[PATCH] KVM: arm64: Fix kvm_vcpu_{set,is}_be() to deal with EL2 state

[PATCH] KVM: arm64: Fix kvm_vcpu_{set,is}_be() to deal with EL2 state

[Marc Zyngier]

Nobody really cares about BE, but KVM currently only deals with
SCTLR_EL1 when evaluating or setting the endianness in PSCI,
meaning that we evaluate whatever the L2 state has been at some point.

Teach these primitives about SCTLR_EL2, and forget about BE...

Signed-off-by: Marc Zyngier <maz@kernel.org>
---
 arch/arm64/include/asm/kvm_emulate.h | 20 ++++++++++++++------
 1 file changed, 14 insertions(+), 6 deletions(-)

diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h
index fa8a08a1ccd5c..85f998dffea5d 100644
--- a/arch/arm64/include/asm/kvm_emulate.h
+++ b/arch/arm64/include/asm/kvm_emulate.h
@@ -511,21 +511,29 @@ static inline void kvm_vcpu_set_be(struct kvm_vcpu *vcpu)
 	if (vcpu_mode_is_32bit(vcpu)) {
 		*vcpu_cpsr(vcpu) |= PSR_AA32_E_BIT;
 	} else {
-		u64 sctlr = vcpu_read_sys_reg(vcpu, SCTLR_EL1);
+		enum vcpu_sysreg r;
+		u64 sctlr;
+
+		r = vcpu_has_nv(vcpu) ? SCTLR_EL2 : SCTLR_EL1;
+
+		sctlr = vcpu_read_sys_reg(vcpu, r);
 		sctlr |= SCTLR_ELx_EE;
-		vcpu_write_sys_reg(vcpu, sctlr, SCTLR_EL1);
+		vcpu_write_sys_reg(vcpu, sctlr, r);
 	}
 }
 
 static inline bool kvm_vcpu_is_be(struct kvm_vcpu *vcpu)
 {
+	enum vcpu_sysreg r;
+	u64 bit;
+
 	if (vcpu_mode_is_32bit(vcpu))
 		return !!(*vcpu_cpsr(vcpu) & PSR_AA32_E_BIT);
 
-	if (vcpu_mode_priv(vcpu))
-		return !!(vcpu_read_sys_reg(vcpu, SCTLR_EL1) & SCTLR_ELx_EE);
-	else
-		return !!(vcpu_read_sys_reg(vcpu, SCTLR_EL1) & SCTLR_EL1_E0E);
+	r = is_hyp_ctxt(vcpu) ? SCTLR_EL2 : SCTLR_EL1;
+	bit = vcpu_mode_priv(vcpu) ? SCTLR_ELx_EE : SCTLR_EL1_E0E;
+
+	return vcpu_read_sys_reg(vcpu, r) & bit;
 }
 
 static inline unsigned long vcpu_data_guest_to_host(struct kvm_vcpu *vcpu,
-- 
2.39.2

Re: [PATCH] KVM: arm64: Fix kvm_vcpu_{set,is}_be() to deal with EL2 state

[Oliver Upton]

On Tue, Sep 16, 2025 at 05:11:03PM +0100, Marc Zyngier wrote:
> Nobody really cares about BE, but KVM currently only deals with
> SCTLR_EL1 when evaluating or setting the endianness in PSCI,
> meaning that we evaluate whatever the L2 state has been at some point.
> 
> Teach these primitives about SCTLR_EL2, and forget about BE...
> 
> Signed-off-by: Marc Zyngier <maz@kernel.org>

You know, I was about to say "don't we need to sanitise this?" although
that seems to already be in place. Nice.

Reviewed-by: Oliver Upton <oliver.upton@linux.dev>

Thanks,
Oliver

Re: [PATCH] KVM: arm64: Fix kvm_vcpu_{set,is}_be() to deal with EL2 state

[Marc Zyngier]

On Wed, 17 Sep 2025 06:21:09 +0100,
Oliver Upton <oliver.upton@linux.dev> wrote:
> 
> On Tue, Sep 16, 2025 at 05:11:03PM +0100, Marc Zyngier wrote:
> > Nobody really cares about BE, but KVM currently only deals with
> > SCTLR_EL1 when evaluating or setting the endianness in PSCI,
> > meaning that we evaluate whatever the L2 state has been at some point.
> > 
> > Teach these primitives about SCTLR_EL2, and forget about BE...
> > 
> > Signed-off-by: Marc Zyngier <maz@kernel.org>
> 
> You know, I was about to say "don't we need to sanitise this?" although
> that seems to already be in place. Nice.

Only for SCTLR_EL1.

SCTLR_EL2, annoyingly enough, isn't, owing to plenty of bits depending
on E2H. It's on my long list of things to address, but I'd like a
common mechanism that'd deal with horrors such as CNTHCTL_EL2...

> Reviewed-by: Oliver Upton <oliver.upton@linux.dev>

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.

Re: [PATCH] KVM: arm64: Fix kvm_vcpu_{set,is}_be() to deal with EL2 state

[Marc Zyngier]

On Tue, 16 Sep 2025 17:11:03 +0100, Marc Zyngier wrote:
> Nobody really cares about BE, but KVM currently only deals with
> SCTLR_EL1 when evaluating or setting the endianness in PSCI,
> meaning that we evaluate whatever the L2 state has been at some point.
> 
> Teach these primitives about SCTLR_EL2, and forget about BE...
> 
> 
> [...]

Applied to next, thanks!

[1/1] KVM: arm64: Fix kvm_vcpu_{set,is}_be() to deal with EL2 state
      commit: 6515c612e79949b17ef4b8c4180c07bbeaf01e4d

Cheers,

	M.
-- 
Without deviation from the norm, progress is not possible.