Re: [RFC][PATCH 07/17] x86/cpu/kvm: Provide UNTRAIN_RET_VM

[Peter Zijlstra]

On Wed, Aug 09, 2023 at 10:30:00AM -0400, Josh Poimboeuf wrote:
> On Wed, Aug 09, 2023 at 04:06:44PM +0200, Peter Zijlstra wrote:
> > On Wed, Aug 09, 2023 at 09:50:04AM -0400, Josh Poimboeuf wrote:
> > > On Wed, Aug 09, 2023 at 09:12:25AM +0200, Peter Zijlstra wrote:
> > > > With the difference being that UNTRAIN_RET_VM uses
> > > > X86_FEATURE_IBPB_ON_VMEXIT instead of X86_FEATURE_ENTRY_IBPB.
> > > > 
> > > > This cures VMEXIT doing potentially unret+IBPB or double IBPB.
> > > > 
> > > > Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> > > > ---
> > > >  arch/x86/include/asm/nospec-branch.h |   11 +++++++++++
> > > >  arch/x86/kernel/cpu/bugs.c           |   17 ++++++++++++++++-
> > > >  arch/x86/kvm/svm/vmenter.S           |    7 ++-----
> > > >  3 files changed, 29 insertions(+), 6 deletions(-)
> > > > 
> > > > --- a/arch/x86/include/asm/nospec-branch.h
> > > > +++ b/arch/x86/include/asm/nospec-branch.h
> > > > @@ -299,6 +299,17 @@
> > > >  #endif
> > > >  .endm
> > > >  
> > > > +.macro UNTRAIN_RET_VM
> > > > +#if defined(CONFIG_CPU_UNRET_ENTRY) || defined(CONFIG_CPU_IBPB_ENTRY) || \
> > > > +	defined(CONFIG_CALL_DEPTH_TRACKING) || defined(CONFIG_CPU_SRSO)
> > > 
> > > Maybe can be simplified?
> > > 
> > 
> > See patches 9 and 10 :-)
> 
> Can still be further simplified to CONFIG_RETHUNK?

Hmm, probably, but my brain seems to be giving out. Let me do the whole
dinner and kids to bed thing and I'll look again later.