diff --git a/arch/arm64/include/asm/kvm_hyp.h b/arch/arm64/include/asm/kvm_hyp.h
index 97f21cc666579..5fde137b51505 100644
--- a/arch/arm64/include/asm/kvm_hyp.h
+++ b/arch/arm64/include/asm/kvm_hyp.h
@@ -47,6 +47,13 @@
 #define read_sysreg_el2(r)	read_sysreg_elx(r, _EL2, _EL1)
 #define write_sysreg_el2(v,r)	write_sysreg_elx(v, r, _EL2, _EL1)
 
+/*
+ * Without an __arch_swab32(), we fall back to ___constant_swab32(), but the
+ * static inline can allow the compiler to out-of-line this. KVM always wants
+ * the macro version as its always inlined.
+ */
+#define __kvm_swab32(x)	___constant_swab32(x)
+
 int __vgic_v2_perform_cpuif_access(struct kvm_vcpu *vcpu);
 
 void __vgic_v3_save_state(struct kvm_vcpu *vcpu);
diff --git a/arch/arm64/kvm/hyp/vgic-v2-cpuif-proxy.c b/arch/arm64/kvm/hyp/vgic-v2-cpuif-proxy.c
index 29ee1feba4eb7..4f3a087e36d51 100644
--- a/arch/arm64/kvm/hyp/vgic-v2-cpuif-proxy.c
+++ b/arch/arm64/kvm/hyp/vgic-v2-cpuif-proxy.c
@@ -69,14 +69,14 @@ int __hyp_text __vgic_v2_perform_cpuif_access(struct kvm_vcpu *vcpu)
 		u32 data = vcpu_get_reg(vcpu, rd);
 		if (__is_be(vcpu)) {
 			/* guest pre-swabbed data, undo this for writel() */
-			data = swab32(data);
+			data = __kvm_swab32(data);
 		}
 		writel_relaxed(data, addr);
 	} else {
 		u32 data = readl_relaxed(addr);
 		if (__is_be(vcpu)) {
 			/* guest expects swabbed data */
-			data = swab32(data);
+			data = __kvm_swab32(data);
 		}
 		vcpu_set_reg(vcpu, rd, data);
 	}