BUS_SPACE(9) NetBSD Kernel Developer's Manual BUS_SPACE(9)
NAME
bus_space, bus_space_barrier, bus_space_copy_region_1, bus_space_copy_region_2, bus_space_copy_region_4, bus_space_copy_region_8, bus_space_free, bus_space_map, bus_space_peek_1, bus_space_peek_2, bus_space_peek_4, bus_space_peek_8, bus_space_poke_1, bus_space_poke_2, bus_space_poke_4, bus_space_poke_8, bus_space_read_1, bus_space_read_2, bus_space_read_4, bus_space_read_8, bus_space_read_multi_1, bus_space_read_multi_2, bus_space_read_multi_4, bus_space_read_multi_8, bus_space_read_multi_stream_1, bus_space_read_multi_stream_2, bus_space_read_multi_stream_4, bus_space_read_multi_stream_8, bus_space_read_region_1, bus_space_read_region_2, bus_space_read_region_4, bus_space_read_region_8, bus_space_read_region_stream_1, bus_space_read_region_stream_2, bus_space_read_region_stream_4, bus_space_read_region_stream_8, bus_space_read_stream_1, bus_space_read_stream_2, bus_space_read_stream_4, bus_space_read_stream_8, bus_space_set_region_1, bus_space_set_region_2, bus_space_set_region_4, bus_space_set_region_8, bus_space_subregion, bus_space_unmap, bus_space_vaddr, bus_space_mmap, bus_space_write_1, bus_space_write_2, bus_space_write_4, bus_space_write_8, bus_space_write_multi_1, bus_space_write_multi_2, bus_space_write_multi_4, bus_space_write_multi_8, bus_space_write_multi_stream_1, bus_space_write_multi_stream_2, bus_space_write_multi_stream_4, bus_space_write_multi_stream_8, bus_space_write_region_1, bus_space_write_region_2, bus_space_write_region_4, bus_space_write_region_8, bus_space_write_region_stream_1, bus_space_write_region_stream_2, bus_space_write_region_stream_4, bus_space_write_region_stream_8, bus_space_write_stream_1, bus_space_write_stream_2, bus_space_write_stream_4, bus_space_write_stream_8 -- bus space manipula- tion functions
SYNOPSIS
#include <machine/bus.h> int bus_space_map(bus_space_tag_t space, bus_addr_t address, bus_size_t size, int flags, bus_space_handle_t *handlep); void bus_space_unmap(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t size); int 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); int bus_space_alloc(bus_space_tag_t space, bus_addr_t reg_start, bus_addr_t reg_end, bus_size_t size, bus_size_t alignment, bus_size_t boundary, int flags, bus_addr_t *addrp, bus_space_handle_t *handlep); void bus_space_free(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t size); void * bus_space_vaddr(bus_space_tag_t space, bus_space_handle_t handle); paddr_t bus_space_mmap(bus_space_tag_t space, bus_addr_t addr, off_t off, int prot, int flags); int bus_space_peek_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint8_t *datap); int bus_space_peek_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint16_t *datap); int bus_space_peek_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint32_t *datap); int bus_space_peek_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint64_t *datap); int bus_space_poke_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint8_t data); int bus_space_poke_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint16_t data); int bus_space_poke_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint32_t data); int bus_space_poke_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint64_t data); uint8_t bus_space_read_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset); uint16_t bus_space_read_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset); uint32_t bus_space_read_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset); uint64_t bus_space_read_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset); void bus_space_write_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint8_t value); void bus_space_write_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint16_t value); void bus_space_write_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint32_t value); void bus_space_write_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint64_t value); void bus_space_barrier(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, bus_size_t length, int flags); void bus_space_read_region_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint8_t *datap, bus_size_t count); void bus_space_read_region_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint16_t *datap, bus_size_t count); void bus_space_read_region_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint32_t *datap, bus_size_t count); void bus_space_read_region_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint64_t *datap, bus_size_t count); void bus_space_read_region_stream_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint8_t *datap, bus_size_t count); void bus_space_read_region_stream_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint16_t *datap, bus_size_t count); void bus_space_read_region_stream_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint32_t *datap, bus_size_t count); void bus_space_read_region_stream_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint64_t *datap, bus_size_t count); void bus_space_write_region_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint8_t *datap, bus_size_t count); void bus_space_write_region_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint16_t *datap, bus_size_t count); void bus_space_write_region_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint32_t *datap, bus_size_t count); void bus_space_write_region_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint64_t *datap, bus_size_t count); void bus_space_write_region_stream_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint8_t *datap, bus_size_t count); void bus_space_write_region_stream_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint16_t *datap, bus_size_t count); void bus_space_write_region_stream_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint32_t *datap, bus_size_t count); void bus_space_write_region_stream_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint64_t *datap, bus_size_t count); void bus_space_copy_region_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); void bus_space_copy_region_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); void bus_space_copy_region_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); void