feat(tdx): restrict KVM features to a safe subset for TDX guests

alioth/src/board/board_x86_64/board_x86_64.rs

@@ -78,7 +78,7 @@ impl<V: Vm> ArchBoard<V> {
     where
         H: Hypervisor<Vm = V>,
     {
-        let mut cpuids = hv.get_supported_cpuids()?;
+        let mut cpuids = hv.get_supported_cpuids(config.coco.as_ref())?;
 
         let threads_per_core = 1 + config.cpu.topology.smt as u16;
         let threads_per_socket = config.cpu.topology.cores * threads_per_core;

alioth/src/hv/hv.rs

@@ -431,7 +431,7 @@ pub trait Hypervisor {
     fn create_vm(&self, config: &VmConfig) -> Result<Self::Vm, Error>;
 
     #[cfg(target_arch = "x86_64")]
-    fn get_supported_cpuids(&self) -> Result<HashMap<CpuidIn, CpuidResult>>;
+    fn get_supported_cpuids(&self, coco: Option<&Coco>) -> Result<HashMap<CpuidIn, CpuidResult>>;
 }
 
 #[derive(Debug, Clone, PartialEq, Eq)]

alioth/src/hv/kvm/kvm.rs

@@ -13,15 +13,11 @@
 // limitations under the License.
 
 #[cfg(target_arch = "x86_64")]
-#[path = "kvm_x86_64.rs"]
+#[path = "kvm_x86_64/kvm_x86_64.rs"]
 mod x86_64;
 
 #[cfg(target_arch = "aarch64")]
 mod device;
-#[cfg(target_arch = "x86_64")]
-mod sev;
-#[cfg(target_arch = "x86_64")]
-mod tdx;
 #[path = "vcpu/vcpu.rs"]
 mod vcpu;
 #[path = "vm/vm.rs"]
@@ -47,6 +43,8 @@ use snafu::{ResultExt, Snafu};
 use crate::arch::cpuid::CpuidIn;
 use crate::errors::{DebugTrace, trace_error};
 use crate::ffi;
+#[cfg(target_arch = "x86_64")]
+use crate::hv::Coco;
 use crate::hv::{Hypervisor, MemMapOption, Result, VmConfig, error};
 #[cfg(target_arch = "aarch64")]
 use crate::sys::kvm::KvmDevType;
@@ -153,8 +151,8 @@ impl Hypervisor for Kvm {
     }
 
     #[cfg(target_arch = "x86_64")]
-    fn get_supported_cpuids(&self) -> Result<HashMap<CpuidIn, CpuidResult>> {
-        Kvm::get_supported_cpuids(self)
+    fn get_supported_cpuids(&self, coco: Option<&Coco>) -> Result<HashMap<CpuidIn, CpuidResult>> {
+        Kvm::get_supported_cpuids(self, coco)
     }
 }
 

alioth/src/hv/kvm/kvm_x86_64.rs → alioth/src/hv/kvm/kvm_x86_64/kvm_x86_64.rs

@@ -12,12 +12,17 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+pub(crate) mod sev;
+pub(crate) mod tdx;
+
 use std::arch::x86_64::CpuidResult;
 use std::collections::HashMap;
 
 use snafu::ResultExt;
 
 use crate::arch::cpuid::CpuidIn;
+#[cfg(target_arch = "x86_64")]
+use crate::hv::Coco;
 use crate::hv::{Kvm, Result, error};
 use crate::sys::kvm::{
     KVM_CPUID_FEATURES, KVM_MAX_CPUID_ENTRIES, KvmCap, KvmCpuid2, KvmCpuid2Flag, KvmCpuidEntry2,
@@ -45,7 +50,10 @@ impl From<KvmCpuidEntry2> for (CpuidIn, CpuidResult) {
 }
 
 impl Kvm {
-    pub fn get_supported_cpuids(&self) -> Result<HashMap<CpuidIn, CpuidResult>> {
+    pub fn get_supported_cpuids(
+        &self,
+        coco: Option<&Coco>,
+    ) -> Result<HashMap<CpuidIn, CpuidResult>> {
         let mut kvm_cpuid2 = KvmCpuid2 {
             nent: KVM_MAX_CPUID_ENTRIES as u32,
             padding: 0,
@@ -64,14 +72,18 @@ impl Kvm {
             func: KVM_CPUID_FEATURES,
             index: None,
         };
-        if let Some(entry) = cpuids.get_mut(&leaf_features)
-            && let Ok(ext) = self.check_extension(KvmCap::X2APIC_API)
-            && KvmX2apicApiFlag::from_bits_retain(ext.get() as u64).contains(
-                KvmX2apicApiFlag::USE_32BIT_IDS | KvmX2apicApiFlag::DISABLE_BROADCAST_QUIRK,
-            )
-        {
-            // Enable KVM_FEATURE_MSI_EXT_DEST_ID if KVM_CAP_X2APIC_API is supported
-            entry.eax |= KvmCpuidFeature::MSI_EXT_DEST_ID.bits();
+        if let Some(entry) = cpuids.get_mut(&leaf_features) {
+            if let Ok(ext) = self.check_extension(KvmCap::X2APIC_API)
+                && KvmX2apicApiFlag::from_bits_retain(ext.get() as u64).contains(
+                    KvmX2apicApiFlag::USE_32BIT_IDS | KvmX2apicApiFlag::DISABLE_BROADCAST_QUIRK,
+                )
+            {
+                // Enable KVM_FEATURE_MSI_EXT_DEST_ID if KVM_CAP_X2APIC_API is supported
+                entry.eax |= KvmCpuidFeature::MSI_EXT_DEST_ID.bits();
+            }
+            if matches!(coco, Some(Coco::IntelTdx { .. })) {
+                entry.eax &= tdx::SUPPORTED_KVM_FEATURES;
+            }
         }
 
