On Fri, 2018-01-05 at 09:28 -0800, Linus Torvalds wrote:
> 
> Yes, I would suggest against expecting altinstructions to have
> relocation information. They are generated in a different place, so..
> 
> That said, I honestly like the inline version (the one that is in the
> google paper first) of the retpoline more than the out-of-line one.
> And that one shouldn't have any relocagtion issues, because all the
> offsets are relative.
> 
> We want to use that one for the entry stub anyway, can't we just
> standardize on that one for all our assembly?

Sure, that's just a case of tweaking the macros a little to do it
inline instead of jumping to the thunk. I thunk judicious use of
__stringify() has resolved any issues I might have had with that
approach in the first place. However...

> If the *compiler* uses the out-of-line version, that's a separate
> thing. But for our asm cases, let's just make it all be the inline
> case, ok?
> 
> It also should simplify the whole target generation. None of this
> silly "__x86.indirect_thunk.\reg" crap with different targets for
> different register choices.

If we're going to let the compiler use the out-of-line version (which
we *want* to because otherwise we don't get to ALTERNATIVE it away as
appropriate), then we still need to emit the various
__x86.indirect_thunk.\reg thunks for the compiler to use anyway.

At that point, there isn't a *huge* benefit to doing the retpoline
inline in our own asm, when it might as well just be a jump to the
thunks which exist anyway. But neither do I have an argument *against*
doing so, so I'll happily tweak the NOSPEC_CALL/NOSPEC_JMP macros
accordingly if you really want...