bus_space_copy_region_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); void bus_space_set_region_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint8_t value, bus_size_t count); void bus_space_set_region_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint16_t value, bus_size_t count); void bus_space_set_region_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint32_t value, bus_size_t count); void bus_space_set_region_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint64_t value, bus_size_t count); void bus_space_read_multi_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint8_t *datap, bus_size_t count); void bus_space_read_multi_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint16_t *datap, bus_size_t count); void bus_space_read_multi_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint32_t *datap, bus_size_t count); void bus_space_read_multi_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint64_t *datap, bus_size_t count); void bus_space_read_multi_stream_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint8_t *datap, bus_size_t count); void bus_space_read_multi_stream_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint16_t *datap, bus_size_t count); void bus_space_read_multi_stream_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint32_t *datap, bus_size_t count); void bus_space_read_multi_stream_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, uint64_t *datap, bus_size_t count); void bus_space_write_multi_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint8_t *datap, bus_size_t count); void bus_space_write_multi_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint16_t *datap, bus_size_t count); void bus_space_write_multi_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint32_t *datap, bus_size_t count); void bus_space_write_multi_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint64_t *datap, bus_size_t count); void bus_space_write_multi_stream_1(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint8_t *datap, bus_size_t count); void bus_space_write_multi_stream_2(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint16_t *datap, bus_size_t count); void bus_space_write_multi_stream_4(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint32_t *datap, bus_size_t count); void bus_space_write_multi_stream_8(bus_space_tag_t space, bus_space_handle_t handle, bus_size_t offset, const uint64_t *datap, bus_size_t count);
DESCRIPTION
The bus_space 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 <machine/bus.h> header file. 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 mem- ory space on a second system. On a third system, architectural limita- tions 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 bus_space 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. Not all busses have to implement all functions described in this docu- ment, though that is encouraged if the operations are logically supported by the bus. Unimplemented functions should cause compile-time errors if possible. All of the interface definitions described in this document are shown as function prototypes and discussed as if they were required to be func- tions. 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 <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.
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 busses 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. Architec- tures may have several different tags which represent the same type of space, for instance because of multiple different host bus interface chipsets. 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. Busses which are not byte addressable may require use of bus space ranges with appropriately aligned addresses and properly rounded sizes. Access to regions of bus space is facilitated by use of bus space han- dles, 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. 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 busses 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. Because some architectures' memory systems use buffering to improve mem- ory and device access performance, there is a mechanism which can be used to create ``barriers'' in the bus space read and write stream. There are two types of barriers: ordering barriers and completion barri- ers. Ordering barriers prevent some operations from bypassing other opera- tions. They are relatively light weight and described in terms of the operations they are intended to order. The important thing to note is that they create specific ordering constraint surrounding bus accesses but do not necessarily force any synchronization themselves. So, if there is enough distance between the memory operations being ordered, the preceding ones could complete by themselves resulting in no performance penalty. For instance, a write before read barrier will force any writes issued before the barrier instruction to complete before any reads after the barrier are issued. This forces processors with write buffers to read data from memory rather than from the pending write in the write buffer. Ordering barriers are usually sufficient for most circumstances, and can be combined together. For instance a read before write barrier can be combined with a write before write barrier to force all memory operations to complete before the next write is started. Completion barriers force all memory operations and any pending excep- tions to be completed before any instructions after the barrier may be issued. Completion barriers are extremely expensive and almost never required in device driver code. A single completion barrier can force the processor to stall on memory for hundreds of cycles on some machines. Correctly-written drivers will include all appropriate barriers, and assume only the read/write ordering imposed by the barrier operations. People trying to write portable drivers with the bus_space 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). The descriptions of the bus_space 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 behav- iour 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 immedi- ately 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 count is zero.
