Thread (9 messages) 9 messages, 3 authors, 2025-02-04

Re: [PATCH 3/3] KVM: arm64: timer: Consolidate NV configuration of virtual timers

From: Marc Zyngier <maz@kernel.org>
Date: 2025-01-31 08:47:50
Also in: kvmarm

On Thu, 30 Jan 2025 21:41:04 +0000,
Oliver Upton [off-list ref] wrote:
quoted
+void kvm_timer_vcpu_nv_init(struct kvm_vcpu *vcpu)
+{
+	/*
+	 * A vcpu running at EL2 is in charge of the offset applied to
+	 * the virtual timer, so use the physical VM offset, and point
+	 * the vcpu offset to CNTVOFF_EL2.
+	 *
+	 * The virtual offset behaviour is "interesting", as it always
+	 * applies when HCR_EL2.E2H==0, but only when accessed from EL1 when
+	 * HCR_EL2.E2H==1. Apply it to the HV timer when E2H==0.
+	 */
I'm definitely being pedantic, but all the talk of an HV timer when
E2H==0 isn't sitting well with me. Since a programmable E2H has gone
out the window there isn't such thing as an HV timer when E2H==0, as
FEAT_VHE isn't implemented for the VM.
Ah, that's a very good point!
And along those lines, accesses to CNTHV_*_EL2 registers should undef
when FEAT_VHE isn't implemented for the VM but I don't think we have any
enforcement of that.
Indeed. Let me fix this and add the required UNDEF behaviour.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help