In D102912#2775303, @kitaisreal wrote:

/usr/bin/clang-11 --version
Ubuntu clang version 11.0.0-2~ubuntu20.04.1

Seems it is fixed in 11.0.1 and trunk or at least I was not able to reproduce the crash, codegen looks fine.

LGTM just please run once more git clang-format HEAD^1 on the patch.

address review comments.

In D89570#2749259, @MaskRay wrote:

In D89570#2744598, @danielkiss wrote:

In D89570#2740272, @MaskRay wrote:

ARM EHABI is very different from the Itanium implementation. Someone should move the implementation to a separate file.

Sorry, do you mean the content of the libunwind/include/unwind.h ?

I mean all the #if defined(_LIBCXXABI_ARM_EHABI) branches make the file difficult to read. reduce sharing and move the implementation to a separate file.
Many sanitizer files are organized this way.

Adding a fronted flag and tests.
comdat is still the default, feels fine but I'm not sure.

@jroelofs Thanks for the review, all addressed.

In D89570#2740272, @MaskRay wrote:

ARM EHABI is very different from the Itanium implementation. Someone should move the implementation to a separate file.

Sorry, do you mean the content of the libunwind/include/unwind.h ?

Rebased, retested on ToT.

Wondering if AArch64InstrInfo::getOutliningType and the outline should do the same because it might emit PACIASP.
see the old change:
https://reviews.llvm.org/D77565?id=256025#change-SC49e8S7FnMX

Ah, right. EMITBKEY should be a meta one but it seems as a problematic change.
Let's simplify this now and hope no other need to be added too.

LGTM

see D99417, way different approach is needed to reduce the number of landing pads.

Implementing the review comment.

fix for lint

The current think is that the unstripped value of the return addresses is not too problematic.
Let's keep it simple now.

In D98008#2614592, @kubamracek wrote:

Keep the PAC in the stack trace and only strip it when necessary like comparing traces, looking up symbols.

Interesting, I'm not saying it's necessarily a bad idea, but could you explain the motivation? The previous approach (stripping the stack trace PCs at the point of capturing stack trace) seemed simpler, easier to implement and presumably faster (less stripping operations overall).

Storing the PAC protected return addresses sounds safer because the values in trace can't be used directly as jump addresses. Assuming some might use sanitisers in production.

Keep the PAC in the stack trace and only strip it when necessary like comparing traces, looking up symbols.
Moved the strip to the sanitizer_ptrauth.h.

In D98008#2607313, @kubamracek wrote:

Could we reuse the macros/builtins from https://github.com/llvm/llvm-project/blob/main/compiler-rt/lib/sanitizer_common/sanitizer_ptrauth.h instead?

Could you add [libunwind] to the beginning of title?
but otherwise LGTM as the author of the corresponding dwarf spec change.

LGTM, Thanks.

Reenables tests ( revert of a4fa667dee60 )

@MaskRay Thanks for the test case.

LGTM, thanks.

rebased, fixed my own comments.

In D93428#2461343, @apazos wrote:

Thanks Daniel for the explanation.

No problem at all.

Is there a reason why assembly files have a different flag (i.e. -mmark-bti-property) to create the .note.gnu.property with the BTI entry?

The .note.gnu.property is already generated when C/C++ files are compiled with -mbranch-protection=bti.
-mmark-bti-property is only for assembly file where the .note.gnu.property should be added manually otherwise.

@manojgupta Thanks for the feedback, let me know if you have updates.

unwind_leaffunction.pass.cpp could be enable too for aarch64.

LGTM

LGTM, Thanks.

Done see: D91182

LGTM, Thanks!

https://bugs.llvm.org/show_bug.cgi?id=47142 this effects this. the missing unwind info make forceunwind fragile.

self.cxx has a function to test compiler flags addFlagIfSupported but there is no support for linker flag (yet).

rebase, add MINGW32

ping

Add a simple test and fixes.
This version passes on arm32 - armv7l-unknown-linux-gnueabihf -

_Unwind_Stop_Fn shall not have different signature for EHABI.