TYPES
Several types are defined in <machine/bus.h> to facilitate use of the bus_space functions by drivers. bus_addr_t The 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 unmap- ping bus space. bus_size_t The 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 bus_space functions, describing sizes when map- ping regions and offsets into regions when performing space access opera- tions. bus_space_tag_t The 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 bus_space functions to name the space on which they're operating. bus_space_handle_t The 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.
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 bus_space_map() and bus_space_unmap() functions provide these capabilities. Some drivers need to be able to pass a subregion of already-mapped bus space to another driver or module within a driver. The bus_space_subregion() function allows such subregions to be created. bus_space_map(space, address, size, flags, handlep) The bus_space_map() function maps the region of bus space named by the space, address, and size arguments. If successful, it returns zero and fills in the bus space handle pointed to by 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 handlep in an unde- fined state. The flags argument controls how the space is to be mapped. Supported flags include: BUS_SPACE_MAP_CACHEABLE Try to map the space so that accesses 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. This flag must have a value of 1 on all implementations for backward compatibil- ity. 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 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. 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 dereferenc- ing and structure accesses). The bus_space_vaddr() method can be used to obtain the kernel virtual address of the mapped range. This is useful when soft- ware 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 bus_space_map() call should fail. If this flag is not speci- fied, 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 bus_space_read/write() methods is strongly recommended. Not all combinations of flags make sense or are supported with all spa- ces. For instance, BUS_SPACE_MAP_CACHEABLE may be meaningless when used on many systems' I/O port spaces, and on some systems BUS_SPACE_MAP_LINEAR without 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 prefetch- able one) will cause the request to fail. 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 co-existing with ISA memory and I/O spaces). 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. bus_space_unmap(space, handle, size) The bus_space_unmap() function unmaps a region of bus space mapped with bus_space_map(). When unmapping a region, the size specified should be the same as the size given to bus_space_map() when mapping that region. After 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.) 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, bus_space_unmap() will never return. bus_space_subregion(space, handle, offset, size, nhandlep) The 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 offset into the region described by handle, with the size given by size, and must be wholly contained within the original region. If successful, bus_space_subregion() returns zero and fills in the bus space handle pointed to by nhandlep. If unsuccessful, it returns non- zero and leaves the bus space handle pointed to by nhandlep in an unde- fined state. In either case, the handle described by handle remains valid and is unmodified. When done with a handle created by bus_space_subregion(), the handle should be thrown away. Under no circumstances should 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 bus_space_unmap() or bus_space_free() is called on a handle, all subre- gions of that handle become invalid. bus_space_vaddr(tag, handle) This method returns the kernel virtual address of a mapped bus space if and only if it was mapped with the BUS_SPACE_MAP_LINEAR flag. The range can be accessed by normal (volatile) pointer dereferences. If mapped with the BUS_SPACE_MAP_PREFETCHABLE flag, the bus_space_barrier() method must be used to force a particular access order. bus_space_mmap(tag, addr, off, prot, flags) This method is used to provide support for memory mapping bus space into user applications. If an address space is addressable via volatile pointer dereferences, bus_space_mmap() will return the physical address (possibly encoded as a machine-dependent cookie) of the bus space indi- cated by addr and off. addr is the base address of the device or device region, and off is the offset into that region that is being requested. If the request is made with BUS_SPACE_MAP_LINEAR as a flag, then a linear region must be returned to the caller. If the region cannot be mapped (either the address does not exist, or the constraints can not be met), bus_space_mmap() returns -1 to indicate failure. Note that it is not necessary that the region being requested by a bus_space_mmap() call be mapped into a bus_space_handle_t. bus_space_mmap() is called once per PAGE_SIZE page in the range. The prot argument indicates the memory protection requested by the user application for the range.
