1			DMA attributes
  2			==============
  3
  4This document describes the semantics of the DMA attributes that are
  5defined in linux/dma-attrs.h.
  6
  7DMA_ATTR_WRITE_BARRIER
  8----------------------
  9
 10DMA_ATTR_WRITE_BARRIER is a (write) barrier attribute for DMA.  DMA
 11to a memory region with the DMA_ATTR_WRITE_BARRIER attribute forces
 12all pending DMA writes to complete, and thus provides a mechanism to
 13strictly order DMA from a device across all intervening busses and
 14bridges.  This barrier is not specific to a particular type of
 15interconnect, it applies to the system as a whole, and so its
 16implementation must account for the idiosyncrasies of the system all
 17the way from the DMA device to memory.
 18
 19As an example of a situation where DMA_ATTR_WRITE_BARRIER would be
 20useful, suppose that a device does a DMA write to indicate that data is
 21ready and available in memory.  The DMA of the "completion indication"
 22could race with data DMA.  Mapping the memory used for completion
 23indications with DMA_ATTR_WRITE_BARRIER would prevent the race.
 24
 25DMA_ATTR_WEAK_ORDERING
 26----------------------
 27
 28DMA_ATTR_WEAK_ORDERING specifies that reads and writes to the mapping
 29may be weakly ordered, that is that reads and writes may pass each other.
 30
 31Since it is optional for platforms to implement DMA_ATTR_WEAK_ORDERING,
 32those that do not will simply ignore the attribute and exhibit default
 33behavior.
 34
 35DMA_ATTR_WRITE_COMBINE
 36----------------------
 37
 38DMA_ATTR_WRITE_COMBINE specifies that writes to the mapping may be
 39buffered to improve performance.
 40
 41Since it is optional for platforms to implement DMA_ATTR_WRITE_COMBINE,
 42those that do not will simply ignore the attribute and exhibit default
 43behavior.
 44
 45DMA_ATTR_NON_CONSISTENT
 46-----------------------
 47
 48DMA_ATTR_NON_CONSISTENT lets the platform to choose to return either
 49consistent or non-consistent memory as it sees fit.  By using this API,
 50you are guaranteeing to the platform that you have all the correct and
 51necessary sync points for this memory in the driver.
 52
 53DMA_ATTR_NO_KERNEL_MAPPING
 54--------------------------
 55
 56DMA_ATTR_NO_KERNEL_MAPPING lets the platform to avoid creating a kernel
 57virtual mapping for the allocated buffer. On some architectures creating
 58such mapping is non-trivial task and consumes very limited resources
 59(like kernel virtual address space or dma consistent address space).
 60Buffers allocated with this attribute can be only passed to user space
 61by calling dma_mmap_attrs(). By using this API, you are guaranteeing
 62that you won't dereference the pointer returned by dma_alloc_attr(). You
 63can treat it as a cookie that must be passed to dma_mmap_attrs() and
 64dma_free_attrs(). Make sure that both of these also get this attribute
 65set on each call.
 66
 67Since it is optional for platforms to implement
 68DMA_ATTR_NO_KERNEL_MAPPING, those that do not will simply ignore the
 69attribute and exhibit default behavior.
 70
 71DMA_ATTR_SKIP_CPU_SYNC
 72----------------------
 73
 74By default dma_map_{single,page,sg} functions family transfer a given
 75buffer from CPU domain to device domain. Some advanced use cases might
 76require sharing a buffer between more than one device. This requires
 77having a mapping created separately for each device and is usually
 78performed by calling dma_map_{single,page,sg} function more than once
 79for the given buffer with device pointer to each device taking part in
 80the buffer sharing. The first call transfers a buffer from 'CPU' domain
 81to 'device' domain, what synchronizes CPU caches for the given region
 82(usually it means that the cache has been flushed or invalidated
 83depending on the dma direction). However, next calls to
 84dma_map_{single,page,sg}() for other devices will perform exactly the
 85same synchronization operation on the CPU cache. CPU cache synchronization
 86might be a time consuming operation, especially if the buffers are
 87large, so it is highly recommended to avoid it if possible.
 88DMA_ATTR_SKIP_CPU_SYNC allows platform code to skip synchronization of
 89the CPU cache for the given buffer assuming that it has been already
 90transferred to 'device' domain. This attribute can be also used for
 91dma_unmap_{single,page,sg} functions family to force buffer to stay in
 92device domain after releasing a mapping for it. Use this attribute with
 93care!
 94
 95DMA_ATTR_FORCE_CONTIGUOUS
 96-------------------------
 97
 98By default DMA-mapping subsystem is allowed to assemble the buffer
 99allocated by dma_alloc_attrs() function from individual pages if it can
100be mapped as contiguous chunk into device dma address space. By
101specifying this attribute the allocated buffer is forced to be contiguous
102also in physical memory.
103
104DMA_ATTR_ALLOC_SINGLE_PAGES
105---------------------------
106
107This is a hint to the DMA-mapping subsystem that it's probably not worth
108the time to try to allocate memory to in a way that gives better TLB
109efficiency (AKA it's not worth trying to build the mapping out of larger
110pages).  You might want to specify this if:
111- You know that the accesses to this memory won't thrash the TLB.
112  You might know that the accesses are likely to be sequential or
113  that they aren't sequential but it's unlikely you'll ping-pong
114  between many addresses that are likely to be in different physical
115  pages.
116- You know that the penalty of TLB misses while accessing the
117  memory will be small enough to be inconsequential.  If you are
118  doing a heavy operation like decryption or decompression this
119  might be the case.
120- You know that the DMA mapping is fairly transitory.  If you expect
121  the mapping to have a short lifetime then it may be worth it to
122  optimize allocation (avoid coming up with large pages) instead of
123  getting the slight performance win of larger pages.
124Setting this hint doesn't guarantee that you won't get huge pages, but it
125means that we won't try quite as hard to get them.
126
127NOTE: At the moment DMA_ATTR_ALLOC_SINGLE_PAGES is only implemented on ARM,
128though ARM64 patches will likely be posted soon.