Re: [PATCH v9 1/5] KVM: arm64: Add a vcpu flag to control ptrauth for guest

[Dave Martin]

On Wed, Apr 17, 2019 at 04:54:32PM +0100, Marc Zyngier wrote:
> On 17/04/2019 15:52, Dave Martin wrote:
> > On Wed, Apr 17, 2019 at 03:19:11PM +0100, Marc Zyngier wrote:
> >> On 17/04/2019 14:08, Amit Daniel Kachhap wrote:
> >>> Hi,
> >>>
> >>> On 4/17/19 2:05 PM, Marc Zyngier wrote:
> >>>> On 12/04/2019 04:20, Amit Daniel Kachhap wrote:
> >>>>> A per vcpu flag is added to check if pointer authentication is
> >>>>> enabled for the vcpu or not. This flag may be enabled according to
> >>>>> the necessary user policies and host capabilities.
> >>>>>
> >>>>> This patch also adds a helper to check the flag.
> >>>>>
> >>>>> Signed-off-by: Amit Daniel Kachhap <amit.kachhap@xxxxxxx>
> >>>>> Cc: Mark Rutland <mark.rutland@xxxxxxx>
> >>>>> Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
> >>>>> Cc: Christoffer Dall <christoffer.dall@xxxxxxx>
> >>>>> Cc: kvmarm@xxxxxxxxxxxxxxxxxxxxx
> >>>>> ---
> >>>>>
> >>>>> Changes since v8:
> >>>>> * Added a new per vcpu flag which will store Pointer Authentication enable
> >>>>>    status instead of checking them again. [Dave Martin]
> >>>>>
> >>>>>   arch/arm64/include/asm/kvm_host.h | 4 ++++
> >>>>>   1 file changed, 4 insertions(+)
> >>>>>
> >>>>> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> >>>>> index 9d57cf8..31dbc7c 100644
> >>>>> --- a/arch/arm64/include/asm/kvm_host.h
> >>>>> +++ b/arch/arm64/include/asm/kvm_host.h
> >>>>> @@ -355,10 +355,14 @@ struct kvm_vcpu_arch {
> >>>>>   #define KVM_ARM64_HOST_SVE_ENABLED	(1 << 4) /* SVE enabled for EL0 */
> >>>>>   #define KVM_ARM64_GUEST_HAS_SVE		(1 << 5) /* SVE exposed to guest */
> >>>>>   #define KVM_ARM64_VCPU_SVE_FINALIZED	(1 << 6) /* SVE config completed */
> >>>>> +#define KVM_ARM64_GUEST_HAS_PTRAUTH	(1 << 7) /* PTRAUTH exposed to guest */
> >>>>>   
> >>>>>   #define vcpu_has_sve(vcpu) (system_supports_sve() && \
> >>>>>   			    ((vcpu)->arch.flags & KVM_ARM64_GUEST_HAS_SVE))
> >>>>>   
> >>>>> +#define vcpu_has_ptrauth(vcpu)	\
> >>>>> +			((vcpu)->arch.flags & KVM_ARM64_GUEST_HAS_PTRAUTH)
> >>>>> +
> >>>>
> >>>> Just as for SVE, please first check that the system has PTRAUTH.
> >>>> Something like:
> >>>>
> >>>> 		(cpus_have_const_cap(ARM64_HAS_GENERIC_AUTH_ARCH) && \
> >>>> 		 ((vcpu)->arch.flags & KVM_ARM64_GUEST_HAS_PTRAUTH))
> >>>
> >>> In the subsequent patches, vcpu->arch.flags is only set to 
> >>> KVM_ARM64_GUEST_HAS_PTRAUTH when all host capability check conditions 
> >>> matches such as system_supports_address_auth(), 
> >>> system_supports_generic_auth() so doing them again is repetitive in my view.
> >>
> >> It isn't the setting of the flag I care about, but the check of that
> >> flag. Checking a flag for a feature that cannot be used on the running
> >> system should have a zero cost, which isn't the case here.
> >>
> >> Granted, the impact should be minimal and it looks like it mostly happen
> >> on the slow path, but at the very least it would be consistent. So even
> >> if you don't buy my argument about efficiency, please change it in the
> >> name of consistency.
> > 
> > One of the annoyances here is there is no single static key for ptrauth.
> > 
> > I'm assuming we don't want to check both static keys (for address and
> > generic auth) on hot paths.
> 
> They both just branches, so I don't see why not. Of course, for people
> using a lesser compiler (gcc 4.8 or clang), things will suck. But they
> got it coming anyway.

I seem to recall Christoffer expressing concerns about this at some
point: even unconditional branches branches to a fixed address are not
free (or even correctly predicted).

I don't think any compiler can elide static key checks of merge them
together.

Maybe I am misremembering.

Cheers
---Dave