Re: [PATCH v4 2/6] arm64/kvm: context-switch ptrauth registers

[James Morse]

Hi Amit,

On 18/12/2018 07:56, Amit Daniel Kachhap wrote:
> When pointer authentication is supported, a guest may wish to use it.
> This patch adds the necessary KVM infrastructure for this to work, with
> a semi-lazy context switch of the pointer auth state.
> 
> Pointer authentication feature is only enabled when VHE is built
> into the kernel and present into CPU implementation so only VHE code
> paths are modified.
> 
> When we schedule a vcpu, we disable guest usage of pointer
> authentication instructions and accesses to the keys. While these are
> disabled, we avoid context-switching the keys. When we trap the guest
> trying to use pointer authentication functionality, we change to eagerly
> context-switching the keys, and enable the feature. The next time the
> vcpu is scheduled out/in, we start again.

Why start again?
Taking the trap at all suggests the guest knows about pointer-auth, if it uses
it, its probably in some context switch code. It would be good to avoid trapping
that.


> Pointer authentication consists of address authentication and generic
> authentication, and CPUs in a system might have varied support for
> either. Where support for either feature is not uniform, it is hidden
> from guests via ID register emulation, as a result of the cpufeature
> framework in the host.
> 
> Unfortunately, address authentication and generic authentication cannot
> be trapped separately, as the architecture provides a single EL2 trap
> covering both. If we wish to expose one without the other, we cannot
> prevent a (badly-written) guest from intermittently using a feature
> which is not uniformly supported (when scheduled on a physical CPU which
> supports the relevant feature).

Yuck!


> When the guest is scheduled on a
> physical CPU lacking the feature, these attemts will result in an UNDEF

(attempts)

> being taken by the guest.

Can we only expose the feature to a guest if both address and generic
authentication are supported? (and hide it from the id register otherwise).

This avoids having to know if this cpu supports address/generic when the system
doesn't. We would need to scrub the values to avoid guest-values being left
loaded when something else is running.


> diff --git a/arch/arm64/include/asm/cpufeature.h b/arch/arm64/include/asm/cpufeature.h
> index 1c8393f..ac7d496 100644
> --- a/arch/arm64/include/asm/cpufeature.h
> +++ b/arch/arm64/include/asm/cpufeature.h
> @@ -526,6 +526,12 @@ static inline bool system_supports_generic_auth(void)
>  		cpus_have_const_cap(ARM64_HAS_GENERIC_AUTH);
>  }
>  
> +static inline bool system_supports_ptrauth(void)
> +{
> +	return IS_ENABLED(CONFIG_ARM64_PTR_AUTH) &&
> +		cpus_have_const_cap(ARM64_HAS_ADDRESS_AUTH);
> +}

(mainline has a system_supports_address_auth(), I guess this will be folded
around a bit during the rebase).

kvm_supports_ptrauth() that checks the two architected algorithm caps?


> diff --git a/arch/arm64/kvm/handle_exit.c b/arch/arm64/kvm/handle_exit.c
> index ab35929..5c47a8f47 100644
> --- a/arch/arm64/kvm/handle_exit.c
> +++ b/arch/arm64/kvm/handle_exit.c
> @@ -174,19 +174,25 @@ static int handle_sve(struct kvm_vcpu *vcpu, struct kvm_run *run)
>  }
>  
>  /*
> + * Handle the guest trying to use a ptrauth instruction, or trying to access a
> + * ptrauth register. This trap should not occur as we enable ptrauth during
> + * vcpu schedule itself but is anyway kept here for any unfortunate scenario.

... so we don't need this? Or if it ever runs its indicative of a bug?


