On Wed, Dec 29, 2021, Yang Zhong wrote:

+static void handle_nm_fault_irqoff(struct kvm_vcpu *vcpu)
+{
+	rdmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
+	kvm_queue_exception(vcpu, NM_VECTOR);

This is still wrong, even though no additional supported is needed to support
nested XFD.  If L1 wants to intercept #NM, then KVM must not inject the #NM and
must not read XFD_ERR.

That this was posted multiple times is disturbing, because kvm-unit-tests has a
test for exactly this, and running it against this series on a host without XFD
yields:

  unchecked MSR access error: RDMSR from 0x1c5 at rIP: 0xffffffffa02478ee (vmx_handle_exit_irqoff+0x1de/0x240 [kvm_intel])
  Call Trace:
   <TASK>
   kvm_arch_vcpu_ioctl_run+0x11a0/0x1fb0 [kvm]
   kvm_vcpu_ioctl+0x279/0x690 [kvm]
   __x64_sys_ioctl+0x83/0xb0
   do_syscall_64+0x3b/0xc0
   entry_SYSCALL_64_after_hwframe+0x44/0xae
   </TASK>