bus space.
The subregion described by the new handle starts at byte offset
.Fa offset
into the region described by
.Fa handle ,
with the size give by
.Fa size ,
and must be wholly contained within the original region.
.Pp
If successful,
.Fn bus_space_subregion
returns zero and fills in the bus space handle pointed to by
.Fa nhandlep .
If unsuccessful, it returns non-zero and leaves the bus space handle
pointed to by
.Fa nhandlep
in an undefined state.
In either case, the handle described by
.Fa handle
remains valid and is unmodified.
.Pp
When done with a handle created by
.Fn bus_space_subregion ,
the handle should be thrown away.
Under no circumstances should
.Fn bus_space_unmap
be used on the handle.
Doing so may confuse any resource management being done on the space,
and will result in undefined behaviour.
When
.Fn bus_space_unmap
or
.Fn bus_space_free
is called on a handle, all subregions of that handle become invalid.
.Pp
.It Fn bus_space_vaddr "tag" "handle"
.Pp
This method returns the kernel virtual address of a mapped bus space if and
only if it was mapped with the
.Dv BUS_SPACE_MAP_LINEAR
flag.
The range can be accessed by normal (volatile) pointer dereferences.
If mapped with the
.Dv BUS_SPACE_MAP_PREFETCHABLE
flag, the
.Fn bus_space_barrier
method must be used to force a particular access order.
.El
.Sh ALLOCATING AND FREEING BUS SPACE
Some devices require or allow bus space to be allocated by the operating
system for device use.
When the devices no longer need the space, the operating system should
free it for use by other devices.
The
.Fn bus_space_alloc
and
.Fn bus_space_free
functions provide these capabilities.
.Pp
.Bl -ohang -compact
.It Xo
.Fo bus_space_alloc
.Fa "space" "reg_start" "reg_end" "size"
.Fa "alignment" "boundary" "flags" "addrp" "handlep"
.Fc
.Xc
.Pp
The
.Fn bus_space_alloc
function allocates and maps a region of bus space with the size given by
.Fa size ,
corresponding to the given constraints.
If successful, it returns zero, fills in the bus address pointed to by
.Fa addrp
with the bus space address of the allocated region, and fills in the bus
space handle pointed to by
.Fa handlep
with the handle that can be used to access that region.
If unsuccessful, it returns non-zero and leaves the bus address pointed
to by
.Fa addrp
and the bus space handle pointed to by
.Fa handlep
in an undefined state.
.Pp
Constraints on the allocation are given by the
.Fa reg_start ,
.Fa reg_end ,
.Fa alignment ,
and
.Fa boundary
parameters.
The allocated region will start at or after
.Fa reg_start
and end before or at
.Fa reg_end .
The
.Fa alignment
constraint must be a power of two, and the allocated region will start
at an address that is an even multiple of that power of two.
The
.Fa boundary
constraint, if non-zero, ensures that the region is allocated so that
.Fa "first address in region"
/
.Fa boundary
has the same value as
.Fa "last address in region"
/
.Fa boundary .
If the constraints cannot be met,
.Fn bus_space_alloc
will fail.
It is an error to specify a set of constraints that can never be met
.Po
for example,
.Fa size
greater than
.Fa boundary
.Pc .
.Pp
The
.Fa flags
parameter is the same as the like-named parameter to
.Fa bus_space_map ,
the same flag values should be used, and they have the same meanings.
.Pp
Handles created by
.Fn bus_space_alloc
should only be freed with
.Fn bus_space_free .
Trying to use
.Fn bus_space_unmap
on them causes undefined behaviour.
The
.Fn bus_space_subregion
function can be used on handles created by
.Fn bus_space_alloc .
.Pp
.It Fn bus_space_free "space" "handle" "size"
.Pp
The
.Fn bus_space_free
function unmaps and frees a region of bus space mapped and allocated
with
.Fn bus_space_alloc .
When unmapping a region, the
.Fa size
specified should be the same as the size given to
.Fn bus_space_alloc
when allocating the region.
.Pp
After
.Fn bus_space_free
is called on a handle, that handle is no longer valid.
If copies were made of the handle, they are no longer valid, either.
.Pp
This function will never fail.
If it would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case,
.Fn bus_space_free
will never return.
.El
.Sh READING AND WRITING SINGLE DATA ITEMS
The simplest way to access bus space is to read or write a single data
item.