> + */
> +void kvm_arm_vcpu_ptrauth_trap(struct kvm_vcpu *vcpu)
> +{
> +	if (system_supports_ptrauth())
> +		kvm_arm_vcpu_ptrauth_enable(vcpu);
> +	else
> +		kvm_inject_undefined(vcpu);
> +}
> +
> +/*
>   * Guest usage of a ptrauth instruction (which the guest EL1 did not turn into
> - * a NOP).
> + * a NOP), or guest EL1 access to a ptrauth register.
>   */
>  static int kvm_handle_ptrauth(struct kvm_vcpu *vcpu, struct kvm_run *run)
>  {
> -	/*
> -	 * We don't currently support ptrauth in a guest, and we mask the ID
> -	 * registers to prevent well-behaved guests from trying to make use of
> -	 * it.
> -	 *
> -	 * Inject an UNDEF, as if the feature really isn't present.
> -	 */
> -	kvm_inject_undefined(vcpu);
> +	kvm_arm_vcpu_ptrauth_trap(vcpu);
>  	return 1;
>  }

> diff --git a/arch/arm64/kvm/hyp/ptrauth-sr.c b/arch/arm64/kvm/hyp/ptrauth-sr.c
> new file mode 100644
> index 0000000..1bfaf74
> --- /dev/null
> +++ b/arch/arm64/kvm/hyp/ptrauth-sr.c
> @@ -0,0 +1,73 @@

> +{
> +	return vcpu->arch.hcr_el2 & (HCR_API | HCR_APK);
> +}

(If you include the the IS_ENABLED() in here too, the compiler can work out
pointer-auth was compiled out and prune all the unused static functions.)


> +#define __ptrauth_save_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);	\
> +})

(shame we can't re-use the arch code's macros for these)


> +static __always_inline void __hyp_text __ptrauth_save_state(struct kvm_cpu_context *ctxt)
> +{
> +	__ptrauth_save_key(ctxt->sys_regs, APIA);
> +	__ptrauth_save_key(ctxt->sys_regs, APIB);
> +	__ptrauth_save_key(ctxt->sys_regs, APDA);
> +	__ptrauth_save_key(ctxt->sys_regs, APDB);

> +	__ptrauth_save_key(ctxt->sys_regs, APGA);

I thought the generic authentication bits had a separate ID-register-field/cap?
But you only checked address-auth above. Is it possible the CPU doesn't have
this register?

(but I think we should only support this if both generic&address are supported)


> +}

> +void __hyp_text __ptrauth_switch_to_guest(struct kvm_vcpu *vcpu,
> +					  struct kvm_cpu_context *host_ctxt,
> +					  struct kvm_cpu_context *guest_ctxt)
> +{
> +	if (!__ptrauth_is_enabled(vcpu))
> +		return;
> +
> +	__ptrauth_save_state(host_ctxt);
> +	__ptrauth_restore_state(guest_ctxt);
> +}
> +
> +void __hyp_text __ptrauth_switch_to_host(struct kvm_vcpu *vcpu,
> +					 struct kvm_cpu_context *host_ctxt,
> +					 struct kvm_cpu_context *guest_ctxt)
> +{
> +	if (!__ptrauth_is_enabled(vcpu))
> +		return;
> +
> +	__ptrauth_save_state(guest_ctxt);
> +	__ptrauth_restore_state(host_ctxt);
> +}


> diff --git a/arch/arm64/kvm/hyp/switch.c b/arch/arm64/kvm/hyp/switch.c
> index 85a2a5c..6c57552 100644
> --- a/arch/arm64/kvm/hyp/switch.c
> +++ b/arch/arm64/kvm/hyp/switch.c
> @@ -508,6 +508,8 @@ int kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
>  	sysreg_restore_guest_state_vhe(guest_ctxt);
>  	__debug_switch_to_guest(vcpu);
>  
> +	__ptrauth_switch_to_guest(vcpu, host_ctxt, guest_ctxt);
> +
>  	__set_guest_arch_workaround_state(vcpu);
>  
>  	do {
> @@ -519,6 +521,8 @@ int kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
>  
>  	__set_host_arch_workaround_state(vcpu);
>  
> +	__ptrauth_switch_to_host(vcpu, host_ctxt, guest_ctxt);
> +
>  	sysreg_save_guest_state_vhe(guest_ctxt);
>  
>  	__deactivate_traps(vcpu);

This is deep in the world-switch code, meaning we save/restore 10 sysregs for
every entry/exit. As the kernel isn't using pointer-auth, wouldn't it be better
to defer the save/restore to the preempt notifier using kvm_arch_vcpu_load()?

