+Chao

On Fri, Aug 01, 2025, Adrian Hunter wrote:

On 29/07/2025 22:33, Sean Christopherson wrote:

quoted

+static int tdx_terminate_vm(struct kvm *kvm)
+{
+	if (kvm_trylock_all_vcpus(kvm))
+		return -EBUSY;
+
+	kvm_vm_dead(kvm);
+	to_kvm_tdx(kvm)->vm_terminated = true;
+
+	kvm_unlock_all_vcpus(kvm);
+
+	tdx_mmu_release_hkid(kvm);
+
+	return 0;
+}

As I think I mentioned when removing vm_dead first came up,
I think we need more checks.  I spent some time going through
the code and came up with what is below:

First, we need to avoid TDX VCPU sub-IOCTLs from racing with
tdx_mmu_release_hkid().  But having any TDX sub-IOCTL run after
KVM_TDX_TERMINATE_VM raises questions of what might happen, so
it is much simpler to understand, if that is not possible.
There are 3 options:

1. Require that KVM_TDX_TERMINATE_VM is valid only if
kvm_tdx->state == TD_STATE_RUNNABLE.  Since currently all
the TDX sub-IOCTLs are for initialization, that would block
the opportunity for any to run after KVM_TDX_TERMINATE_VM.

2. Check vm_terminated in tdx_vm_ioctl() and tdx_vcpu_ioctl()

3. Test KVM_REQ_VM_DEAD in tdx_vm_ioctl() and tdx_vcpu_ioctl()

[ Note cannot check is_hkid_assigned() because that is racy ]

Secondly, I suggest we avoid SEAMCALLs that will fail and
result in KVM_BUG_ON() if HKID has been released.

There are 2 groups of those: MMU-related and TDVPS_ACCESSORS.

For the MMU-related, the following 2 functions should return
an error immediately if vm_terminated:

	tdx_sept_link_private_spt()
	tdx_sept_set_private_spte()

For that not be racy, extra synchronization is needed so that
vm_terminated can be reliably checked when holding mmu lock
i.e.

static int tdx_terminate_vm(struct kvm *kvm)
{
	if (kvm_trylock_all_vcpus(kvm))
		return -EBUSY;

	kvm_vm_dead(kvm);
+
+       write_lock(&kvm->mmu_lock);
	to_kvm_tdx(kvm)->vm_terminated = true;
+       write_unlock(&kvm->mmu_lock);

	kvm_unlock_all_vcpus(kvm);

	tdx_mmu_release_hkid(kvm);

	return 0;
}

Finally, there are 2 TDVPS_ACCESSORS that need avoiding:

	tdx_load_mmu_pgd()
		skip td_vmcs_write64() if vm_terminated

	tdx_protected_apic_has_interrupt()
		skip td_state_non_arch_read64() if vm_terminated

Oof.  And as Chao pointed out[*], removing the vm_dead check would allow creating
and running vCPUs in a dead VM, which is most definitely not desirable.  Squashing
the vCPU creation case is easy enough if we keep vm_dead but still generally allow
ioctls, and it's probably worth doing that no matter what (to plug the hole where
pending vCPU creations could succeed):

diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index d477a7fda0ae..941d2c32b7dc 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c

@@ -4207,6 +4207,11 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, unsigned long id)
 
        mutex_lock(&kvm->lock);
 
+       if (kvm->vm_dead) {
+               r = -EIO;
+               goto unlock_vcpu_destroy;
+       }
+
        if (kvm_get_vcpu_by_id(kvm, id)) {
                r = -EEXIST;
                goto unlock_vcpu_destroy;

And then to ensure vCPUs can't do anything, check KVM_REQ_VM_DEAD after acquiring
vcpu->mutex.

diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 6c07dd423458..883077eee4ce 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c

@@ -4433,6 +4433,12 @@ static long kvm_vcpu_ioctl(struct file *filp,
 
        if (mutex_lock_killable(&vcpu->mutex))
                return -EINTR;
+
+       if (kvm_test_request(KVM_REQ_VM_DEAD, vcpu)) {
+               r = -EIO;
+               goto out;
+       }
+
        switch (ioctl) {
        case KVM_RUN: {
                struct pid *oldpid;

That should address all TDVPS paths (I hope), and I _think_ would address all
MMU-related paths as well?  E.g. prefault requires a vCPU.

Disallowing (most) vCPU ioctls but not all VM ioctls on vm_dead isn't great ABI
(understatement), but I think we need/want the above changes even if we keep the
general vm_dead restriction.  And given the extremely ad hoc behavior of taking
kvm->lock for VM ioctls, trying to enforce vm_dead for "all" VM ioctls seems like
a fool's errand.

So I'm leaning toward keeping "KVM: Reject ioctls only if the VM is bugged, not
simply marked dead" (with a different shortlog+changelog), but keeping vm_dead
(and not introducing kvm_tdx.vm_terminated).

Thoughts?

[*] https://lore.kernel.org/all/aIlzeT+yFG2Tvb3%2F@intel.com (local)