[PATCH] x86: Support GNU_PROPERTY_X86_ISA_1_V[234] marker [BZ #26717]

[H.J. Lu]

On Wed, Oct 28, 2020 at 4:08 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu:
>
> >> It's not clear based on this description if the expectation is that the
> >> search for alternative implementations continues after an attempt to
> >> load an object fails due to an architecture level mismatch.
> >>
> >
> > 1. ldconfig should issue an error for shared objects with mismatched ISA
> > level and mark such shared objects in ld.so cache.
>
> But what is the algorithm for that?

The algorithm should be the same as handling EM_386 shared objects in
x86-64 directories.

> > 2. ld.so should skip shared objects with incompatible ISA level similar to
> > EM_386 shared objects.   If ld.so detects it after loading such a shared
> > object, ld.so should issue a hard error.
>
> These two sentences seem to contradict each other.  Do you mean that
> with a relative path/soname, ld.so should continue searching?  But if it
> is not searching at all because ld.so was passed an absolute path, it
> should produce an error?

If a shared object is marked as incompatible in ld.so cache, ld.so should
skip it.  Otherwise ld.so should issue an error when detecting a loaded
shared object is incompatible.


-- 
H.J.