The
.Fn bus_space_read_N
and
.Fn bus_space_write_N
families of functions provide the ability to read and write 1, 2, 4, and
8 byte data items on buses which support those access sizes.
.Pp
.Bl -ohang -compact
.It Fn bus_space_read_1 "space" "handle" "offset"
.It Fn bus_space_read_2 "space" "handle" "offset"
.It Fn bus_space_read_4 "space" "handle" "offset"
.It Fn bus_space_read_8 "space" "handle" "offset"
.Pp
The
.Fn bus_space_read_N
family of functions reads a 1, 2, 4, or 8 byte data item from
the offset specified by
.Fa offset
into the region specified by
.Fa handle
of the bus space specified by
.Fa space .
The location being read must lie within the bus space region specified
by
.Fa handle .
.Pp
For portability, the starting address of the region specified by
.Fa handle
plus the offset should be a multiple of the size of data item being
read.
On some systems, not obeying this requirement may cause incorrect data
to be read, on others it may cause a system crash.
.Pp
Read operations done by the
.Fn bus_space_read_N
functions may be executed out of order with respect to other pending
read and write operations unless order is enforced by use of the
.Fn bus_space_barrier
function.
.Pp
These functions will never fail.
If they would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case, they will never return.
.Pp
.It Fn bus_space_write_1 "space" "handle" "offset" "value"
.It Fn bus_space_write_2 "space" "handle" "offset" "value"
.It Fn bus_space_write_4 "space" "handle" "offset" "value"
.It Fn bus_space_write_8 "space" "handle" "offset" "value"
.Pp
The
.Fn bus_space_write_N
family of functions writes a 1, 2, 4, or 8 byte data item to the offset
specified by
.Fa offset
into the region specified by
.Fa handle
of the bus space specified by
.Fa space .
The location being written must lie within the bus space region
specified by
.Fa handle .
.Pp
For portability, the starting address of the region specified by
.Fa handle
plus the offset should be a multiple of the size of data item being
written.
On some systems, not obeying this requirement may cause incorrect data
to be written, on others it may cause a system crash.
.Pp
Write operations done by the
.Fn bus_space_write_N
functions may be executed out of order with respect to other pending
read and write operations unless order is enforced by use of the
.Fn bus_space_barrier
function.
.Pp
These functions will never fail.
If they would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case, they will never return.
.El
.Sh BARRIERS
In order to allow high-performance buffering implementations to avoid
bus activity on every operation, read and write ordering should be
specified explicitly by drivers when necessary.
The
.Fn bus_space_barrier
function provides that ability.
.Pp
.Bl -ohang -compact
.It Fn bus_space_barrier "space" "handle" "offset" "length" "flags"
.Pp
The
.Fn bus_space_barrier
function enforces ordering of bus space read and write operations for
the specified subregion (described by the
.Fa offset
and
.Fa length
parameters) of the region named by
.Fa handle
in the space named by
.Fa space .
.Pp
The
.Fa flags
argument controls what types of operations are to be ordered.
Supported flags are:
.Bl -tag -width BUS_SPACE_BARRIER_WRITE -offset indent
.It Dv BUS_SPACE_BARRIER_READ
Synchronize read operations.
.It Dv BUS_SPACE_BARRIER_WRITE
Synchronize write operations.
.El
.Pp
Those flags can be combined (or-ed together) to enforce ordering on both
read and write operations.
.Pp
All of the specified type(s) of operation which are done to the region
before the barrier operation are guaranteed to complete before any of
the specified type(s) of operation done after the barrier.
.Pp
Example: Consider a hypothetical device with two single-byte ports, one
write-only input port (at offset 0) and a read-only output port (at
offset 1).
Operation of the device is as follows: data bytes are written to the
input port, and are placed by the device on a stack, the top of which is
read by reading from the output port.
The sequence to correctly write two data bytes to the device then read
those two data bytes back would be:
.Bd -literal
/*
 * t and h are the tag and handle for the mapped device's
 * space.
