[RFC PATCH 2/2] x86: Alternative <sys/platform/x86.h> implementation

[H.J. Lu]

On Sat, Dec 26, 2020 at 7:34 AM Florian Weimer <fw@deneb.enyo.de> wrote:
>
> * H. J. Lu via Libc-alpha:
>
> > We need to remove
> >
> > 000000000040ddd0  0000002f00000005 R_X86_64_COPY
> > 000000000040ddd0 __x86_cpu_array@GLIBC_2.33 + 0
>
> So the issue is that we initialize __x86_cpu_array in libc.so.6, but
> the active copy is actually in the main program?  And that copy reloc

Yes.

> is processed after IFUNC resolvers in libm.so.6 read __x86_cpu_array,
> observing zero data there?

You mentioned that the current <sys/platform.h> couldn't be used in
IFUNC resolver with BIND_NOW.  Do you have a testcase?

-- 
H.J.