Re: [RFC 8/9] arm64/kvm: context-switch PAC registers

From: Marc Zyngier
Date: Fri Apr 07 2017 - 11:41:50 EST


Hi Mark,

On 03/04/17 16:19, Mark Rutland wrote:
> If we have pointer authentication support, a guest may wish to use it.
> This patch adds the infrastructure to allow it to do so.
>
> This is sufficient for basic testing, but not for real-world usage. A
> guest will still see pointer authentication support advertised in the ID
> registers, and we will need to trap accesses to these to provide
> santized values.
>
> Signed-off-by: Mark Rutland <mark.rutland@xxxxxxx>
> Cc: Christoffer Dall <christoffer.dall@xxxxxxxxxx>
> Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
> Cc: kvmarm@xxxxxxxxxxxxxxxxxxxxx
> ---
> arch/arm64/include/asm/kvm_emulate.h | 15 +++++++++++++
> arch/arm64/include/asm/kvm_host.h | 12 ++++++++++
> arch/arm64/kvm/hyp/sysreg-sr.c | 43 ++++++++++++++++++++++++++++++++++++
> 3 files changed, 70 insertions(+)
>
> diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h
> index f5ea0ba..0c3cb43 100644
> --- a/arch/arm64/include/asm/kvm_emulate.h
> +++ b/arch/arm64/include/asm/kvm_emulate.h
> @@ -28,6 +28,8 @@
> #include <asm/kvm_arm.h>
> #include <asm/kvm_mmio.h>
> #include <asm/ptrace.h>
> +#include <asm/cpucaps.h>
> +#include <asm/cpufeature.h>
> #include <asm/cputype.h>
> #include <asm/virt.h>
>
> @@ -49,6 +51,19 @@ static inline void vcpu_reset_hcr(struct kvm_vcpu *vcpu)
> vcpu->arch.hcr_el2 |= HCR_E2H;
> if (test_bit(KVM_ARM_VCPU_EL1_32BIT, vcpu->arch.features))
> vcpu->arch.hcr_el2 &= ~HCR_RW;
> +
> + /*
> + * Address auth and generic auth share the same enable bits, so we have
> + * to ensure both are uniform before we can enable support in a guest.
> + * Until we have the infrastructure to detect uniform absence of a
> + * feature, only permit the case when both are supported.
> + *
> + * Note that a guest will still see the feature in ID_AA64_ISAR1 until
> + * we introduce code to emulate the ID registers.
> + */
> + if (cpus_have_const_cap(ARM64_HAS_ADDRESS_AUTH) &&
> + cpus_have_const_cap(ARM64_HAS_GENERIC_AUTH))
> + vcpu->arch.hcr_el2 |= (HCR_API | HCR_APK);

Instead of unconditionally allowing the guest to access this feature...

> }
>
> static inline unsigned long vcpu_get_hcr(struct kvm_vcpu *vcpu)
> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index e7705e7..b25f710 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -133,6 +133,18 @@ enum vcpu_sysreg {
> PMSWINC_EL0, /* Software Increment Register */
> PMUSERENR_EL0, /* User Enable Register */
>
> + /* Pointer Authentication Registers */
> + APIAKEYLO_EL1,
> + APIAKEYHI_EL1,
> + APIBKEYLO_EL1,
> + APIBKEYHI_EL1,
> + APDAKEYLO_EL1,
> + APDAKEYHI_EL1,
> + APDBKEYLO_EL1,
> + APDBKEYHI_EL1,
> + APGAKEYLO_EL1,
> + APGAKEYHI_EL1,
> +
> /* 32bit specific registers. Keep them at the end of the range */
> DACR32_EL2, /* Domain Access Control Register */
> IFSR32_EL2, /* Instruction Fault Status Register */
> diff --git a/arch/arm64/kvm/hyp/sysreg-sr.c b/arch/arm64/kvm/hyp/sysreg-sr.c
> index 9341376..3440b42 100644
> --- a/arch/arm64/kvm/hyp/sysreg-sr.c
> +++ b/arch/arm64/kvm/hyp/sysreg-sr.c
> @@ -18,6 +18,8 @@
> #include <linux/compiler.h>
> #include <linux/kvm_host.h>
>
> +#include <asm/cpucaps.h>
> +#include <asm/cpufeature.h>
> #include <asm/kvm_asm.h>
> #include <asm/kvm_hyp.h>
>
> @@ -31,6 +33,24 @@ static void __hyp_text __sysreg_do_nothing(struct kvm_cpu_context *ctxt) { }
> * pstate, and guest must save everything.
> */
>
> +#define __save_ap_key(regs, key) \
> + regs[key ## KEYLO_EL1] = read_sysreg_s(SYS_ ## key ## KEYLO_EL1); \
> + regs[key ## KEYHI_EL1] = read_sysreg_s(SYS_ ## key ## KEYHI_EL1)
> +
> +static void __hyp_text __sysreg_save_ap_keys(struct kvm_cpu_context *ctxt)
> +{
> + if (cpus_have_const_cap(ARM64_HAS_ADDRESS_AUTH)) {
> + __save_ap_key(ctxt->sys_regs, APIA);
> + __save_ap_key(ctxt->sys_regs, APIB);
> + __save_ap_key(ctxt->sys_regs, APDA);
> + __save_ap_key(ctxt->sys_regs, APDB);
> + }
> +
> + if (cpus_have_const_cap(ARM64_HAS_GENERIC_AUTH)) {
> + __save_ap_key(ctxt->sys_regs, APGA);
> + }
> +}
> +

...which immediately translates in quite a bit of sysreg churn on both
host and guest (specially given that these are not banked by exception
level), could we make it a bit more lazy instead?

Even an "enable on first use" would be good, given that it is likely
that we'll have non PAC-enabled VMs for quite a while.

Thoughts?

M.
--
Jazz is not dead. It just smells funny...