https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121572
--- Comment #10 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Sam James from comment #7)
> Created attachment 62135 [details]
> pr121572-elfutils.tar.xz
>
> At r16-3190-g5cf1b9a03ec5b6, I noticed some bugginess with consumers of
> elfutils/libelf (libdwfl) error handling. With your patch (the one posted
> last night), I get an ICE with elfutils at a different place to Sergei's
> above testcase.
>
> I didn't clean it up, cvise only now finished. I'll try the new patch.
>
> But for now:
> ```
> + gcc -O3 -flto=auto -fPIC -o libdw.so -shared -o libdw.so
> -Wl,--version-script=libdw.map -Wl,--whole-archive dwarf_error.os
> dwfl_begin.os dwfl_error.os argp-std.os -Wl,--no-whole-archive
> + cat x.txt
> argp-std.i:13:6: warning: type of ‘dwfl_begin’ does not match original
> declaration [-Wlto-type-mismatch]
> 13 | int *dwfl_begin(Dwfl_Callbacks *);
> | ^
> dwfl_begin.i:9:6: note: type mismatch in parameter 1
> 9 | int *dwfl_begin() { __libdwfl_seterrno(DWFL_E_LIBELF); }
> | ^
> dwfl_begin.i:9:6: note: ‘dwfl_begin’ was previously declared here
> during RTL pass: x86_cse
> argp-std.i: In function ‘parse_opt’:
> argp-std.i:29:1: internal compiler error: in ix86_place_single_tls_call, at
> config/i386/i386-features.cc:3873
> 29 | }
> | ^`
> ```
I can't reproduce it with the updated v2 patch.
