Re: [PATCH] elf: Update GNU_PROPERTY_X86_XXX macros [BZ #23797]

[Mark Wielaard <mark at klomp dot org>]

Hi,

On Wed, 2018-10-24 at 13:03 +0200, Mark Wielaard wrote:
> Some questions about how to interpret the new constant values vs the
> old ones. I am not sure I understand how this is fully backwards
> compatible.

Currently the elfutils code just handles GNU_PROPERTY_STACK_SIZE,
GNU_PROPERTY_NO_COPY_ON_PROTECTED and GNU_PROPERTY_X86_FEATURE_1_AND
with IBT and SHSTK. But none of the other types/data that would be
renamed/renumbered with this patch.

It would be really helpful if you could answer the questions below to
better understand how this new data fits in and what the backwards
compatibility story is.

Thanks,

Mark

> On Fri, 2018-10-19 at 15:47 -0700, H.J. Lu wrote:
> > GNU_PROPERTY_X86_FEATURE_1_AND is
> > unchanged.  GNU_PROPERTY_X86_ISA_1_USED
> > and GNU_PROPERTY_X86_FEATURE_2_USED are redefined to better support
> > targeted processors since GNU_PROPERTY_X86_ISA_1_?86 aren't isn't
> > very
> > useful.  A new set of GNU_PROPERTY_X86_ISA_1_XXX bits are
> > defined.  The
> > previous GNU_PROPERTY_X86_ISA_1_XXX macros are removed.
> 
> So GNU_PROPERTY_X86_FEATURE_1_AND is kept at the same constant value,
> but now falls into the GNU_PROPERTY_X86_UINT32_AND_XXX range which
> has
> specific semantics for how linkers combine the note data. Isn't this
> a
> backwards compatible issue? There is no guarantee the note data was
> constructed by a linker that knows about this, but there is no way to
> detect that?
> 
> GNU_PROPERTY_X86_ISA_1_USED and GNU_PROPERTY_X86_FEATURE_2_USED are
> renumbered. And the the GNU_PROPERTY_X86_ISA_1_XXX macros are
> renamed,
> but some are kept the same with new constant values. Were the old
> constants/names not used at all?
> 
> > +/* Set by linker to indicate that the property is valid.  */
> > +#define GNU_PROPERTY_X86_UINT32_VALID		(1U << 31)
> 
> This isn't mentioned in your commit message, nor is it mentioned in
> the
> ChangeLog entry. Should it be added? And if so, how does it interact
> with the interpretation of GNU_PROPERTY_X86_FEATURE_1_AND generated
> by
> older linkers or interpreted by older loaders (does it have to be
> checked by newer loaders?)
> 
> Thanks,
> 
> Mark