[PATCH] D55423: [LLD][ELF] - A fix for "linker script assignment loses relative nature of section" bug.

[Peter Smith via Phabricator via llvm-commits]

peter.smith added a comment.

I quite like the idea of renaming maybeNeedsThunks(), for example addAddressDependentContent()? We could make sure that there is always at least one pass of assignAddresses done there.

The test case was derived from the linux kernel linker script for AArch64 which would have assignAddresses() called more than once due to AArch64 needing thunks. It is tempting to say just rearrange your linker script but this is not always easy as complex linker scripts are often machine generated. For example the kernel linker script is actually generated from https://github.com/torvalds/linux/blob/master/arch/arm64/kernel/vmlinux.lds.S using macros from https://github.com/torvalds/linux/blob/master/arch/arm64/kernel/image.h .

My understanding is that AArch64 (Arm64 in kernel parlance) is the priority target for using LLD to link the kernel, so I guess the Relocations.cpp would be sufficient on AArch64 and Arm. I don't know if the X86 kernel linker script would suffer from this problem, but it is certainly a possibility.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D55423/new/

https://reviews.llvm.org/D55423