Re: [PATCH v2 2/2] x86/bugs: Don't fill RSB on context switch with eIBRS

["Shah, Amit" <Amit.Shah@amd.com>]

On Fri, 2024-12-06 at 15:02 -0800, Josh Poimboeuf wrote:
> On Thu, Dec 05, 2024 at 04:53:03PM -0800, Josh Poimboeuf wrote:
> > On Thu, Dec 05, 2024 at 03:32:47PM -0800, Josh Poimboeuf wrote:
> > > On Thu, Nov 21, 2024 at 12:07:19PM -0800, Josh Poimboeuf wrote:
> > > > User->user Spectre v2 attacks (including RSB) across context
> > > > switches
> > > > are already mitigated by IBPB in cond_mitigation(), if enabled
> > > > globally
> > > > or if either the prev or the next task has opted in to
> > > > protection.  RSB
> > > > filling without IBPB serves no purpose for protecting user
> > > > space, as
> > > > indirect branches are still vulnerable.
> > > 
> > > Question for Intel/AMD folks: where is it documented that IBPB
> > > clears
> > > the RSB?  I thought I'd seen this somewhere but I can't seem to
> > > find it.
> > 
> > For Intel, I found this:
> > 
> >  
> > https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/advisory-guidance/post-barrier-return-stack-buffer-predictions.html
> > 
> >   "Software that executed before the IBPB command cannot control
> > the
> >   predicted targets of indirect branches executed after the command
> > on
> >   the same logical processor. The term indirect branch in this
> > context
> >   includes near return instructions, so these predicted targets may
> > come
> >   from the RSB.
> > 
> >   This article uses the term RSB-barrier to refer to either an IBPB
> >   command event, or (on processors which support enhanced IBRS)
> > either a
> >   VM exit with IBRS set to 1 or setting IBRS to 1 after a VM exit."
> > 
> > I haven't seen anything that explicit for AMD.
> 
> Found it.  As Andrew mentioned earlier, AMD IBPB only clears RSB if
> the
> IBPB_RET CPUID bit is set.  From APM vol 3:
> 
> CPUID Fn8000_0008_EBX Extended Feature Identifiers:
> 
> 30	IBPB_RET	The processor clears the return address
> 			predictor when MSR PRED_CMD.IBPB is written
> to 1.
> 
> We check that already for the IBPB entry mitigation, but now we'll
> also
> need to do so for the context switch IBPB.
> 
> Question for AMD, does SBPB behave the same way, i.e. does it clear
> RSB
> if IBPB_RET?

That's correct - SBPB clears the RSB only when IBPB_RET is set.

		Amit