 */
bus_space_write_1(t, h, 0, data0);
bus_space_barrier(t, h, 0, 1, BUS_SPACE_BARRIER_WRITE);  /* 1 */
bus_space_write_1(t, h, 0, data1);
bus_space_barrier(t, h, 0, 2,
    BUS_SPACE_BARRIER_READ|BUS_SPACE_BARRIER_WRITE);     /* 2 */
ndata1 = bus_space_read_1(t, h, 1);
bus_space_barrier(t, h, 1, 1, BUS_SPACE_BARRIER_READ);   /* 3 */
ndata0 = bus_space_read_1(t, h, 1);
/* data0 == ndata0, data1 == ndata1 */
.Ed
.Pp
The first barrier makes sure that the first write finishes before the
second write is issued, so that two writes to the input port are done in
order and are not collapsed into a single write.
This ensures that the data bytes are written to the device correctly and
in order.
.Pp
The second barrier makes sure that the writes to the output port finish
before any of the reads to the input port are issued, thereby making
sure that all of the writes are finished before data is read.
This ensures that the first byte read from the device really is the last
one that was written.
.Pp
The third barrier makes sure that the first read finishes before the
second read is issued, ensuring that data is read correctly and in order.
.Pp
The barriers in the example above are specified to cover the absolute
minimum number of bus space locations.
It is correct (and often easier) to make barrier operations cover the
device's whole range of bus space, that is, to specify an offset of zero
and the size of the whole region.
.El
.Sh REGION OPERATIONS
Some devices use buffers which are mapped as regions in bus space.
Often, drivers want to copy the contents of those buffers to or from
memory, e.g. into mbufs which can be passed to higher levels of the
system or from mbufs to be output to a network.
In order to allow drivers to do this as efficiently as possible, the
.Fn bus_space_read_region_N
and
.Fn bus_space_write_region_N
families of functions are provided.
.Pp
Drivers occasionally need to copy one region of a bus space to another,
or to set all locations in a region of bus space to contain a single
value.
The
.Fn bus_space_copy_N
family of functions and the
.Fn bus_space_set_region_N
family of functions allow drivers to perform these operations.
.Pp
.Bl -ohang -compact
.It Fn bus_space_read_region_1 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_read_region_2 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_read_region_4 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_read_region_8 "space" "handle" "offset" "datap" "count"
.Pp
The
.Fn bus_space_read_region_N
family of functions reads
.Fa count
1, 2, 4, or 8 byte data items from bus space starting at byte offset
.Fa offset
in the region specified by
.Fa handle
of the bus space specified by
.Fa space
and writes them into the array specified by
.Fa datap .
Each successive data item is read from an offset
1, 2, 4, or 8 bytes after the previous data item (depending on which
function is used).
All locations being read must lie within the bus space region specified
by
.Fa handle .
.Pp
For portability, the starting address of the region specified by
.Fa handle
plus the offset should be a multiple of the size of data items being
read and the data array pointer should be properly aligned.
On some systems, not obeying these requirements may cause incorrect data
to be read, on others it may cause a system crash.
.Pp
Read operations done by the
.Fn bus_space_read_region_N
functions may be executed in any order.
They may also be executed out of order with respect to other pending
read and write operations unless order is enforced by use of the
.Fn bus_space_barrier
function.
There is no way to insert barriers between reads of individual bus space
locations executed by the
.Fn bus_space_read_region_N
functions.
.Pp
These functions will never fail.
If they would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case, they will never return.
.Pp
.It Fn bus_space_write_region_1 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_write_region_2 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_write_region_4 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_write_region_8 "space" "handle" "offset" "datap" "count"
.Pp
The
.Fn bus_space_write_region_N
family of functions reads
.Fa count
1, 2, 4, or 8 byte data items from the array specified by
.Fa datap
and writes them to bus space starting at byte offset
.Fa offset
in the region specified by
.Fa handle
of the bus space specified by
.Fa space .
Each successive data item is written to an offset 1, 2, 4, or 8 bytes
after the previous data item (depending on which function is used).
All locations being written must lie within the bus space region
specified by
.Fa handle .
.Pp
For portability, the starting address of the region specified by
.Fa handle
plus the offset should be a multiple of the size of data items being
written and the data array pointer should be properly aligned.
On some systems, not obeying these requirements may cause incorrect data
to be written, on others it may cause a system crash.
.Pp
Write operations done by the
.Fn bus_space_write_region_N
functions may be executed in any order.
They may also be executed out of order with respect to other pending
read and write operations unless order is enforced by use of the
.Fn bus_space_barrier
function.
There is no way to insert barriers between writes of individual bus
space locations executed by the
.Fn bus_space_write_region_N
functions.
.Pp
These functions will never fail.
If they would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case, they will never return.
