Re: [PATCH] x86: Generate PLT relocations for -z now

[Carlos O'Donell <carlos at redhat dot com>]

On 05/11/2017 05:41 AM, Szabolcs Nagy wrote:
> i think this is not a binutils issue,
> gcc can do an optimization to load
> foo from the got and call it indirectly
> and later return it, then the linker
> has no chance to emit plt reloc for foo.

I do not disagree with you.

Function pointers are a blind spot in PLT-based auditing, and
some day we might be able to solve that.

That does not mean we should continue to erode that support
without having a discussion.

If we are going to deprecate LD_AUDIT, LD_PROFILE, and ltrace,
what do we replace them with? Are those replacements mature
technology?

Keep in mind that LD_AUDIT has worked fora long time and it's
only in binutils 2.26 where we've started to introduce
non-optional optimizations that might break things.

> this is why i said plt should not be
> considered part of the abi because it's
> unreliable anyway
That is not true. You need to think big and consider yourself
part of a large GNU community. We are not isolated silos.

The GNU implementation includes a compiler, linker, and dynamic
loader, and all of these parts can collude to provide a better
than average developer experience.

We made PLTs part of the implementation by implementing LD_AUDIT, 
LD_PROFILE, and by supporting ltrace.

-- 
Cheers,
Carlos.