Re: [PATCH] KVM: VMX: Require KVM_SET_TSS_ADDR being called prior to running a VCPU

[Sean Christopherson <seanjc@google.com>]

Dropped all the old maintainers from Cc.  This is one of the more impressive
displays of thread necromancy I've seen :-)

On Wed, Jun 28, 2023, Thorsten Glaser wrote:
> Hi,
> 
> I just saw a new message after reboot today and searched for it and got
> https://patchwork.kernel.org/project/kvm/patch/5142D010.7060303@web.de/
> and you might be interested that I ran into it.
> 
> [   70.747068] kvm: KVM_SET_TSS_ADDR need to be called before entering vcpu
> 
> Full dmesg attached. This is on reboot, no VMs are running yet.

Heh, there are no VMs that _you_ deliberately created, but that doesn't mean there
aren't VMs in the system.  IIRC, libvirt (or maybe systemd?) probes KVM by doing
modprobe *and* creating a dummy VM.  If whatever is creating a VM also creates a
vCPU, then the "soft" warning about KVM_SET_TSS_ADDR will trigger.

Another possibility is that KVM sefltests are being run during boot.  KVM's selftests
stuff register state to force the vCPU into 64-bit mode, and so they don't bother
setting KVM_SET_TSS_ADDR, e.g. the soft warning can be triggered by doing

  sudo modprobe kvm_intel unrestricted_guest=0

and running pretty much any KVM selftest.

> I don’t get any extra weird messages on VM start, and they come up fine.

So long as the VMs you care about don't have issues, the message is completely
benign, and expected since you are running on Nehalem, which doesn't support
unrestricted guest.

> [    0.000000] DMI: Gigabyte Technology Co., Ltd. X58-USB3/X58-USB3, BIOS F5 09/07/2011

> [    0.330068] smpboot: CPU0: Intel(R) Core(TM) i7 CPU         950  @ 3.07GHz (family: 0x6, model: 0x1a, stepping: 0x5)

> [   70.747068] kvm: KVM_SET_TSS_ADDR need to be called before entering vcpu