On 10/19/23 11:56, Nathan Chancellor wrote:
On Thu, Oct 19, 2023 at 02:53:01PM +0200, Arnd Bergmann wrote:
On Thu, Oct 19, 2023, at 12:04, Alexander Potapenko wrote:
So the remaining option would be to just increase the frame size every time a new function surpasses the limit.
That is clearly not an option, though we could try to add Kconfig dependencies that avoid the known bad combinations, such as annotating the AMD GPU driver as
depends on (CC_IS_GCC || CLANG_VERSION >=180000) || !(KASAN || KCSAN)This would effectively disable the AMDGPU driver for allmodconfig, which is somewhat unfortunate as it is an easy testing target.
Taking a step back, this is all being done because of a couple of warnings in the AMDGPU code. If fixing those in the source is too much effort (I did note [1] that GCC is at the current limit for that file even with Rodrigo's series applied [2]), couldn't we just take the existing workaround that this Makefile has for this file and its high stack usage and just extend it slightly for clang?
I personally don't mind fixing these issues in the driver, but the fact that they the creep back every time a new major version of Clang rolls out (that has been true for the past couple of years at the very least), makes it rather annoying to deal with.
diff --git a/drivers/gpu/drm/amd/display/dc/dml2/Makefile b/drivers/gpu/drm/amd/display/dc/dml2/Makefile index 66431525f2a0..fd49e3526c0d 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/Makefile +++ b/drivers/gpu/drm/amd/display/dc/dml2/Makefile @@ -58,7 +58,7 @@ endif endif ifneq ($(CONFIG_FRAME_WARN),0) -frame_warn_flag := -Wframe-larger-than=2048 +frame_warn_flag := -Wframe-larger-than=$(if $(CONFIG_CC_IS_CLANG),3072,2048) endif CFLAGS_$(AMDDALPATH)/dc/dml2/display_mode_core.o := $(dml2_ccflags) $(frame_warn_flag)
That would address the immediate concern of the warning breaking builds with CONFIG_WERROR=y while not raising the limit for other files in the kernel (just this one file in AMDGPU) and avoiding disabling the whole driver. The number could be lower, I think ~2500 bytes is the most usage I see with Rodrigo's series applied, so maybe 2800 would be a decent limit? Once there is a fix in the compiler, this expression could be changed to use clang-min-version or something of that sort.
Cheers, Nathan