The compiler option -Wmissing-prototypes is being enabled globally in
the upstream Linux kernel and this causes build failures for nvgpu. The
build failures occur because either the driver is missing an include
file which has the prototype or because the function is not declared
statically when it should be (ie. there are no external users).
Fix the various build failures and enable -Wmissing-prototypes to
prevent any new instances from occurring.
Bug 4404965
Change-Id: I551922836e37b0c94c158232d6277f4053e9d2d3
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/3027483
(cherry picked from commit e8cbf90db2d0db7277db9e3eec9fb88d69c7fcc7)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/3035518
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
- As part of mmap call handler mmap_sem is being acquired which is
causing an BUG: scheduling while atomic:. This is happening because
mmap callback gets called with mmap_sem held. The mmap file handler
is called via call_mmap() via mmap_region() via do_mmap(), and the
caller of do_mmap() must hold down_write(¤t->mm->mmap_sem).
To fix this issue removing the mmap_sem locking from nvs mmap handler.
- If remap_vmalloc_range() is used to map the pages to userspace then
allocated pages should be mapped into virtually contiguous space by
passing VM_USERMAP flag to vmap(). Passing VM_USERMAP to vmap() call
if API nvgpu_dma_alloc_flags_sys() is called with flag
NVGPU_DMA_VM_USERMAP_ADDRESS.
Bug 3884011
Change-Id: I64264f6f6e0dd75b1f828dc58355d740e6ef5ccb
Signed-off-by: prsethi <prsethi@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2820781
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
The simulator ring buffer DMA interface supports buffers of the following sizes:
4, 8, 12 and 16K. At present, it is configured to 4K and it happens to match
with the kernel PAGE_SIZE, which is used to wrap back the GET/PUT pointers once
4K is reached. However, this is not always true; for instance, take 64K pages.
Hence, replace PAGE_SIZE with SIM_BFR_SIZE.
Introduce macro NVGPU_CPU_PAGE_SIZE which aliases to PAGE_SIZE and replace
latter with former.
Bug 200658101
Jira NVGPU-6018
Change-Id: I83cc62b87291734015c51f3e5a98173549e065de
Signed-off-by: Antony Clince Alex <aalex@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2420728
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Mapping of large buffers to GMMU end up needing many
pages for the PTE tables. Allocating these one by one
can end up being a performance bottleneck, particularly
in the virtualized case.
This is adding the following changes:
- As the TLB invalidation doesn't have access to mem_off,
allow top-level allocation by alloc_cache_direct().
- Define NVGPU_PD_CACHE_SIZE, the allocation size for a new slab
for the PD cache, effectively set to 64K bytes
- Use the PD cache for any allocation < NVGPU_PD_CACHE_SIZE
When freeing up cached entries, avoid prefetch errors by
invalidating the entry (memset to 0).
- Try to fall back to direct allocation of smaller chunk for
contiguous allocation failures.
- Unit test changes.
Bug 200649243
Change-Id: I0a667af0ba01d9147c703e64fc970880e52a8fbc
Signed-off-by: dt <dt@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2404371
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Rule 2.2 doesn't allow unused variable assignments. The reason is
presence of unused variable assignments may indicate error in program's
logic.
Rule 21.x doesn't allow reserved identifier or macro names starting with
'_' to be reused or defined.
Jira NVGPU-3864
Change-Id: I8ee31c0ee522cd4de00b317b0b4463868ac958ef
Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/2163723
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Safety build does not support vidmem. This patch compiles out vidmem
related changes - vidmem, dma alloc, cbc/acr/pmu alloc based on
vidmem and corresponding tests like pramin, page allocator &
gmmu_map_unmap_vidmem..
As vidmem is applicable only in case of DGPUs the code is compiled
out using CONFIG_NVGPU_DGPU.
JIRA NVGPU-3524
Change-Id: Ic623801112484ffc071195e828ab9f290f945d4d
Signed-off-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/2132773
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vaibhav Kachore <vkachore@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The latest GPU uses nvlink and its own MMU to access memory,
instead of SMMU like others. So it doesn't go through IOMMU
framework to allocate physically non-contiguous memory. The
DMA API had a pair of downstream functions to allocate the
memory for this situation, but it is removed since it's not
likely acceptable for upstream kernel.
In order not to hack the dma-direct ops that by its meaning
is supposed to provide contiguous memory, this patch adds a
pair of memory-allocation functions inside the gpu driver,
since nvgpu is the only user.
This pair of functions are only used when GPU driver doesn't
go through either dma-direct (FORCE_CONTIGUOUS) or iommu. It
also requires GPU driver to map the non-contiguous pages.
Bug 200444660
Change-Id: I26678a3f8d63bba340872beeecbb7b0e1e7a35fa
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/2029680
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The flag NVGPU_DMA_FORCE_CONTIGUOUS simply means that the memory
or the pages should be forced contiguous. Meanwhile, the other
flag NVGPU_DMA_PHYSICALLY_ADDRESSED means that the memory should
be contiguous from GPU perspective, either physically contiguous
when IOMMU is not used, or virtually contiguous by IOMMU.
Thus the NVGPU_DMA_FORCE_CONTIGUOUS flag is now redundant.
This patch cleans up the NVGPU_DMA_FORCE_CONTIGUOUS flag.
