1 # SPDX-License-Identifier: GPL-2.0-only
16 # IOMMU drivers that can bypass the IOMMU code and optionally use the direct
17 # mapping fast path should select this option and set the dma_ops_bypass
18 # flag in struct device where applicable
23 # Lets platform IOMMU driver choose between bypass and IOMMU
24 config ARCH_HAS_DMA_MAP_DIRECT
27 config NEED_SG_DMA_LENGTH
30 config NEED_DMA_MAP_STATE
33 config ARCH_DMA_ADDR_T_64BIT
34 def_bool 64BIT || PHYS_ADDR_T_64BIT
36 config ARCH_HAS_DMA_SET_MASK
40 # Select this option if the architecture needs special handling for
41 # DMA_ATTR_WRITE_COMBINE. Normally the "uncached" mapping should be what
42 # people thing of when saying write combine, so very few platforms should
43 # need to enable this.
45 config ARCH_HAS_DMA_WRITE_COMBINE
49 # Select if the architectures provides the arch_dma_mark_clean hook
51 config ARCH_HAS_DMA_MARK_CLEAN
54 config DMA_DECLARE_COHERENT
57 config ARCH_HAS_SETUP_DMA_OPS
60 config ARCH_HAS_TEARDOWN_DMA_OPS
63 config ARCH_HAS_SYNC_DMA_FOR_DEVICE
66 config ARCH_HAS_SYNC_DMA_FOR_CPU
68 select NEED_DMA_MAP_STATE
70 config ARCH_HAS_SYNC_DMA_FOR_CPU_ALL
73 config ARCH_HAS_DMA_PREP_COHERENT
76 config ARCH_HAS_FORCE_DMA_UNENCRYPTED
81 select NEED_DMA_MAP_STATE
84 # Should be selected if we can mmap non-coherent mappings to userspace.
85 # The only thing that is really required is a way to set an uncached bit
88 config DMA_NONCOHERENT_MMAP
92 config DMA_COHERENT_POOL
93 select GENERIC_ALLOCATOR
96 config DMA_GLOBAL_POOL
97 select DMA_DECLARE_COHERENT
103 select DMA_NONCOHERENT_MMAP
105 config DMA_DIRECT_REMAP
108 select DMA_COHERENT_POOL
111 bool "DMA Contiguous Memory Allocator"
112 depends on HAVE_DMA_CONTIGUOUS && CMA
114 This enables the Contiguous Memory Allocator which allows drivers
115 to allocate big physically-contiguous blocks of memory for use with
116 hardware components that do not support I/O map nor scatter-gather.
118 You can disable CMA by specifying "cma=0" on the kernel's command
121 For more information see <kernel/dma/contiguous.c>.
126 config DMA_PERNUMA_CMA
127 bool "Enable separate DMA Contiguous Memory Area for each NUMA Node"
128 default NUMA && ARM64
130 Enable this option to get pernuma CMA areas so that devices like
131 ARM64 SMMU can get local memory by DMA coherent APIs.
133 You can set the size of pernuma CMA by specifying "cma_pernuma=size"
134 on the kernel's command line.
136 comment "Default contiguous memory area size:"
138 config CMA_SIZE_MBYTES
139 int "Size in Mega Bytes"
140 depends on !CMA_SIZE_SEL_PERCENTAGE
144 Defines the size (in MiB) of the default memory area for Contiguous
145 Memory Allocator. If the size of 0 is selected, CMA is disabled by
146 default, but it can be enabled by passing cma=size[MG] to the kernel.
149 config CMA_SIZE_PERCENTAGE
150 int "Percentage of total memory"
151 depends on !CMA_SIZE_SEL_MBYTES
155 Defines the size of the default memory area for Contiguous Memory
156 Allocator as a percentage of the total memory in the system.
157 If 0 percent is selected, CMA is disabled by default, but it can be
158 enabled by passing cma=size[MG] to the kernel.
161 prompt "Selected region size"
162 default CMA_SIZE_SEL_MBYTES
164 config CMA_SIZE_SEL_MBYTES
165 bool "Use mega bytes value only"
167 config CMA_SIZE_SEL_PERCENTAGE
168 bool "Use percentage value only"
170 config CMA_SIZE_SEL_MIN
171 bool "Use lower value (minimum)"
173 config CMA_SIZE_SEL_MAX
174 bool "Use higher value (maximum)"
179 int "Maximum PAGE_SIZE order of alignment for contiguous buffers"
183 DMA mapping framework by default aligns all buffers to the smallest
184 PAGE_SIZE order which is greater than or equal to the requested buffer
185 size. This works well for buffers up to a few hundreds kilobytes, but
186 for larger buffers it just a memory waste. With this parameter you can
187 specify the maximum PAGE_SIZE order for contiguous buffers. Larger
188 buffers will be aligned only to this specified order. The order is
189 expressed as a power of two multiplied by the PAGE_SIZE.
191 For example, if your system defaults to 4KiB pages, the order value
192 of 8 means that the buffers will be aligned up to 1MiB only.
194 If unsure, leave the default value "8".
199 bool "Enable debugging of DMA-API usage"
200 select NEED_DMA_MAP_STATE
202 Enable this option to debug the use of the DMA API by device drivers.
203 With this option you will be able to detect common bugs in device
204 drivers like double-freeing of DMA mappings or freeing mappings that
205 were never allocated.
207 This option causes a performance degradation. Use only if you want to
208 debug device drivers and dma interactions.
212 config DMA_API_DEBUG_SG
213 bool "Debug DMA scatter-gather usage"
215 depends on DMA_API_DEBUG
217 Perform extra checking that callers of dma_map_sg() have respected the
218 appropriate segment length/boundary limits for the given device when
219 preparing DMA scatterlists.
221 This is particularly likely to have been overlooked in cases where the
222 dma_map_sg() API is used for general bulk mapping of pages rather than
223 preparing literal scatter-gather descriptors, where there is a risk of
224 unexpected behaviour from DMA API implementations if the scatterlist
225 is technically out-of-spec.
229 config DMA_MAP_BENCHMARK
230 bool "Enable benchmarking of streaming DMA mapping"
233 Provides /sys/kernel/debug/dma_map_benchmark that helps with testing
234 performance of dma_(un)map_page.
236 See tools/testing/selftests/dma/dma_map_benchmark.c