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 bus_space_alloc() and bus_space_free() functions provide these capabili- ties. bus_space_alloc(space, reg_start, reg_end, size, alignment, boundary, flags, addrp, handlep) The bus_space_alloc() function allocates and maps a region of bus space with the size given by size, corresponding to the given constraints. If successful, it returns zero, fills in the bus address pointed to by addrp with the bus space address of the allocated region, and fills in the bus space handle pointed to by 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 addrp and the bus space handle pointed to by handlep in an undefined state. Constraints on the allocation are given by the reg_start, reg_end, alignment, and boundary parameters. The allocated region will start at or after reg_start and end before or at reg_end. The alignment con- straint 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 boundary con- straint, if non-zero, ensures that the region is allocated so that first address in region / boundary has the same value as last address in region / boundary. If the constraints cannot be met, bus_space_alloc() will fail. It is an error to specify a set of constraints that can never be met (for example, size greater than boundary). The flags parameter is the same as the like-named parameter to bus_space_map, the same flag values should be used, and they have the same meanings. Handles created by bus_space_alloc() should only be freed with bus_space_free(). Trying to use bus_space_unmap() on them causes unde- fined behaviour. The bus_space_subregion() function can be used on han- dles created by bus_space_alloc(). bus_space_free(space, handle, size) The bus_space_free() function unmaps and frees a region of bus space mapped and allocated with bus_space_alloc(). When unmapping a region, the size specified should be the same as the size given to bus_space_alloc() when allocating the region. After 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.) 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, bus_space_free() will never return.
READING AND WRITING SINGLE DATA ITEMS
The simplest way to access bus space is to read or write a single data item. The bus_space_read_N() and bus_space_write_N() families of func- tions provide the ability to read and write 1, 2, 4, and 8 byte data items on busses which support those access sizes. bus_space_read_1(space, handle, offset) bus_space_read_2(space, handle, offset) bus_space_read_4(space, handle, offset) bus_space_read_8(space, handle, offset) The bus_space_read_N() family of functions reads a 1, 2, 4, or 8 byte data item from the offset specified by offset into the region specified by handle of the bus space specified by space. The location being read must lie within the bus space region specified by handle. For portability, the starting address of the region specified by 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. Read operations done by the 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 bus_space_barrier() function. 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. bus_space_write_1(space, handle, offset, value) bus_space_write_2(space, handle, offset, value) bus_space_write_4(space, handle, offset, value) bus_space_write_8(space, handle, offset, value) The bus_space_write_N() family of functions writes a 1, 2, 4, or 8 byte data item to the offset specified by offset into the region specified by handle of the bus space specified by space. The location being written must lie within the bus space region specified by handle. For portability, the starting address of the region specified by handle plus the offset should be a multiple of the size of data item being writ- ten. On some systems, not obeying this requirement may cause incorrect data to be written, on others it may cause a system crash. Write operations done by the bus_space_write_N() functions may be exe- cuted out of order with respect to other pending read and write opera- tions unless order is enforced by use of the bus_space_barrier() func- tion. 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.
