RISC-V: KVM: Flush VS-stage TLB after VCPU migration for Andes cores

Most implementations cache the combined result of two-stage translation,
but some, like Andes cores, use split TLBs that store VS-stage and
G-stage entries separately.

On such systems, when a VCPU migrates to another CPU, an additional
HFENCE.VVMA is required to avoid using stale VS-stage entries, which
could otherwise cause guest faults.

Introduce a static key to identify CPUs with split two-stage TLBs.
When enabled, KVM issues an extra HFENCE.VVMA on VCPU migration to
prevent stale VS-stage mappings.

Signed-off-by: Hui Min Mina Chou <minachou@andestech.com>
Signed-off-by: Ben Zong-You Xie <ben717@andestech.com>
Reviewed-by: Radim Krčmář <rkrcmar@ventanamicro.com>
Reviewed-by: Nutty Liu <nutty.liu@hotmail.com>
Link: https://lore.kernel.org/r/20251117084555.157642-1-minachou@andestech.com
Signed-off-by: Anup Patel <anup@brainfault.org>

arch/riscv/include/asm/kvm_host.h

@@ -330,4 +330,7 @@ bool kvm_riscv_vcpu_stopped(struct kvm_vcpu *vcpu);
 
 void kvm_riscv_vcpu_record_steal_time(struct kvm_vcpu *vcpu);
 
+/* Flags representing implementation specific details */
+DECLARE_STATIC_KEY_FALSE(kvm_riscv_vsstage_tlb_no_gpa);
+
 #endif /* __RISCV_KVM_HOST_H__ */

arch/riscv/include/asm/kvm_tlb.h

@@ -49,6 +49,7 @@ void kvm_riscv_local_hfence_vvma_gva(unsigned long vmid,
 				     unsigned long gva, unsigned long gvsz,
 				     unsigned long order);
 void kvm_riscv_local_hfence_vvma_all(unsigned long vmid);
+void kvm_riscv_local_tlb_sanitize(struct kvm_vcpu *vcpu);
 
 void kvm_riscv_tlb_flush_process(struct kvm_vcpu *vcpu);
 

arch/riscv/include/asm/kvm_vmid.h

@@ -22,6 +22,5 @@ unsigned long kvm_riscv_gstage_vmid_bits(void);
 int kvm_riscv_gstage_vmid_init(struct kvm *kvm);
 bool kvm_riscv_gstage_vmid_ver_changed(struct kvm_vmid *vmid);
 void kvm_riscv_gstage_vmid_update(struct kvm_vcpu *vcpu);
-void kvm_riscv_gstage_vmid_sanitize(struct kvm_vcpu *vcpu);
 
 #endif

arch/riscv/kvm/main.c

@@ -15,6 +15,18 @@
 #include <asm/kvm_nacl.h>
 #include <asm/sbi.h>
 
+DEFINE_STATIC_KEY_FALSE(kvm_riscv_vsstage_tlb_no_gpa);
+
+static void kvm_riscv_setup_vendor_features(void)
+{
+	/* Andes AX66: split two-stage TLBs */
+	if (riscv_cached_mvendorid(0) == ANDES_VENDOR_ID &&
+	    (riscv_cached_marchid(0) & 0xFFFF) == 0x8A66) {
+		static_branch_enable(&kvm_riscv_vsstage_tlb_no_gpa);
+		kvm_info("VS-stage TLB does not cache guest physical address and VMID\n");
+	}
+}
+
 long kvm_arch_dev_ioctl(struct file *filp,
 			unsigned int ioctl, unsigned long arg)
 {
@@ -160,6 +172,8 @@ static int __init riscv_kvm_init(void)
 		kvm_info("AIA available with %d guest external interrupts\n",
 			 kvm_riscv_aia_nr_hgei);
 
+	kvm_riscv_setup_vendor_features();
+
 	kvm_register_perf_callbacks(NULL);
 
 	rc = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);

arch/riscv/kvm/tlb.c

@@ -158,6 +158,36 @@ void kvm_riscv_local_hfence_vvma_all(unsigned long vmid)
 	csr_write(CSR_HGATP, hgatp);
 }
 
+void kvm_riscv_local_tlb_sanitize(struct kvm_vcpu *vcpu)
+{
+	unsigned long vmid;
+
+	if (!kvm_riscv_gstage_vmid_bits() ||
+	    vcpu->arch.last_exit_cpu == vcpu->cpu)
+		return;
+
+	/*
+	 * On RISC-V platforms with hardware VMID support, we share same
+	 * VMID for all VCPUs of a particular Guest/VM. This means we might
+	 * have stale G-stage TLB entries on the current Host CPU due to
+	 * some other VCPU of the same Guest which ran previously on the
+	 * current Host CPU.
+	 *
+	 * To cleanup stale TLB entries, we simply flush all G-stage TLB
+	 * entries by VMID whenever underlying Host CPU changes for a VCPU.
+	 */
+
+	vmid = READ_ONCE(vcpu->kvm->arch.vmid.vmid);
+	kvm_riscv_local_hfence_gvma_vmid_all(vmid);
+
+	/*
+	 * Flush VS-stage TLB entries for implementation where VS-stage
+	 * TLB does not cahce guest physical address and VMID.
+	 */
+	if (static_branch_unlikely(&kvm_riscv_vsstage_tlb_no_gpa))
+		kvm_riscv_local_hfence_vvma_all(vmid);
+}
+
 void kvm_riscv_fence_i_process(struct kvm_vcpu *vcpu)
 {
 	kvm_riscv_vcpu_pmu_incr_fw(vcpu, SBI_PMU_FW_FENCE_I_RCVD);

arch/riscv/kvm/vcpu.c

@@ -968,7 +968,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
 		 * Note: This should be done after G-stage VMID has been
 		 * updated using kvm_riscv_gstage_vmid_ver_changed()
 		 */
-		kvm_riscv_gstage_vmid_sanitize(vcpu);
+		kvm_riscv_local_tlb_sanitize(vcpu);
 
 		trace_kvm_entry(vcpu);
 

arch/riscv/kvm/vmid.c

@@ -122,26 +122,3 @@ void kvm_riscv_gstage_vmid_update(struct kvm_vcpu *vcpu)
 	kvm_for_each_vcpu(i, v, vcpu->kvm)
 		kvm_make_request(KVM_REQ_UPDATE_HGATP, v);
 }
-
-void kvm_riscv_gstage_vmid_sanitize(struct kvm_vcpu *vcpu)
-{
-	unsigned long vmid;
-
-	if (!kvm_riscv_gstage_vmid_bits() ||
-	    vcpu->arch.last_exit_cpu == vcpu->cpu)
-		return;
-
-	/*
-	 * On RISC-V platforms with hardware VMID support, we share same
-	 * VMID for all VCPUs of a particular Guest/VM. This means we might
-	 * have stale G-stage TLB entries on the current Host CPU due to
-	 * some other VCPU of the same Guest which ran previously on the
-	 * current Host CPU.
-	 *
-	 * To cleanup stale TLB entries, we simply flush all G-stage TLB
-	 * entries by VMID whenever underlying Host CPU changes for a VCPU.
-	 */
-
-	vmid = READ_ONCE(vcpu->kvm->arch.vmid.vmid);
-	kvm_riscv_local_hfence_gvma_vmid_all(vmid);
-}