I've seen the discussion that the kernel may start using this 'soon':
lore.kernel.org/r/20181112223212.5o4ipc5kt5ziuupt@localhost

(does anyone know when soon is?)

... but it doesn't today, and save/restoring these in C becomes impossible when
that happens, which the arch code is doing too, so we aren't inventing a new
problem.


probable-tangent: {
If the kernel starts using ptrauth, it will need to switch these registers in
the entry asm, and in __cpu_switch_to()... there will be new asm macro's to do this.

Can we make it so that the same series that does that, can easily do KVM too?

If we use the preempt-notifier stuff the GET/SET_ONE_REG code knows that today
the ptrauth values are always loaded on the cpu, it doesn't need to look in the
sys_reg_descs[] array.
This means the preempt notifier could save/restore them to a struct ptrauth_keys
in struct kvm_vcpu_arch. This lets us use the arch codes __ptrauth_key_install()
and ptrauth_keys_switch() helpers today, instead of duplicating them because the
layout is slightly different.

Once these become asm macros, the same structure is already in use, so we can
(cough) probably feed it a different base pointer for guest-entry/ret_to_user.
guest-exit would almost be the same as kernel_enter, its only the
save-guest-values that would be specific to kvm.

(oh, and the GET/SET_ONE_REG code would need to know to look elsewhere for the
guest values, but it looks like your patch 5 is already doing some of this).

}


> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
> index 1ca592d..6af6c7d 100644
> --- a/arch/arm64/kvm/sys_regs.c
> +++ b/arch/arm64/kvm/sys_regs.c
> @@ -986,6 +986,32 @@ static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
>  	{ SYS_DESC(SYS_PMEVTYPERn_EL0(n)),					\
>  	  access_pmu_evtyper, reset_unknown, (PMEVTYPER0_EL0 + n), }
>  
> +
> +void kvm_arm_vcpu_ptrauth_enable(struct kvm_vcpu *vcpu)
> +{
> +	vcpu->arch.hcr_el2 |= (HCR_API | HCR_APK);
> +}
> +
> +void kvm_arm_vcpu_ptrauth_disable(struct kvm_vcpu *vcpu)
> +{
> +	vcpu->arch.hcr_el2 &= ~(HCR_API | HCR_APK);
> +}
> +

> @@ -1040,14 +1066,6 @@ static u64 read_id_reg(struct sys_reg_desc const *r, bool raz)
>  			kvm_debug("SVE unsupported for guests, suppressing\n");
>  
>  		val &= ~(0xfUL << ID_AA64PFR0_SVE_SHIFT);
> -	} else if (id == SYS_ID_AA64ISAR1_EL1) {
> -		const u64 ptrauth_mask = (0xfUL << ID_AA64ISAR1_APA_SHIFT) |
> -					 (0xfUL << ID_AA64ISAR1_API_SHIFT) |
> -					 (0xfUL << ID_AA64ISAR1_GPA_SHIFT) |
> -					 (0xfUL << ID_AA64ISAR1_GPI_SHIFT);

Don't we still want to remove the imp-def bits? and the APA/GPA bits if we
decide there is incomplete support.


> -		if (val & ptrauth_mask)
> -			kvm_debug("ptrauth unsupported for guests, suppressing\n");
> -		val &= ~ptrauth_mask;
>  	} else if (id == SYS_ID_AA64MMFR1_EL1) {
>  		if (val & (0xfUL << ID_AA64MMFR1_LOR_SHIFT))
>  			kvm_debug("LORegions unsupported for guests, suppressing\n");


> diff --git a/virt/kvm/arm/arm.c b/virt/kvm/arm/arm.c
> index feda456..1f3b6ed 100644
> --- a/virt/kvm/arm/arm.c
> +++ b/virt/kvm/arm/arm.c
> @@ -388,6 +388,8 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
>  		vcpu_clear_wfe_traps(vcpu);
>  	else
>  		vcpu_set_wfe_traps(vcpu);
> +
> +	kvm_arm_vcpu_ptrauth_config(vcpu);

(doesn't this unconfigure ptrauth?!)

>  }

Thanks,

James