Web lists-archives.com

Re: [PATCH] x86/retpoline: Fill RSB on context switch for affected CPUs




On Fri, 2018-01-12 at 10:02 -0800, Andi Kleen wrote:
> > +     if ((!boot_cpu_has(X86_FEATURE_PTI) &&
> > +          !boot_cpu_has(X86_FEATURE_SMEP)) || is_skylake_era()) {
> > +             setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
> > +             pr_info("Filling RSB on context switch\n");
> 
> We need to do more things for Skylake (like idle and interrupt fill
> and possibly deep call cahin), so I don't think it makes sense to
> 
> - have an individual flag for each of these. It can be just a single
> flag that enables all of this for Skylake
> 
> - print something for each of them. that will just be very noisy
> without any useful benefit to the user.

I still think we are better off using IBRS by default on Skylake.

This patch wasn't really for Skylake; the real use case was for AMD
CPUs (!PTI) without SMEP. Since it happens to needed on Skylake too we
might as well enable it there... but that doesn't mean I was planning
to do all the other horrible crap we need for Skylake.

Attachment: smime.p7s
Description: S/MIME cryptographic signature