PROBING BUS SPACE FOR HARDWARE WHICH MAY NOT RESPOND
One problem with the bus_space_read_N() and bus_space_write_N() family of functions is that they provide no protection against exceptions which can occur when no physical hardware or device responds to the read or write cycles. In such a situation, the system typically would panic due to a kernel-mode bus error. The bus_space_peek_N() and bus_space_poke_N() family of functions provide a mechanism to handle these exceptions grace- fully without the risk of crashing the system. As with bus_space_read_N() and bus_space_write_N(), the peek and poke functions provide the ability to read and write 1, 2, 4, and 8 byte data items on busses which support those access sizes. All of the constraints specified in the descriptions of the bus_space_read_N() and bus_space_write_N() functions also apply to bus_space_peek_N() and bus_space_poke_N(). In addition, explicit calls to the bus_space_barrier() function are not required as the implementation will ensure all pending operations com- plete before the peek or poke operation starts. The implementation will also ensure that the peek or poke operations complete before returning. The return value indicates the outcome of the peek or poke operation. A return value of zero implies that a hardware device is responding to the operation at the specified offset in the bus space. A non-zero return value indicates that the kernel intercepted a hardware exception (e.g., bus error) when the peek or poke operation was attempted. Note that some busses are incapable of generating exceptions when non-existent hardware is accessed. In such cases, these functions will always return zero and the value of the data read by bus_space_peek_N() will be unspecified. Finally, it should be noted that at this time the bus_space_peek_N() and bus_space_poke_N() functions are not re-entrant and should not, there- fore, be used from within an interrupt service routine. This constraint may be removed at some point in the future. bus_space_peek_1(space, handle, offset, datap) bus_space_peek_2(space, handle, offset, datap) bus_space_peek_4(space, handle, offset, datap) bus_space_peek_8(space, handle, offset, datap) The bus_space_peek_N() family of functions cautiously read a 1, 2, 4, or 8 byte data item from the offset specified by offset in the region speci- fied by handle of the bus space specified by space. The data item read is stored in the location pointed to by datap. It is permissible for datap to be NULL, in which case the data item will be discarded after being read. bus_space_poke_1(space, handle, offset, value) bus_space_poke_2(space, handle, offset, value) bus_space_poke_4(space, handle, offset, value) bus_space_poke_8(space, handle, offset, value) The bus_space_poke_N() family of functions cautiously write a 1, 2, 4, or 8 byte data item specified by value to the offset specified by offset in the region specified by handle of the bus space specified by space.
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 bus_space_barrier() function provides that ability. bus_space_barrier(space, handle, offset, length, flags) The bus_space_barrier() function enforces ordering of bus space read and write operations for the specified subregion (described by the offset and length parameters) of the region named by handle in the space named by space. The flags argument controls what types of operations are to be ordered. Supported flags are: BUS_SPACE_BARRIER_READ_BEFORE_READ Force all reads before the barrier to complete before any reads after the barrier may be issued. BUS_SPACE_BARRIER_READ_BEFORE_WRITE Force all reads before the barrier to complete before any writes after the barrier may be issued. BUS_SPACE_BARRIER_WRITE_BEFORE_READ Force all writes before the barrier to complete before any reads after the barrier may be issued. BUS_SPACE_BARRIER_WRITE_BEFORE_WRITE Force all writes before the barrier to complete before any writes after the barrier may be issued. BUS_SPACE_BARRIER_SYNC Force all memory operations and any pending exceptions to be completed before any instructions after the bar- rier may be issued. Those flags can be combined (or-ed together) to enforce ordering on dif- ferent combinations of read and write operations. 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. 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 off- set 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: /* * 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_BEFORE_WRITE); /* 1 */ bus_space_write_1(t, h, 0, data1); bus_space_barrier(t, h, 0, 2, BUS_SPACE_BARRIER_WRITE_BEFORE_READ); /* 2 */ ndata1 = bus_space_read_1(t, h, 1); bus_space_barrier(t, h, 1, 1, BUS_SPACE_BARRIER_READ_BEFORE_READ); /* 3 */ ndata0 = bus_space_read_1(t, h, 1); /* data0 == ndata0, data1 == ndata1 */ 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. The second barrier forces 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. The third barrier makes sure that the first read finishes before the sec- ond read is issued, ensuring that data is read correctly and in order. 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. The following barrier operations are obsolete and should be removed from existing code: BUS_SPACE_BARRIER_READ Synchronize read operations. BUS_SPACE_BARRIER_WRITE Synchronize write operations.
