0001 The io_mapping functions in linux/io-mapping.h provide an abstraction for
0002 efficiently mapping small regions of an I/O device to the CPU. The initial
0003 usage is to support the large graphics aperture on 32-bit processors where
0004 ioremap_wc cannot be used to statically map the entire aperture to the CPU
0005 as it would consume too much of the kernel address space.
0007 A mapping object is created during driver initialization using
0009 struct io_mapping *io_mapping_create_wc(unsigned long base,
0010 unsigned long size)
0012 'base' is the bus address of the region to be made
0013 mappable, while 'size' indicates how large a mapping region to
0014 enable. Both are in bytes.
0016 This _wc variant provides a mapping which may only be used
0017 with the io_mapping_map_atomic_wc or io_mapping_map_wc.
0019 With this mapping object, individual pages can be mapped either atomically
0020 or not, depending on the necessary scheduling environment. Of course, atomic
0021 maps are more efficient:
0023 void *io_mapping_map_atomic_wc(struct io_mapping *mapping,
0024 unsigned long offset)
0026 'offset' is the offset within the defined mapping region.
0027 Accessing addresses beyond the region specified in the
0028 creation function yields undefined results. Using an offset
0029 which is not page aligned yields an undefined result. The
0030 return value points to a single page in CPU address space.
0032 This _wc variant returns a write-combining map to the
0033 page and may only be used with mappings created by
0036 Note that the task may not sleep while holding this page
0039 void io_mapping_unmap_atomic(void *vaddr)
0041 'vaddr' must be the value returned by the last
0042 io_mapping_map_atomic_wc call. This unmaps the specified
0043 page and allows the task to sleep once again.
0045 If you need to sleep while holding the lock, you can use the non-atomic
0046 variant, although they may be significantly slower.
0048 void *io_mapping_map_wc(struct io_mapping *mapping,
0049 unsigned long offset)
0051 This works like io_mapping_map_atomic_wc except it allows
0052 the task to sleep while holding the page mapped.
0054 void io_mapping_unmap(void *vaddr)
0056 This works like io_mapping_unmap_atomic, except it is used
0057 for pages mapped with io_mapping_map_wc.
0059 At driver close time, the io_mapping object must be freed:
0061 void io_mapping_free(struct io_mapping *mapping)
0063 Current Implementation:
0065 The initial implementation of these functions uses existing mapping
0066 mechanisms and so provides only an abstraction layer and no new
0069 On 64-bit processors, io_mapping_create_wc calls ioremap_wc for the whole
0070 range, creating a permanent kernel-visible mapping to the resource. The
0071 map_atomic and map functions add the requested offset to the base of the
0072 virtual address returned by ioremap_wc.
0074 On 32-bit processors with HIGHMEM defined, io_mapping_map_atomic_wc uses
0075 kmap_atomic_pfn to map the specified page in an atomic fashion;
0076 kmap_atomic_pfn isn't really supposed to be used with device pages, but it
0077 provides an efficient mapping for this usage.
0079 On 32-bit processors without HIGHMEM defined, io_mapping_map_atomic_wc and
0080 io_mapping_map_wc both use ioremap_wc, a terribly inefficient function which
0081 performs an IPI to inform all processors about the new mapping. This results
0082 in a significant performance penalty.