KVM: arm64: Write arch.mdcr_el2 changes since last vcpu_load on VHE

On VHE systems arch.mdcr_el2 is written to mdcr_el2 at vcpu_load time to
set options for self-hosted debug and the performance monitors
extension.

Unfortunately the value of arch.mdcr_el2 is not calculated until
kvm_arm_setup_debug() in the run loop after the vcpu has been loaded.
This means that the initial brief iterations of the run loop use a zero
value of mdcr_el2 - until the vcpu is preempted. This also results in a
delay between changes to vcpu->guest_debug taking effect.

Fix this by writing to mdcr_el2 in kvm_arm_setup_debug() on VHE systems
when a change to arch.mdcr_el2 has been detected.

Fixes: d5a21bcc ("KVM: arm64: Move common VHE/non-VHE trap config in separate functions")
Cc: <stable@vger.kernel.org> # 4.17.x-
Suggested-by: NJames Morse <james.morse@arm.com>
Acked-by: NWill Deacon <will@kernel.org>
Reviewed-by: NMarc Zyngier <maz@kernel.org>
Signed-off-by: NAndrew Murray <andrew.murray@arm.com>
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>

arch/arm64/kvm/debug.c

@@ -101,7 +101,7 @@ void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu)
 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu)
 {
 	bool trap_debug = !(vcpu->arch.flags & KVM_ARM64_DEBUG_DIRTY);
-	unsigned long mdscr;
+	unsigned long mdscr, orig_mdcr_el2 = vcpu->arch.mdcr_el2;
 
 	trace_kvm_arm_setup_debug(vcpu, vcpu->guest_debug);
 
@@ -197,6 +197,10 @@ void kvm_arm_setup_debug(struct kvm_vcpu *vcpu)
 	if (vcpu_read_sys_reg(vcpu, MDSCR_EL1) & (DBG_MDSCR_KDE | DBG_MDSCR_MDE))
 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
 
+	/* Write mdcr_el2 changes since vcpu_load on VHE systems */
+	if (has_vhe() && orig_mdcr_el2 != vcpu->arch.mdcr_el2)
+		write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
+
 	trace_kvm_arm_set_dreg32("MDCR_EL2", vcpu->arch.mdcr_el2);
 	trace_kvm_arm_set_dreg32("MDSCR_EL1", vcpu_read_sys_reg(vcpu, MDSCR_EL1));
 }