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 mem- ory, 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 bus_space_read_region_N() and bus_space_write_region_N() families of functions are provided. 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 bus_space_copy_region_N() family of functions and the bus_space_set_region_N() family of functions allow drivers to perform these operations. bus_space_read_region_1(space, handle, offset, datap, count) bus_space_read_region_2(space, handle, offset, datap, count) bus_space_read_region_4(space, handle, offset, datap, count) bus_space_read_region_8(space, handle, offset, datap, count) The bus_space_read_region_N() family of functions reads count 1, 2, 4, or 8 byte data items from bus space starting at byte offset offset in the region specified by handle of the bus space specified by space and writes them into the array specified by 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 handle. For portability, the starting address of the region specified by 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. Read operations done by the 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 bus_space_barrier() function. There is no way to insert barriers between reads of individual bus space locations executed by the bus_space_read_region_N() functions. 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. bus_space_write_region_1(space, handle, offset, datap, count) bus_space_write_region_2(space, handle, offset, datap, count) bus_space_write_region_4(space, handle, offset, datap, count) bus_space_write_region_8(space, handle, offset, datap, count) The bus_space_write_region_N() family of functions reads count 1, 2, 4, or 8 byte data items from the array specified by datap and writes them to bus space starting at byte offset offset in the region specified by handle of the bus space specified by 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 handle. For portability, the starting address of the region specified by 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. Write operations done by the 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 bus_space_barrier() function. There is no way to insert barriers between writes of individual bus space locations executed by the bus_space_write_region_N() functions. 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. bus_space_copy_region_1(space, srchandle, srcoffset, dsthandle, dstoffset, count) bus_space_copy_region_2(space, srchandle, srcoffset, dsthandle, dstoffset, count) bus_space_copy_region_4(space, srchandle, srcoffset, dsthandle, dstoffset, count) bus_space_copy_region_8(space, srchandle, srcoffset, dsthandle, dstoffset, count) The bus_space_copy_region_N() family of functions copies count 1, 2, 4, or 8 byte data items in bus space from the area starting at byte offset srcoffset in the region specified by srchandle of the bus space specified by space to the area starting at byte offset dstoffset in the region specified by 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 respec- tive handles. 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. Read and write operations done by the bus_space_copy_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 bus_space_barrier(function). There is no way to insert barriers between reads or writes of individual bus space locations executed by the bus_space_copy_region_N() functions. Overlapping copies between different subregions of a single region of bus space are handled correctly by the bus_space_copy_region_N() functions. 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. bus_space_set_region_1(space, handle, offset, value, count) bus_space_set_region_2(space, handle, offset, value, count) bus_space_set_region_4(space, handle, offset, value, count) bus_space_set_region_8(space, handle, offset, value, count) The bus_space_set_region_N() family of functions writes the given value to count 1, 2, 4, or 8 byte data items in bus space starting at byte off- set offset in the region specified by handle of the bus space specified by 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 handle. For portability, the starting address of the region specified by 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 incor- rect data to be written, on others it may cause a system crash. Write operations done by the 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 bus_space_barrier() function. There is no way to insert barriers between writes of individual bus space locations executed by the bus_space_set_region_N() functions. 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.
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 bus_space_read_multi_N() and bus_space_write_multi_N() families of func- tions are provided. bus_space_read_multi_1(space, handle, offset, datap, count) bus_space_read_multi_2(space, handle, offset, datap, count) bus_space_read_multi_4(space, handle, offset, datap, count) bus_space_read_multi_8(space, handle, offset, datap, count) The bus_space_read_multi_N() family of functions reads count 1, 2, 4, or 8 byte data items from bus space at byte offset offset in the region specified by handle of the bus space specified by space and writes them into the array specified by 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 handle. For portability, the starting address of the region specified by 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. Read operations done by the bus_space_read_multi_N() functions may be executed out of order with respect to other pending read and write opera- tions unless order is enforced by use of the bus_space_barrier() func- tion. Because the 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. 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. bus_space_write_multi_1(space, handle, offset, datap, count) bus_space_write_multi_2(space, handle, offset, datap, count) bus_space_write_multi_4(space, handle, offset, datap, count) bus_space_write_multi_8(space, handle, offset, datap, count) The bus_space_write_multi_N() family of functions reads count 1, 2, 4, or 8 byte data items from the array specified by datap and writes them into bus space at byte offset offset in the region specified by handle of the bus space specified by 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 handle. For portability, the starting address of the region specified by 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. Write operations done by the bus_space_write_multi_N() functions may be executed out of order with respect to other pending read and write opera- tions unless order is enforced by use of the bus_space_barrier() func- tion. Because the 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. 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.
