When using a coherent DMA API wee must make sure to program
any aperture fields with the coherent aperture setting. To
do this the nvgpu_aperture_mask() function was modified to
take a third aperture mask argument, a coherent setting, so
that code can use this function to generate coherent aperture
settings.
The aperture choice is some what tricky: the default version
of this function uses the state of the DMA API to determine
what aperture to use for SYSMEM: either coherent or
non-coherent internally. Thus a kernel user need only specify
the normal nvgpu_mem struct and the correct mask should be
chosen. Due to many uses of nvgpu_mem structs not created
directly from the DMA API wrapper it's easier to translate
SYSMEM to SYSMEM_COH after creation.
However, the GMMU mapping code, will encounter buffers from
userspace with difference coerency attributes than the DMA
API. Thus the __nvgpu_aperture_mask() really respects the
aperture setting passed in regardless of the DMA API state.
This aperture setting is pulled from NVGPU_VM_MAP_IO_COHERENT
since this is either passed in from userspace or set by the
kernel when using coherent DMA. The aperture field in attrs
is upgraded to coh if this flag is set.
This change also adds a coherent sysmem mask everywhere that
it can. There's a couple places that do not have a coherent
register field defined yet. These need to eventually be
defined and added.
Lastly the aperture mask code has been mvoed from the Linux
vm.c code to the general vm.c code since this function has
no Linux dependencies.
Note: depends on https://git-master.nvidia.com/r/1664536 for
new register fields.
JIRA EVLR-2333
Change-Id: I4b347911ecb7c511738563fe6c34d0e6aa380d71
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1655220
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The GPU has multiple different operating modes in respect to IOMMU'ability.
As such there needs to be a clean way to tell the driver whether it is
IOMMU'able or not. This state also does not always reflect what is possible:
all becasue the GPU can generate IOMMU'ed memory requests doesn't mean it
wants to.
The nvgpu_iommuable() API has now existed for a little while which is a
useful way to convey whether nvgpu should consider the GPU as IOMMU'able.
However, there is also the g->mm.bypass_smmu flag which used to be able to
override what the GPU decided it should do. Typically it was assigned
the same value as nvgpu_iommuable() but that was not necessarily a
requirment.
This patch removes all the usages of g->mm.bypass_smmu and instead uses the
nvgpu_iommuable() function. All places where the check against
g->mm.bypass_smmu have been replaced with nvgpu_iommuable(). The code
should now be much cleaner.
Subsequently other checks can also be placed in the nvgpu_iommuable()
function. For example, when NVLINK comes online and the GPU should no
longer consider DMA addresses and instead use scatter-gather lists
directly the ngpu_iommuable() function will be able to check the state of
NVLINK and then act accordingly.
Change-Id: I0da6262386de15709decac89d63d3eecfec20cd7
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1648332
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Drastically simplify and move the aligment computation for buffers
getting mapped into the SGT code. An SGT is all that is needed for
computing the alignment.
However, this did require that a new SGT op was added:
nvgpu_sgt_iommuable()
This function returns true if the passed SGT is IOMMU'able and must
be implemented by an SGT implementation that has IOMMU'able buffers.
If this function is left as NULL then it is assumed that the buffer
is not IOMMU'able.
Also cleanup the parameter ordering convention among all nvgpu_sgt
functions. Previously there was a mishmash of different parameter
orderings. This patch now standardizes on the gk20a first approach
seen everywhere else in the driver.
JIRA NVGPU-30
JIRA NVGPU-246
JIRA NVGPU-71
Change-Id: Ic4ab7b752847cf795c7cfafed5a07818217bba86
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1583985
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Split VIDMEM support into its own code files organized as such:
common/mm/vidmem.c - Base vidmem support
common/linux/vidmem.c - Linux specific user-space interaction
include/nvgpu/vidmem.h - Vidmem API definitions
Also use the config to enable/disable VIDMEM support in the makefile
and remove as many CONFIG_GK20A_VIDMEM preprocessor checks as possible
from the source code.
And lastly update a while-loop that iterated over an SGT to use the
new for_each construct for iterating over SGTs.
Currently this organization is not perfectly adhered to. More patches
will fix that.
JIRA NVGPU-30
JIRA NVGPU-138
Change-Id: Ic0f4d2cf38b65849c7dc350a69b175421477069c
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1540705
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Rename get_physical_addr_bits and related functions to something that
more clearly conveys what they are doing. The basic idea of these
functions is to translate from a physical GPU address to a IOMMU GPU
address. To do that a particular bit (that varies from chip to chip)
is added to the physical address.
JIRA NVGPU-68
Change-Id: I536cc595c4397aad69a24f740bc74db03f52bc0a
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1542966
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The basic nvgpu_mem_sgl implementation provides support
for OS specific scatter-gather list implementations by
simply copying them node by node. This is inefficient,
taking extra time and memory.
This patch implements an nvgpu_mem_sgt struct to act as
a header which is inserted at the front of any scatter-
gather list implementation. This labels every struct
with a set of ops which can be used to interact with
the attached scatter gather list.
Since nvgpu common code only has to interact with these
function pointers, any sgl implementation can be used.
Initialization only requires the allocation of a single
struct, removing the need to copy or iterate through the
sgl being converted.
Jira NVGPU-186
Change-Id: I2994f804a4a4cc141b702e987e9081d8560ba2e8
Signed-off-by: Sunny He <suhe@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1541426
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The last major item preventing the core MM code in the nvgpu
driver from being platform agnostic is the usage of Linux
scattergather tables and scattergather lists. These data
structures are used throughout the mapping code to handle
discontiguous DMA allocations and also overloaded to represent
VIDMEM allocs.
The notion of a scatter gather table is crucial to a HW device
that can handle discontiguous DMA. The GPU has a MMU which
allows the GPU to do page gathering and present a virtually
contiguous buffer to the GPU HW. As a result it makes sense
for the GPU driver to use some sort of scatter gather concept
so maximize memory usage efficiency.
To that end this patch keeps the notion of a scatter gather
list but implements it in the nvgpu common code. It is based
heavily on the Linux SGL concept. It is a singly linked list
of blocks - each representing a chunk of memory. To map or
use a DMA allocation SW must iterate over each block in the
SGL.
This patch implements the most basic level of support for this
data structure. There are certainly easy optimizations that
could be done to speed up the current implementation. However,
this patches' goal is to simply divest the core MM code from
any last Linux'isms. Speed and efficiency come next.
Change-Id: Icf44641db22d87fa1d003debbd9f71b605258e42
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1530867
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
