On Wed, 2018-01-10 at 12:45 +0100, Borislav Petkov wrote:
> On Wed, Jan 10, 2018 at 11:36:41AM +0000, David Woodhouse wrote:
> > 
> > That fixed and understood, I shall remove the offending NOPs anyway,
> > because aligning instructions in the *altinstr* section is entirely
> > pointless as they might still not be aligned when they get copied into
> > place anyway.
>
> Yap. It was still a good experiment because we found this shortcoming!

Don't suppose you want to make the alignment actually *work*? :)

Paul, how much of a win was it really? And not just in a microbenchmark
of the retpoline itself, saying "look Ma! This goes faster if I make it
take three times as many icache lines!", but overall system benchmarks?
:)