LLD silently misinterprets the LMA in linkerscripts

[andrewrk]

Bugzilla Link	45078
Resolution	FIXED
Resolved on	Apr 13, 2020 16:11
Version	unspecified
OS	All
Blocks	#44654
CC	@zmodem,@MaskRay,@smithp35,@tstellar

Extended Description

I'm filing this in case nobody pays attention to github issues
#163

[andrewrk]

assigned to @MaskRay

[smithp35]

I don't know whether anyone is actively looking at github issues. My reading of llvm-dev was that there wasn't any agreement on whether to use them or not. Anyhow definitely worth raising a PR as well.

Can you check master lld. I think that https://reviews.llvm.org/D74297 should prevent at least the worst case behaviour seen in the PR. There is still only one PHDR created, but that is probably because it is empty.

[MaskRay]

Already fixed by https://reviews.llvm.org/D74297 (will be included in lld 11)

[zmodem]

Should this be considered for 10.0.1? (Reopening for now if that's the case)

[MaskRay]

Should this be considered for 10.0.1? (Reopening for now if that's the case)

It should be safe to merge. It is ok if the original reporter (unfortunately https://github.com/llvm/llvm-project/issues seems to have been deleted) or Andrew is strong about this issue.

[andrewrk]

Having the fix in 10.0.1 would be nice. Either way is fine with me.

[MaskRay]

Merging D74297 is tricky. The test file it touched has been touched several other unmerged commits. I am a bit confident that we have achieved very good compatibility with GNU ld in terms of LMA semantics after https://reviews.llvm.org/D76995

We probably should not merge D74297 into 10.0.1

[tstellar]

mentioned in issue #44654