         Ok(cpuids)

alioth/src/hv/kvm/kvm_x86_64_test.rs → alioth/src/hv/kvm/kvm_x86_64/kvm_x86_64_test.rs

@@ -25,7 +25,7 @@ use crate::sys::kvm::{KVM_CPUID_SIGNATURE, KvmCpuid2Flag, KvmCpuidEntry2};
 fn test_get_supported_cpuid() {
     let kvm = Kvm::new(KvmConfig::default()).unwrap();
     let mut kvm_cpuid_exist = false;
-    let supported_cpuids = kvm.get_supported_cpuids().unwrap();
+    let supported_cpuids = kvm.get_supported_cpuids(None).unwrap();
     for (in_, out) in &supported_cpuids {
         if in_.func == KVM_CPUID_SIGNATURE
             && out.ebx.to_le_bytes() == *b"KVMK"

alioth/src/hv/kvm/tdx.rs → alioth/src/hv/kvm/kvm_x86_64/tdx.rs

@@ -17,7 +17,7 @@ use std::os::fd::OwnedFd;
 use snafu::ResultExt;
 
 use crate::hv::{Result, error};
-use crate::sys::kvm::kvm_memory_encrypt_op;
+use crate::sys::kvm::{KvmCpuidFeature, kvm_memory_encrypt_op};
 use crate::sys::tdx::{KvmTdxCmd, KvmTdxCmdId};
 
 pub fn tdx_op<T>(fd: &OwnedFd, cmd: KvmTdxCmdId, flags: u32, data: Option<&mut T>) -> Result<()> {
@@ -33,3 +33,11 @@ pub fn tdx_op<T>(fd: &OwnedFd, cmd: KvmTdxCmdId, flags: u32, data: Option<&mut T
     }
     Ok(())
 }
+
+pub const SUPPORTED_KVM_FEATURES: u32 = KvmCpuidFeature::NOP_IO_DELAY.bits()
+    | KvmCpuidFeature::PV_UNHALT.bits()
+    | KvmCpuidFeature::PV_TLB_FLUSH.bits()
+    | KvmCpuidFeature::PV_SEND_IPI.bits()
+    | KvmCpuidFeature::POLL_CONTROL.bits()
+    | KvmCpuidFeature::PV_SCHED_YIELD.bits()
+    | KvmCpuidFeature::MSI_EXT_DEST_ID.bits();

alioth/src/hv/kvm/vcpu/vcpu_x86_64/tdx.rs

@@ -13,8 +13,8 @@
 // limitations under the License.
 
 use crate::hv::Result;
-use crate::hv::kvm::tdx::tdx_op;
 use crate::hv::kvm::vcpu::KvmVcpu;
+use crate::hv::kvm::x86_64::tdx::tdx_op;
 use crate::sys::tdx::{KvmTdxCmdId, KvmTdxInitMemRegion, KvmTdxInitMemRegionFlag};
 
 impl KvmVcpu {

alioth/src/hv/kvm/vm/vm_x86_64/tdx.rs

@@ -19,8 +19,8 @@ use std::mem::MaybeUninit;
 use crate::arch::cpuid::CpuidIn;
 use crate::arch::tdx::TdAttr;
 use crate::hv::Result;
-use crate::hv::kvm::tdx::tdx_op;
 use crate::hv::kvm::vm::KvmVm;
+use crate::hv::kvm::x86_64::tdx::tdx_op;
 use crate::sys::kvm::{KvmCap, KvmCpuid2Flag, KvmCpuidEntry2, KvmHypercall};
 use crate::sys::tdx::{KvmTdxCapabilities, KvmTdxCmdId, KvmTdxInitVm};
 

alioth/src/hv/kvm/vm/vm_x86_64/vm_x86_64.rs

@@ -22,7 +22,7 @@ use snafu::ResultExt;
 
 use crate::arch::intr::{MsiAddrHi, MsiAddrLo};
 use crate::arch::ioapic::NUM_PINS;
-use crate::hv::kvm::sev::SevFd;
+use crate::hv::kvm::x86_64::sev::SevFd;
 use crate::hv::kvm::{KvmVm, kvm_error};
 use crate::hv::{Coco, Kvm, Result, VmConfig, error};
 use crate::sys::kvm::{