.Pp
.It Fn bus_space_copy_1 "space" "srchandle" "srcoffset" "dsthandle" \
"dstoffset" "count"
.It Fn bus_space_copy_2 "space" "srchandle" "srcoffset" "dsthandle" \
"dstoffset" "count"
.It Fn bus_space_copy_4 "space" "srchandle" "srcoffset" "dsthandle" \
"dstoffset" "count"
.It Fn bus_space_copy_8 "space" "srchandle" "srcoffset" "dsthandle" \
"dstoffset" "count"
.Pp
The
.Fn bus_space_copy_N
family of functions copies
.Fa count
1, 2, 4, or 8 byte data items in bus space from the area starting at
byte offset
.Fa srcoffset
in the region specified by
.Fa srchandle
of the bus space specified by
.Fa space
to the area starting at byte offset
.Fa dstoffset
in the region specified by
.Fa dsthandle
in the same bus space.
Each successive data item read or written has an offset 1, 2, 4, or 8
bytes after the previous data item (depending on which function is
used).
All locations being read and written must lie within the bus space
region specified by their respective handles.
.Pp
For portability, the starting addresses of the regions specified by
each handle plus its respective offset should be a multiple of the size
of data items being copied.
On some systems, not obeying this requirement may cause incorrect data
to be copied, on others it may cause a system crash.
.Pp
Read and write operations done by the
.Fn bus_space_copy_N
functions may be executed in any order.
They may also be executed out of order with respect to other pending
read and write operations unless order is enforced by use of the
.Fn bus_space_barrier function .
There is no way to insert barriers between reads or writes of individual
bus space locations executed by the
.Fn bus_space_copy_N
functions.
.Pp
Overlapping copies between different subregions of a single region of
bus space are handled correctly by the
.Fn bus_space_copy_N
functions.
.Pp
These functions will never fail.
If they would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case, they will never return.
.Pp
.It Fn bus_space_set_region_1 "space" "handle" "offset" "value" "count"
.It Fn bus_space_set_region_2 "space" "handle" "offset" "value" "count"
.It Fn bus_space_set_region_4 "space" "handle" "offset" "value" "count"
.It Fn bus_space_set_region_8 "space" "handle" "offset" "value" "count"
.Pp
The
.Fn bus_space_set_region_N
family of functions writes the given
.Fa value
to
.Fa count
1, 2, 4, or 8 byte data items in bus space starting at byte offset
.Fa offset
in the region specified by
.Fa handle
of the bus space specified by
.Fa space .
Each successive data item has an offset 1, 2, 4, or 8 bytes after the
previous data item (depending on which function is used).
All locations being written must lie within the bus space region
specified by
.Fa handle .
.Pp
For portability, the starting address of the region specified by
.Fa handle
plus the offset should be a multiple of the size of data items being
written.
On some systems, not obeying this requirement may cause incorrect data
to be written, on others it may cause a system crash.
.Pp
Write operations done by the
.Fn bus_space_set_region_N
functions may be executed in any order.
They may also be executed out of order with respect to other pending
read and write operations unless order is enforced by use of the
.Fn bus_space_barrier
function.
There is no way to insert barriers between writes of individual bus
space locations executed by the
.Fn bus_space_set_region_N
functions.
.Pp
These functions will never fail.
If they would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case, they will never return.
.El
.Sh READING AND WRITING A SINGLE LOCATION MULTIPLE TIMES
Some devices implement single locations in bus space which are to be
read or written multiple times to communicate data, e.g. some Ethernet
devices' packet buffer FIFOs.
In order to allow drivers to manipulate these types of devices as
efficiently as possible, the
.Fn bus_space_read_multi_N ,
.Fn bus_space_write_multi_N ,
and
.Fn bus_space_set_multi_N
families of functions are provided.
.Pp
.Bl -ohang -compact
.It Fn bus_space_read_multi_1 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_read_multi_2 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_read_multi_4 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_read_multi_8 "space" "handle" "offset" "datap" "count"
.Pp
The
.Fn bus_space_read_multi_N
family of functions reads
.Fa count
1, 2, 4, or 8 byte data items from bus space at byte offset
.Fa offset
in the region specified by
.Fa handle
of the bus space specified by
.Fa space
and writes them into the array specified by
.Fa datap .
Each successive data item is read from the same location in bus space.
The location being read must lie within the bus space region specified
by
.Fa handle .
