Re: [PATCH kernel v3 2/3] KVM: SEV: Enable data breakpoints in SEV-ES

[Sean Christopherson <seanjc@google.com>]

On Wed, Feb 01, 2023, Sean Christopherson wrote:
> On Tue, Jan 31, 2023, Borislav Petkov wrote:
> > Hey Sean,
> > 
> > On Fri, Jan 20, 2023 at 02:10:46PM +1100, Alexey Kardashevskiy wrote:
> > > Prior to SEV-ES, KVM stored/loaded host debug registers upon switching
> > > to/from a VM. Changing those registers inside a running SEV VM
> > > triggered #VC exit to KVM.
> > > 
> > > SEV-ES added the encrypted state (ES) which uses an encrypted guest page
> > > for the VM state (VMSA). The hardware saves/restores certain registers on
> > > VMRUN/VMEXIT according to a swap type (A, B, C), see
> > > "Table B-3. Swap Types" in the AMD Architecture Programmer’s Manual
> > > volume 2.
> > > 
> > > AMD Milan (Fam 19h) introduces support for the debug registers swapping.
> > > DR6 and DR7 are always swapped. DR[0-3] and DR[0-3]_ADDR_MASK are swapped
> > > a type B when SEV_FEATURES[5] ("DebugSwap") is set.
> > > 
> > > Enable DebugSwap in VMSA. But only do so if CPUID Fn80000021_EAX[0]
> > > ("NoNestedDataBp", "Processor ignores nested data breakpoints") is
> > > supported by the SOC as otherwise a malicious SEV-ES guest can set up
> > > data breakpoints on the #VC IDT entry/stack and cause an infinite loop.
> > > 
> > > Eliminate DR7 and #DB intercepts as:
> > > - they are not needed when DebugSwap is supported;
> > > - #VC for these intercepts is most likely not supported anyway and
> > > kills the VM.
> > > Keep DR7 intercepted unless DebugSwap enabled to prevent the infinite #DB
> > > loop DoS.
> > 
> > ...
> > 
> > ok to take this through the tip tree?
> 
> I would prefer to take this through KVM, there's enough subtle complexity in this
> code that it'd be nice to have it close by.
> 
> If you're happy with patch 1, maybe ack that one and take it through KVM, and
> route patch 3 through tip?

Ah, you've already applied 1.  That works too.  I don't think KVM support for
DebugSwap is going to make v6.3 no matter who takes what, so just base on the
next version of this patch on tip/x86/cpu and I'll make a mental note to not try
to grab this until after v6.3-rc1.