Re: [PATCH v14 011/113] KVM: TDX: Add C wrapper functions for SEAMCALLs to the TDX module

[Isaku Yamahata]

On Thu, Jun 08, 2023 at 01:43:27AM +0000,
"Wang, Wei W" <wei.w.wang@xxxxxxxxx> wrote:

> On Thursday, June 8, 2023 2:16 AM, Isaku Yamahata wrote:
> > On Mon, Jun 05, 2023 at 03:20:19PM +0000, "Wang, Wei W"
> > <wei.w.wang@xxxxxxxxx> wrote:
> > > > +static inline u64 kvm_seamcall(u64 op, u64 rcx, u64 rdx, u64 r8, u64 r9,
> > > > +			       struct tdx_module_output *out) {
> > > > +	u64 ret;
> > > > +
> > > > +	ret = __seamcall(op, rcx, rdx, r8, r9, out);
> > > > +	if (unlikely(ret == TDX_SEAMCALL_UD)) {
> > > > +		/*
> > > > +		 * TDX requires VMXON or #UD.  In the case of reboot or
> > > > kexec,
> > > > +		 * VMX is made off (VMXOFF) by kvm reboot notifier,
> > > > +		 * kvm_reboot(), while TDs are still running.  The callers
> > > > check
> > > > +		 * the returned error and complain.  Suppress it by returning 0.
> > > > +		 */
> > >
> > > Curious how do the callers check the returned error when " Suppress it
> > > by returning 0" here.
> > 
> > It doesn't make sense for the caller to check the error and warn when
> > kvm_rebooting = true.
> > Let's make it "return kvm_rebooting ? 0 : ret;" instread of "return 0;".
> > Does it make sense?
> 
> Seems no need. The comments look confusing, and not aligned to what
> the code achieves. From what I read:
> - if kvm_rebooting=true there: return 0 to caller and no error or warning happens
> - if kvm_rebooting=false there: crash the system via kvm_spurious_fault.
> In this non-rebooting case, I think the callers don’t get a chance to read the 
> returned value and complain.

How about this comment?

	if (unlikely(ret == TDX_SEAMCALL_UD)) {
		/*
		 * TDX requires VMXON or #UD.  In the case of reboot or kexec,
		 * kvm shutdown notifier, kvm_shutdown(), makes VMX off (VMXOFF)
		 * while TDs can be still running to invoke SEAMCALL.  It
		 * results in superfluous errors or warnings.
		 * If rebooting, return 0 to suppress superfluous messages.
		 * If not rebooting, panic by kvm_spurious_fault().
		 */
		kvm_spurious_fault();
		return 0;
	}

> Another thing is, have you double-checked that invocation of seamcalls
> indeed returns TDX_SEAMCALL_UD when VMX=off?

I observed it several times during testing tdx module initialization. Here is
the example I dug out from my debug logs.

  tdx: SEAMCALL failed: CPU 162 is not in VMX operation.
  tdx: TDX module initialization failed (-22)
  kvm_intel: Failed to initialize TDX module.

-- 
Isaku Yamahata <isaku.yamahata@xxxxxxxxx>