Re: [PATCH v2 0/8] x86/PV: avoid speculation abuse through guest accessors

[Ian Jackson <iwj@xxxxxxxxxxxxxx>]

Jan Beulich writes ("Re: [PATCH v2 0/8] x86/PV: avoid speculation abuse through 
guest accessors"):
> On 19.02.2021 17:13, Ian Jackson wrote:
> > Jan Beulich writes ("Re: [PATCH v2 0/8] x86/PV: avoid speculation abuse 
> > through guest accessors"):
> > I think 4-6 and 8 are good candidates for the reasons you give, and
> > because they seem low risk to me.  Have you used any automatic
> > techniques to check that there is no unintentional codegen change ?
> > (Eg, binary diffs, diffing sedderied versions, or something.)
> 
> I did some manual inspection at the time of putting together that
> work, but nothing further to be honest.

I think that something automatic might be worthwhile, but I would like
an opinion from another hypervisor maintainer about the level of risk
posed by the possibility of manual slips.  Eg, how likely it would be
for the compiler to catch them.

> > To my naive eye patch 7 looks scary because it might be moving the
> > scope of a critical section.  Am I wrong about that ?
> 
> At the source level it moves things, yes. Generated code, again as
> per manual inspection, doesn't change, due to the pieces that the
> compiler is able to eliminate. So I guess it's not as scary as it
> may look.

Oh, you eyeballed the generated code ?  Cool.

Ian.