]> www.infradead.org Git - nvme.git/commitdiff
drm/amd/display: Disable CONFIG_DRM_AMD_DC_FP for RISC-V with clang
authorNathan Chancellor <nathan@kernel.org>
Fri, 14 Jun 2024 19:54:52 +0000 (12:54 -0700)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 19 Jun 2024 18:17:25 +0000 (14:17 -0400)
Commit 77acc6b55ae4 ("riscv: add support for kernel-mode FPU") and
commit a28e4b672f04 ("drm/amd/display: use ARCH_HAS_KERNEL_FPU_SUPPORT")
enabled support for CONFIG_DRM_AMD_DC_FP with RISC-V. Unfortunately,
this exposed -Wframe-larger-than warnings (which become fatal with
CONFIG_WERROR=y) when building ARCH=riscv allmodconfig with clang:

  drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn32/display_mode_vba_32.c:58:13: error: stack frame size (2448) exceeds limit (2048) in 'DISPCLKDPPCLKDCFCLKDeepSleepPrefetchParametersWatermarksAndPerformanceCalculation' [-Werror,-Wframe-larger-than]
     58 | static void DISPCLKDPPCLKDCFCLKDeepSleepPrefetchParametersWatermarksAndPerformanceCalculation(
        |             ^
  1 error generated.

Many functions in this file use a large number of parameters, which must
be passed on the stack at a certain pointer due to register exhaustion,
which can cause high stack usage when inlining and issues with stack
slot analysis get involved. While the compiler can and should do better
(as GCC uses less than half the amount of stack space for the same
function), it is not as simple as a fix as adjusting the functions not
to take a large number of parameters.

Unfortunately, modifying these files to avoid the problem is a difficult
to justify approach because any revisions to the files in the kernel
tree never make it back to the original source (so copies of the code
for newer hardware revisions just reintroduce the issue) and the files
are hard to read/modify due to being "gcc-parsable HW gospel, coming
straight from HW engineers".

Avoid building the problematic code for RISC-V by modifying the existing
condition for arm64 that exists for the same reason. Factor out the
logical not to make the condition a little more readable naturally.

Fixes: a28e4b672f04 ("drm/amd/display: use ARCH_HAS_KERNEL_FPU_SUPPORT")
Reported-by: Palmer Dabbelt <palmer@rivosinc.com>
Closes: https://lore.kernel.org/20240530145741.7506-2-palmer@rivosinc.com/
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/display/Kconfig

index 5fcd4f778dc3dd3e350b9e732ea85043931b6252..47b8b49da8a72b6e2a11a816466678f0a69b7264 100644 (file)
@@ -8,7 +8,7 @@ config DRM_AMD_DC
        depends on BROKEN || !CC_IS_CLANG || ARM64 || RISCV || SPARC64 || X86_64
        select SND_HDA_COMPONENT if SND_HDA_CORE
        # !CC_IS_CLANG: https://github.com/ClangBuiltLinux/linux/issues/1752
-       select DRM_AMD_DC_FP if ARCH_HAS_KERNEL_FPU_SUPPORT && (!ARM64 || !CC_IS_CLANG)
+       select DRM_AMD_DC_FP if ARCH_HAS_KERNEL_FPU_SUPPORT && !(CC_IS_CLANG && (ARM64 || RISCV))
        help
          Choose this option if you want to use the new display engine
          support for AMDGPU. This adds required support for Vega and