Assembly options not recognized when using integrated assembler

[rengolin]

Bugzilla Link	18689
Resolution	WONTFIX
Resolved on	Feb 21, 2014 05:20
Version	trunk
OS	Linux
Blocks	#19300
CC	@walkerkd,@tinti

Extended Description

While compiling the Linux kernel with the IAS, we get this two parameter errors:

/llvmlinux/toolchain/clang/install/bin/clang -gcc-toolchain
/llvmlinux/arch/arm/toolchain/codesourcery/arm-2013.05
-Wp,-MD,arch/arm/vfp/.vfphw.o.d -nostdinc -isystem
/llvmlinux/toolchain/clang/install/bin/../lib/clang/3.5/include
-I/llvmlinux/targets/vexpress/src/linux/arch/arm/include
-Iarch/arm/include/generated -Iinclude
-I/llvmlinux/targets/vexpress/src/linux/arch/arm/include/uapi
-Iarch/arm/include/generated/uapi
-I/llvmlinux/targets/vexpress/src/linux/include/uapi -Iinclude/generated/uapi
-include /llvmlinux/targets/vexpress/src/linux/include/linux/kconfig.h
-D__KERNEL__ -target arm-none-linux-gnueabi -Qunused-arguments
-Wno-unknown-warning-option -D__ASSEMBLY__ -integrated-as -mfpu=vfp
-funwind-tables -marm -D__LINUX_ARM_ARCH__=7 -march=armv7-a
-include asm/unified.h -Wa,-mfpu=softvfp+vfp -mfloat-abi=soft -Wa,--gdwarf-2
-c -o arch/arm/vfp/vfphw.o arch/arm/vfp/vfphw.S

clang-3.5: error: unsupported argument '-mfpu=softvfp+vfp' to option 'Wa,'
clang-3.5: error: unsupported argument '--gdwarf-2' to option 'Wa,'

The easiest solution would be to add them as alias... But the dwarf2 worries me, since, AFAIK, we can't guarantee it.

[rengolin]

assigned to @rengolin

[rengolin]

Explanations about the vfp choices:

http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0472c/BABDGGBI.html

About Dwarf2, I'd rather change that in the Make files, as we don't support Dwarf2 any more and accepting that flag would be an error.

[walkerkd]

Also the clang option is currently -gdwarf-2 and not --gdwarf-2, hence the unsupported argument error.

It appears that GNU as accepts both versions of the option; but interestingly it accepts -gdwarf-3 but not --gdwarf-3.

But I agree that as we don't support DWARF-2 the Makefile should be changed to at least -gdwarf-3

[rengolin]

Also the clang option is currently -gdwarf-2 and not --gdwarf-2, hence the unsupported argument error.

Right, so this is a bug, since the standard now is to use two dashes, not one.

Dwarf 2 support should remain untouched and if debugging the kernel breaks, they've been warned! ;)

[rengolin]

I think Keith is right, we should use -dwarf-2 on both compilers, especially with GCC only accepting -dwarf-3.

Vinicius/Behan, do you think such a patch would be easy to maintain? I think, given that it will work with GCC, upstreaming it would be a no-brainer.

cheers,
--renato

[rengolin]

Proposal for softvfp implementation:

http://llvm-reviews.chandlerc.com/D2817

[rengolin]

The dwarf-2 issue, I hear, has already been changed in the LLVMLinux and Linaro's kernel trees, so we can consider it "fixed".

[rengolin]

I have had long discussions about softvfp+vfp and the conclusion is this:

• In GNU tools -mfpu=fp means: use FP AND change to hard-float ABI.
• The original meaning of softvfp+vfp was that it should use FP
  instructions (but VFP instead of the, then default, FPA) but not
  change the float ABI.

Since we don't support FPA instructions in LLVM, and GCC will not emit
them by default, and we now have -mfloat-abi=softfp, the flag is no
longer relevant and we can simply remove it.

The only reason it might still be in, as Ramana pointed out, is to
work around very old binutils bugs regarding build attributes and
linking soft+hard objects together. In the case where removing the
flag introduces undesirable behaviour, we should fix that behaviour
instead, not continue using an ad-hoc outdated and deprecated flag.

[rengolin]

mentioned in issue #19300