Re: [PATCH 1/3] x86: Initialize CPU info via IFUNC relocation [BZ 26203]

["H.J. Lu" <hjl.tools@gmail.com>]

On Fri, Sep 18, 2020 at 1:06 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu via Libc-alpha:
>
> > X86 CPU features in ld.so are initialized by init_cpu_features, which is
> > invoked by DL_PLATFORM_INIT from _dl_sysdep_start.  But when ld.so is
> > loaded by static executable, DL_PLATFORM_INIT is never called.
>
> I don't think that's accurate.  It's called from elf/dl-support.c in the
> static case.  But I agree that it's too late in this case.

I was referring to the ld.so case.  elf/dl-support.c isn't used for ld.so.

> > Also x86 cache info in libc.o and libc.a is initialized by a
> > constructor which may be called too late.  Instead, we should
> > initialize x86 CPU features and cache info by initializing dummy
> > function pointers via IFUNC relocation.
>
> Why would ld put this special IRELATIVE relocation first?  I'm sorry,
> but this looks very brittle.

This is for ld.so.  ld.so doesn't assume that IFUNC is usable before everything
is setted up properly.   This IRELATIVE relocation is not different from other
dynamic relocations,  IRELATIVE or not.  The key here is that we want to call
init_cacheinfo as early as possible.  Calling from _dl_relocate_object is as
early as we can get.

> I think there's an existing mechanism for this corner case,
> ARCH_SETUP_IREL.  Could you use that?
>

ARCH_SETUP_IREL is for static executable, not ld.so.

-- 
H.J.