87 lines
3.0 KiB
Plaintext
87 lines
3.0 KiB
Plaintext
|
menuconfig TEGRA_NVMAP
|
||
|
|
bool "Tegra GPU memory management driver (nvmap)"
|
||
|
|
select ARM_DMA_USE_IOMMU if IOMMU_API
|
||
|
|
select DMA_SHARED_BUFFER
|
||
|
|
select CRYPTO_LZO
|
||
|
|
default y
|
||
|
|
help
|
||
|
|
Say Y here to include the memory management driver for the Tegra
|
||
|
|
GPU, multimedia and display subsystems
|
||
|
|
|
||
|
|
if TEGRA_NVMAP
|
||
|
|
|
||
|
|
config NVMAP_PAGE_POOLS
|
||
|
|
bool "Use page pools to reduce allocation overhead"
|
||
|
|
default y
|
||
|
|
help
|
||
|
|
say Y here to reduce the alloction overhead, which is significant
|
||
|
|
for uncached, writecombine and inner cacheable memories as it
|
||
|
|
involves changing page attributes during every allocation per page
|
||
|
|
and flushing cache. Alloc time is reduced by allcoating the pages
|
||
|
|
ahead and keeping them aside. The reserved pages would be released
|
||
|
|
when system is low on memory and acquired back during release of
|
||
|
|
memory.
|
||
|
|
|
||
|
|
config NVMAP_PAGE_POOL_DEBUG
|
||
|
|
bool "Debugging for page pools"
|
||
|
|
depends on NVMAP_PAGE_POOLS
|
||
|
|
help
|
||
|
|
Say Y here to include some debugging info in the page pools. This
|
||
|
|
adds a bit of unnecessary overhead so only enable this is you
|
||
|
|
suspect there is an issue with the nvmap page pools.
|
||
|
|
|
||
|
|
config NVMAP_PAGE_POOL_SIZE
|
||
|
|
depends on NVMAP_PAGE_POOLS
|
||
|
|
hex "Page pool size in pages"
|
||
|
|
default 0x0
|
||
|
|
|
||
|
|
config NVMAP_COLOR_PAGES
|
||
|
|
bool "Color pages allocated"
|
||
|
|
default y
|
||
|
|
help
|
||
|
|
Say Y here to enable page coloring.
|
||
|
|
Page coloring rearranges the pages allocated based on the color
|
||
|
|
of the page. It can improve memory access performance.
|
||
|
|
The coloring option enable can optionally overallocate a portion of
|
||
|
|
reqeusted allcoation size to improve the probabilty of better
|
||
|
|
page coloring. If unsure, say Y.
|
||
|
|
|
||
|
|
config NVMAP_CACHE_MAINT_BY_SET_WAYS
|
||
|
|
bool "Enable cache maintenance by set/ways"
|
||
|
|
help
|
||
|
|
Say Y here to reduce cache maintenance overhead by MVA.
|
||
|
|
This helps in reducing cache maintenance overhead in the systems,
|
||
|
|
where inner cache includes only L1. For the systems, where inner cache
|
||
|
|
includes L1 and L2, keep this option disabled.
|
||
|
|
|
||
|
|
config NVMAP_FD_START
|
||
|
|
hex "FD number to start allocation from"
|
||
|
|
default 0x400
|
||
|
|
help
|
||
|
|
NvMap handles are represented with FD's in the user processes.
|
||
|
|
To avoid Linux FD usage limitations, NvMap allocates FD starting
|
||
|
|
from this number.
|
||
|
|
|
||
|
|
config NVMAP_DEFER_FD_RECYCLE
|
||
|
|
bool "Defer FD recycle"
|
||
|
|
help
|
||
|
|
Say Y here to enable deferred FD recycle.
|
||
|
|
A released nvmap handle would release memory and FD. This FD
|
||
|
|
can be reused immediately for subsequent nvmap allocation req in
|
||
|
|
the same process. Any buggy code in client process that continues to
|
||
|
|
use FD of released allocation would continue to use new allocation
|
||
|
|
and can lead to undesired consequences, which can be hard to debug.
|
||
|
|
Enabling this option would defer recycling FD for longer time and
|
||
|
|
allows debugging incorrect FD references by clients by returning errors
|
||
|
|
for the accesses that occur after handle/FD release.
|
||
|
|
|
||
|
|
config NVMAP_DEFER_FD_RECYCLE_MAX_FD
|
||
|
|
hex "FD number to start free FD recycle"
|
||
|
|
depends on NVMAP_DEFER_FD_RECYCLE
|
||
|
|
default 0x8000
|
||
|
|
help
|
||
|
|
Once last allocated FD reaches this number, allocation of subsequent
|
||
|
|
FD's start from NVMAP_START_FD.
|
||
|
|
|
||
|
|
endif
|