.Pp
For portability, the starting address of the region specified by
.Fa handle
plus the offset should be a multiple of the size of data items being
read and the data array pointer should be properly aligned.
On some systems, not obeying these requirements may cause incorrect data
to be read, on others it may cause a system crash.
.Pp
Read operations done by the
.Fn bus_space_read_multi_N
functions may be executed out of order with respect to other pending
read and write operations unless order is enforced by use of the
.Fn bus_space_barrier
function.
Because the
.Fn bus_space_read_multi_N
functions read the same bus space location multiple times, they place an
implicit read barrier between each successive read of that bus space
location.
.Pp
These functions will never fail.
If they would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case, they will never return.
.Pp
.It Fn bus_space_write_multi_1 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_write_multi_2 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_write_multi_4 "space" "handle" "offset" "datap" "count"
.It Fn bus_space_write_multi_8 "space" "handle" "offset" "datap" "count"
.Pp
The
.Fn bus_space_write_multi_N
family of functions reads
.Fa count
1, 2, 4, or 8 byte data items from the array specified by
.Fa datap
and writes them into bus space at byte offset
.Fa offset
in the region specified by
.Fa handle
of the bus space specified by
.Fa space .
Each successive data item is written to the same location in bus space.
The location being written must lie within the bus space region
specified by
.Fa handle .
.Pp
For portability, the starting address of the region specified by
.Fa handle
plus the offset should be a multiple of the size of data items being
written and the data array pointer should be properly aligned.
On some systems, not obeying these requirements may cause incorrect data
to be written, on others it may cause a system crash.
.Pp
Write operations done by the
.Fn bus_space_write_multi_N
functions may be executed out of order with respect to other pending
read and write operations unless order is enforced by use of the
.Fn bus_space_barrier
function.
Because the
.Fn bus_space_write_multi_N
functions write the same bus space location multiple times, they place
an implicit write barrier between each successive write of that bus
space location.
.Pp
These functions will never fail.
If they would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case, they will never return.
.Pp
.It Fn bus_space_set_multi_1 "space" "handle" "offset" "value" "count"
.It Fn bus_space_set_multi_2 "space" "handle" "offset" "value" "count"
.It Fn bus_space_set_multi_4 "space" "handle" "offset" "value" "count"
.It Fn bus_space_set_multi_8 "space" "handle" "offset" "value" "count"
.Pp
The
.Fn bus_space_set_multi_N
family of functions writes the 1, 2, 4, or 8 byte value
.Fa value
into bus space
.Fa count
times at byte offset
.Fa offset
in the region specified by
.Fa handle
of the bus space specified by
.Fa space .
The location being written must lie within the bus space region
specified by
.Fa handle .
.Pp
For portability, the address specified by
.Fa handle
plus the offset should be a multiple of the size of the data value being
written.
On some systems, not obeying these requirements may cause
incorrect data to be written, on others it may cause a system crash.
.Pp
Write operations done by the
.Fn bus_space_set_multi_N
functions may be executed out of order with respect to other pending
read and write operations unless order is enforced by use of the
.Fn bus_space_barrier
function.
Because the
.Fn bus_space_set_multi_N
functions write the same bus space location multiple times, they place
an implicit write barrier between each successive write of that bus
space location.
.Pp
These functions will never fail.
If they would fail (e.g. because of an argument error), that indicates a
software bug which should cause a panic.
In that case, they will never return.
.El
.Sh RAW FUNCTIONS
Most of the
.Nm
functions imply a host byte-order and a bus byte-order and take care of
any translation for the caller.
In some cases, however, hardware may map a FIFO or some other memory
region for which the caller may want to use multi-word, yet untranslated
access.
Access to these types of memory regions should be with the
.Fn bus_space_*_raw_*_N
functions.
