Re: [RFC PATCH 00/11] Kernel FineIBT Support

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Apr 20, 2022 at 03:40:41PM -0700, Joao Moreira wrote:
> > > 
> > > If FineIBT needs it, I could reconsider.  But I think there's a strong
> > > case to be made that the linker should be doing that instead.
> > 
> > That sounds reasonable to me (and reminds me of linker relaxation).
> > Joao, can you please work with Fangrui (LLD) and HJ (GNU binutils) to
> > determine how feasible this would be? I assume code outside the kernel
> > might enjoy such an optimization, too.  When that's the case, then it
> > probably makes more sense to "upstream" such "optimizations" from the
> > kernel-specific objtool into the toolchains.
> 
> Alright, these are the greenlights I was hoping for.
> 
> I went quickly into this with HJ and he mentioned that it should be doable
> in the linker, and that he has a patch for it in gcc (for local function,
> from what I could see):
> https://gcc.gnu.org/pipermail/gcc-patches/2022-February/590832.html
> 
> If @Fangrui is fine with it, I would like to try implementing this myself in
> lld (I'm still learning a lot about lld and having an actual problem to
> solve is the kind of fuel I need). Should take me a while, but I think this
> is not urgent, right? I can also go ahead and replicate HJ's gcc patch into
> clang, so we can also handle the local functions within the compiler (I
> think this makes a lot of sense).
> 
> Once we have these in, I'll revisit FineIBT and extend the features to
> handle the FineIBT instrumentation. Hopefully we'll be released from needing
> objtool (famous last words?!).
> 
> This sounds like a plan, but I'm ofc open to suggestions or different
> ideas/plans.

So trivially the plan sounds like: compiler fixes STB_LOCAL because it
has the scope, and the linker fixes everything else. However, that seems
to assume that !STB_LOCAL will have ENDBR.

This latter isn't true; for one there's __attribute__((nocf_check)) that
can be used to suppress ENDBR generation on a function.

Alternatively the linker will need to 'read' the function to determine
if it has ENDBR, or we need to augment the ELF format such that we can
tell from that.

So what exactly is the plan?