[RFC PATCH v3 0/6] Direct Map Removal for guest_memfd

[Kaplan, David]

[AMD Official Use Only - AMD Internal Distribution Only]

> -----Original Message-----
> From: Sean Christopherson <seanjc at google.com>
> Sent: Friday, November 1, 2024 10:18 AM
> To: Derek Manwaring <derekmn at amazon.com>
> Cc: roypat at amazon.co.uk; ackerleytng at google.com;
> agordeev at linux.ibm.com; aou at eecs.berkeley.edu;
> borntraeger at linux.ibm.com; bp at alien8.de; catalin.marinas at arm.com;
> chenhuacai at kernel.org; corbet at lwn.net; dave.hansen at linux.intel.com;
> david at redhat.com; gerald.schaefer at linux.ibm.com; gor at linux.ibm.com;
> graf at amazon.com; hca at linux.ibm.com; hpa at zytor.com;
> jgowans at amazon.com; jthoughton at google.com; kalyazin at amazon.com;
> kernel at xen0n.name; kvm at vger.kernel.org; linux-arm-
> kernel at lists.infradead.org; linux-doc at vger.kernel.org; linux-
> kernel at vger.kernel.org; linux-kselftest at vger.kernel.org; linux-
> mm at kvack.org; linux-riscv at lists.infradead.org; linux-s390 at vger.kernel.org;
> linux-trace-kernel at vger.kernel.org; loongarch at lists.linux.dev;
> luto at kernel.org; mathieu.desnoyers at efficios.com; mhiramat at kernel.org;
> mingo at redhat.com; palmer at dabbelt.com; paul.walmsley at sifive.com;
> pbonzini at redhat.com; peterz at infradead.org; quic_eberman at quicinc.com;
> rostedt at goodmis.org; rppt at kernel.org; shuah at kernel.org;
> svens at linux.ibm.com; tabba at google.com; tglx at linutronix.de;
> vannapurve at google.com; will at kernel.org; x86 at kernel.org;
> xmarcalx at amazon.com; Kaplan, David <David.Kaplan at amd.com>
> Subject: Re: [RFC PATCH v3 0/6] Direct Map Removal for guest_memfd
>
> Caution: This message originated from an External Source. Use proper
> caution when opening attachments, clicking links, or responding.
>
>
> +David Kaplan
>
> On Thu, Oct 31, 2024, Derek Manwaring wrote:
> > On 2024-10-31 at 10:42+0000 Patrick Roy wrote:
> > > On Thu, 2024-10-31 at 09:50 +0000, David Hildenbrand wrote:
> > > > On 30.10.24 14:49, Patrick Roy wrote:
> > > >> Most significantly, I've reduced the patch series to focus only
> > > >> on direct map removal for guest_memfd for now, leaving the whole
> > > >> "how to do non-CoCo VMs in guest_memfd" for later. If this
> > > >> separation is acceptable, then I think I can drop the RFC tag in
> > > >> the next revision (I've mainly kept it here because I'm not
> > > >> entirely sure what to do with patches 3 and 4).
> > > >
> > > > Hi,
> > > >
> > > > keeping upcoming "shared and private memory in guest_memfd" in
> > > > mind, I assume the focus would be to only remove the direct map for
> private memory?
> > > >
> > > > So in the current upstream state, you would only be removing the
> > > > direct map for private memory, currently translating to
> "encrypted"/"protected"
> > > > memory that is inaccessible either way already.
> > > >
> > > > Correct?
> > >
> > > Yea, with the upcomming "shared and private" stuff, I would expect
> > > the the shared<->private conversions would call the routines from
> > > patch 3 to restore direct map entries on private->shared, and zap
> > > them on
> > > shared->private.
> > >
> > > But as you said, the current upstream state has no notion of "shared"
> > > memory in guest_memfd, so everything is private and thus everything
> > > is direct map removed (although it is indeed already inaccessible
> > > anyway for TDX and friends. That's what makes this patch series a
> > > bit awkward :( )
> >
> > TDX and SEV encryption happens between the core and main memory, so
> > cached guest data we're most concerned about for transient execution
> > attacks isn't necessarily inaccessible.
> >
> > I'd be interested what Intel, AMD, and other folks think on this, but
> > I think direct map removal is worthwhile for CoCo cases as well.
>
> Removal of the direct map entries for guest private PFNs likely won't affect
> the ability of an attacker to glean information from the unencrypted data
> that's in the CPU caches, at least not on x86.  Both TDX and SEV steal physical
> address
> bit(s) for tagging encrypted memory, and unless things have changed on
> recent AMD microarchitectures (I'm 99.9% certain Intel CPUs haven't
> changed), those stolen address bits are propagated into the caches.  I.e. the
> encrypted and unencrypted forms of a given PFN are actually two different
> physical addresses under the hood.
>
> I don't actually know how SEV uses the stolen PA bits though.  I don't see
> how it simply be the ASID, because IIUC, AMD CPUs allow for more unique
> SEV-capable ASIDs than uniquely addressable PAs by the number of stolen
> bits.  But I would be very surprised if the tag for the cache isn't guaranteed to
> be unique per encryption key.
>
> David?

How the stolen PA bits are used is a microarchitectural implementation detail.  It is true that the tag will be unique per encryption key.  Beyond that, I'm not sure what other details are relevant to SW.

--David Kaplan