STREAM FUNCTIONS
Most of the bus_space 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 bus_space_*_stream_N() functions. bus_space_read_stream_1(space, handle, offset) bus_space_read_stream_2(space, handle, offset) bus_space_read_stream_4(space, handle, offset) bus_space_read_stream_8(space, handle, offset) bus_space_read_multi_stream_1(space, handle, offset, datap, count) bus_space_read_multi_stream_2(space, handle, offset, datap, count) bus_space_read_multi_stream_4(space, handle, offset, datap, count) bus_space_read_multi_stream_8(space, handle, offset, datap, count) bus_space_read_region_stream_1(space, handle, offset, datap, count) bus_space_read_region_stream_2(space, handle, offset, datap, count) bus_space_read_region_stream_4(space, handle, offset, datap, count) bus_space_read_region_stream_8(space, handle, offset, datap, count) bus_space_write_stream_1(space, handle, offset, value) bus_space_write_stream_2(space, handle, offset, value) bus_space_write_stream_4(space, handle, offset, value) bus_space_write_stream_8(space, handle, offset, value) bus_space_write_multi_stream_1(space, handle, offset, datap, count) bus_space_write_multi_stream_2(space, handle, offset, datap, count) bus_space_write_multi_stream_4(space, handle, offset, datap, count) bus_space_write_multi_stream_8(space, handle, offset, datap, count) bus_space_write_region_stream_1(space, handle, offset, datap, count) bus_space_write_region_stream_2(space, handle, offset, datap, count) bus_space_write_region_stream_4(space, handle, offset, datap, count) bus_space_write_region_stream_8(space, handle, offset, datap, count) These functions are defined just as their non-stream counterparts, except that they provide no byte-order translation.
EXPECTED CHANGES TO THE BUS_SPACE FUNCTIONS
The definition of the bus_space functions should not yet be considered finalized. There are several changes and improvements which should be explored, including: · 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 #define in the incorrectly-written drivers. Unfortu- nately, at this time, few drivers actually use barriers correctly (or at all). Because of that, bus_space implementations on architectures which do buffering must always do the barriers inside the bus_space calls, to be safe. That has a potentially significant performance impact. · Exporting the bus_space functions to user-land so that applications (such as X servers) have easier, more portable access to device space. · 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. · Converting bus spaces (such as PCI configuration space) which cur- rently use space-specific access methods to use the bus_space func- tions where that is appropriate. · 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 bus_space functions into user space (since mapping in user space would look like it just used a different bus-specific mapping function).
COMPATIBILITY
The current version of the bus_space 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, depre- cated specification can be compiled by defining the __BUS_SPACE_COMPAT_OLDDEFS preprocessor symbol before including <machine/bus.h>.
SEE ALSO
bus_dma(9), mb(9)
HISTORY
The bus_space functions were introduced in a different form (memory and I/O spaces were accessed via different sets of functions) in NetBSD 1.2. The functions were merged to work on generic ``spaces'' early in the NetBSD 1.3 development cycle, and many drivers were converted to use them. This document was written later during the NetBSD 1.3 development cycle and the specification was updated to fix some consistency problems and to add some missing functionality.
AUTHORS
The bus_space interfaces were designed and implemented by the NetBSD developer community. Primary contributors and implementors were Chris Demetriou, Jason Thorpe, and Charles Hannum, but the rest of the NetBSD developers and the user community played a significant role in develop- ment. Chris Demetriou wrote this manual page. NetBSD 5.0 March 1, 2008 NetBSD 5.0
Powered by man-cgi (2024-08-26). Maintained for NetBSD by Kimmo Suominen. Based on man-cgi by Panagiotis Christias.