Bug 200444660
Change-Id: I63bb06fea728b34ec2c6f831504392d42c426d55
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/2035403
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The original free routine has three options:
if (NVGPU_DMA_NO_KERNEL_MAPPING)
dma_free_attrs(d, mem->aligned_size, mem->priv.pages,
else if (other flags)
dma_free_attrs(d, mem->aligned_size, mem->cpu_va,
else /* No flags */
dma_free_coherent(d, mem->aligned_size, mem->cpu_va,
The last dma_free_coherent() can be unwrapped to dma_free_attrs
with its dma_attrs=0, while the former two are identical except
cpu_addr. So this patch merges these three into one single call
but differentiate the cpu_addr and dma_attrs parameters.
Note that the dma_free_attrs returns 0 when flags is not set.
Bug 200444660
Change-Id: I92ec0390138c79c5109973e476ea0ea719d4e2b9
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/2029679
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add error prints for DMA alloc failures. This way when there is a
DMA alloc failure the failure is clear. Without this it's hard to
know what exactly is causing any given -ENOMEM issue or what the
specifics of said ENOMEM case are.
Change-Id: Ia535895ae07bc1704edaed564edbb6f6dfbf6518
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1976441
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vinod Gopalakrishnakurup <vinodg@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Instead of using kmalloc() to get a buffer for storing the
computed flags string during DMA debugging use a stack
buffer. This removes the need for a kmalloc() call. The
problem with kmalloc() is that if a dma_alloc() fails due
to being out of memory the kmalloc may likely fail, too!
Also simplify the logic now that there's no need to do
any error checking for a kmalloc() call.
Change-Id: I45c1fd16658212188a1206a2edf17b28f3c06c9e
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1976440
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vinod Gopalakrishnakurup <vinodg@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Make a physical nvgpu_mem implementation in the common code. This
implementation assumes a single, contiguous, physical range. GMMU
mappability is provided by building a one entry SGT.
Since this is now "common" code the original Linux code has been
moved to commom/mm/nvgpu_mem.c.
Also fix the '__' prefix in the nvgpu_mem function. This is not
necessary as this function, although somewhat tricky, is expected
to be used by arbitrary users within the nvgpu driver.
JIRA NVGPU-1029
Bug 2441531
Change-Id: I42313e5c664df3cd94933cc63ff0528326628683
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1995866
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The commit 3fdd8e38b2 ("gpu: nvgpu: Use our own vmap() for
coherent DMA buffers") added an NVGPU_DMA_NO_KERNEL_MAPPING
flag for coherent chips to work around a memory mapping bug
suspiciously from DMA API.
However, this requires dma-mapping code of ARM64 to support
a legacy DMA_NO_KERNEL_MAPPING attribute for DMA allocation,
which will not likely get upstreamed -- it is not long-term
sustainable. So the plan is to remove this flag from ARM64
part.
The results of 3D benchmarks and GVS sanity tests show that
the system has no regressions in stability, and no mapping
issue being observed after removing this WAR. In case that
GPU code encounters mapping issue in the future, we should
fix from the general DMA API side instead.
Bug 2424160
Change-Id: Ice91f2b2c924beb2f83762cb02efbd53fe7df1c0
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/2001294
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Some buffers like userd are physically addressed. If nvlink is
enabled, or device is not iommuable, this requires buffer to be
physically contiguous.
Add NVGPU_DMA_PHYSICALLY_ADDRESSED to identify such buffers, in
order to force physically contiguous allocation, only in above
cases.
Bug 2422486
Change-Id: I6426e23b064904e812e6b33e6d706391648a51ae
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1959034
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
MISRA rule 14.4 doesn't allow the usage of non-boolean variable as
boolean in the controlling expression of an if statement or an
iteration statement.
Fix violations where a non-boolean variable is used as a boolean in the
controlling expression of if and loop statements.
JIRA NVGPU-1022
Change-Id: I957f8ca1fa0eb00928c476960da1e6e420781c09
Signed-off-by: Amurthyreddy <amurthyreddy@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1941002
GVS: Gerrit_Virtual_Submit
Reviewed-by: Adeel Raza <araza@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
1) Update header path of gk20a.h files present in os/
to <nvgpu/gk20a.h>
2) os_fence_android_sema.c indirectly was dependent on gk20a.h via
semaphore.h. So, added #include <nvgpu/gk20a.h> in
os_fence_android_sema.c and replaced the header with forward
declaration of struct gk20a in semaphore.h
Jira NVGPU-597
Change-Id: I96e23befeb80713f3a399071eb5498f6f580211d
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1842868
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This migrates the common DMA code (os agnostic) to the
common directory. This new unit will be the common DMA
allocator that lets users allocate SYSMEM, VIDMEM, or
either. Other units will be responsible for actually
handling the mechanics of allocating VIDMEM or SYSMEM.
Also update the names of the DMA related files so that
tmake doesn't complain about duplicate C file names. To
do this call the common DMA file dma.c and prepend the
OS to the other DMA files. So now we have:
common/mm/dma.c
os/posix/posix-dma.c
os/linux/linux-dma.c
JIRA NVGPU-990
Change-Id: I22d2d41803ad89be7d9c28f87864ce4fedf10836
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1799807
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