.Pp
.Bl -ohang -compact
.\".It Fn bus_space_read_stream_1 "space" "handle" "offset"
.\".It Fn bus_space_read_stream_2 "space" "handle" "offset"
.\".It Fn bus_space_read_stream_4 "space" "handle" "offset"
.\".It Fn bus_space_read_stream_8 "space" "handle" "offset"
.\".It Fn bus_space_read_raw_multi_1 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_read_raw_multi_2 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_read_raw_multi_4 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_read_raw_multi_8 "space" "handle" "offset" "datap" "size"
.\".It Fn bus_space_read_raw_region_1 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_read_raw_region_2 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_read_raw_region_4 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_read_raw_region_8 "space" "handle" "offset" "datap" "size"
.\".It Fn bus_space_write_stream_1 "space" "handle" "offset" "value"
.\".It Fn bus_space_write_stream_2 "space" "handle" "offset" "value"
.\".It Fn bus_space_write_stream_4 "space" "handle" "offset" "value"
.\".It Fn bus_space_write_stream_8 "space" "handle" "offset" "value"
.\".It Fn bus_space_write_raw_multi_1 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_write_raw_multi_2 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_write_raw_multi_4 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_write_raw_multi_8 "space" "handle" "offset" "datap" "size"
.\".It Fn bus_space_write_raw_region_1 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_write_raw_region_2 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_write_raw_region_4 "space" "handle" "offset" "datap" "size"
.It Fn bus_space_write_raw_region_8 "space" "handle" "offset" "datap" "size"
.El
.Pp
These functions, unlike their non-raw counterparts, all take a
u_int8_t pointer for the
.Fa datap
argument and a byte count for the
.Fa size
argument regardless of the access width being requested.
.Pp
.Fa datap
must reference a buffer that is correctly aligned for the
access width being requested or the results are undefined.
.Pp
.Fa size
must be a multiple of the access width or the results are undefined.
.Pp
In all other respects these functions are the same as their non-raw
counterparts.
Please consult the documentation for those functions for further
information.
.Sh EXPECTED CHANGES TO THE BUS_SPACE FUNCTIONS
The definition of the
.Nm
functions should not yet be considered finalized.
There are several changes and improvements which should be explored,
including:
.Bl -bullet
.It
Providing a mechanism by which incorrectly-written drivers will be
automatically given barriers and properly-written drivers won't be
forced to use more barriers than they need.
This should probably be done via a
.Li #define
in the incorrectly-written drivers.
Unfortunately, at this time, few drivers actually use barriers correctly
(or at all).
Because of that,
.Nm
implementations on architectures which do buffering must always do the
barriers inside the
.Nm
calls, to be safe.
That has a potentially significant performance impact.
.It
Exporting the
.Nm
functions to user-land so that applications (such as X servers) have
easier, more portable access to device space.
.It
Redefining bus space tags and handles so that machine-independent bus
interface drivers (for example PCI to VME bridges) could define and
implement bus spaces without requiring machine-dependent code.
If this is done, it should be done in such a way that machine-dependent
optimizations should remain possible.
.It
Converting bus spaces (such as PCI configuration space) which currently
use space-specific access methods to use the
.Nm
functions where that is appropriate.
.It
Redefining the way bus space is mapped and allocated, so that mapping
and allocation are done with bus specific functions which return bus
space tags.
This would allow further optimization than is currently possible, and
would also ease translation of the
.Nm
functions into user space (since mapping in user space would look like
it just used a different bus-specific mapping function).
.El
.Sh SEE ALSO
.Xr bus_dma 9
.\".Sh STANDARDS
.\".Pp
.\"The current version of the
.\".Nm
.\"interface specification differs slightly from the original
.\"specification that came into wide use.
.\"A few of the function names and arguments have changed
.\"for consistency and increased functionality.
.\"Drivers that were written to the
.\"old, deprecated specification can be compiled by defining the
.\".Dv __BUS_SPACE_COMPAT_OLDDEFS
.\"preprocessor symbol before including
.\".Pa Aq machine/bus.h .
.Sh HISTORY
The
.Nm
functions were introduced in a different form (memory and I/O spaces
were accessed via different sets of functions) in
.Nx 1.2 .
The functions were merged to work on generic
.Dq spaces
early in the
.Nx 1.3
development cycle, and many drivers were converted to use them.
This document was written later during the
.Nx 1.3
development cycle and the specification was updated to fix some
consistency problems and to add some missing functionality.
.Pp
The
.Ox
development team added the *_raw_* API, and discarded the *_stream_*
API.
.Sh AUTHORS
The
.Nm
interfaces were designed and implemented by the
.Nx
developer community.
Primary contributors and implementors were Chris Demetriou, Jason
Thorpe, and Charles Hannum, but the rest of the
.Nx
developers and the user community played a significant role in
development.
.Pp
Chris Demetriou wrote this manual page.
.Pp
Niklas Hallqvist did the *_raw_* API for
.Ox .