cfi_label and cfi_startproc in start.S

[alexfanqi]

Hi,

Can I get some help to clarify the .cfi_label dummy directive here? In ziglang/zig#3340, zig vendors part of glibc header, and it fails to compile because llvm does not support .cfi_label.

The thing confuses me is the macro ENTRY already includes a cfi_startproc, which I thought should start the FDE and makes cfi_label redundant.

I have also read Bug 23125 that introduces this line. The original patch in first comment didn't include cfi_label.

Is it safe to simply remove cfi_label, or there are other workarounds?

[aswaterman]

I know this is not an answer to your question, but note that RISC-V is not alone in using the .cfi_label directive in this context; see https://github.com/bminor/glibc/search?q=cfi_label

As a matter of curiosity, I'd also like to understand why this directive is necessary.

[TommyMurphyTM1234]

Shouldn't this issue be logged against the upstream glibc and not here?

Anyway, doesn't this explain it?

bminor/glibc@2dd12ba

If not then presumably the author of that commit (@andreas-schwab) can explain further?

[alexfanqi]

Anyway, doesn't this explain it?

bminor/glibc@2dd12ba

I read this commit message before. It explains the reason for undefining ra and replacing tail, but doesn't mention cfi_label. Only comment I can find is cfi_label forces start of fde, which seems already done by cfi_startproc.

RISC-V is not alone in using the .cfi_label directive in this context; see https://github.com/bminor/glibc/search?q=cfi_label

I noticed all arches using cfi_label come after riscv, so I tried to ask in riscv community first. I will lodge a bug to upstream while waiting for more help here.

[TommyMurphyTM1234]

I will lodge a bug to upstream while waiting for more help here.

Can you post a link here to the upstream bug please?

[alexfanqi]

Sorry, upstream bug https://sourceware.org/bugzilla/show_bug.cgi?id=29749

[TommyMurphyTM1234]

Sorry, upstream bug https://sourceware.org/bugzilla/show_bug.cgi?id=29749

Thanks @alexfanqi

[TommyMurphyTM1234]

This post seems to address the original question:

https://sourceware.org/bugzilla/show_bug.cgi?id=29749#c2

As such, can this issue be closed?

[alexfanqi]

still want to check some details of cfi_label, but since the discussion is in upstream bugzilla. I think this issue can be closed.

[TommyMurphyTM1234]

still want to check some details of cfi_label, but since the discussion is in upstream bugzilla. I think this issue can be closed.

Thanks @alexfanqi