Re: [PATCH v2 2/7] KVM: x86: Extract VMXON and EFER.SVME enablement to kernel

[Huang, Kai]

On Fri, 2025-12-19 at 07:40 -0800, Sean Christopherson wrote:
> Kai, question for you (or anyone else that might know):
> 
> Is there any **need** for tdx_cpu_enable() and try_init_module_global() to run
> with IRQs disabled?  AFAICT, the lockdep_assert_irqs_disabled() checks added by
> commit 6162b310bc21 ("x86/virt/tdx: Add skeleton to enable TDX on demand") were
> purely because, _when the code was written_, KVM enabled virtualization via IPI
> function calls.
> 
> But by the time the KVM code landed upstream in commit fcdbdf63431c ("KVM: VMX:
> Initialize TDX during KVM module load"), that was no longer true, thanks to
> commit 9a798b1337af ("KVM: Register cpuhp and syscore callbacks when enabling
> hardware") setting the stage for doing everything from task context.

The other intention was to make tdx_cpu_enable() as IRQ safe, since it was
supposed to be able to be called by multiple in-kernel users but not just
KVM (i.e., also for TDX connect).

But right currently we don't need to call them with IRQ disabled.