Re: [PATCH v19 078/130] KVM: TDX: Implement TDX vcpu enter/exit path

[Huang, Kai]

On Thu, 2024-04-04 at 15:45 -0700, Sean Christopherson wrote:
> On Thu, Apr 04, 2024, Kai Huang wrote:
> > On Thu, 2024-04-04 at 16:22 +0300, Kirill A. Shutemov wrote:
> > > On Mon, Feb 26, 2024 at 12:26:20AM -0800, isaku.yamahata@xxxxxxxxx wrote:
> > > > @@ -491,6 +494,87 @@ void tdx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
> > > >  	 */
> > > >  }
> > > >  
> > > > +static noinstr void tdx_vcpu_enter_exit(struct vcpu_tdx *tdx)
> > > > +{
> > > 
> > > ...
> > > 
> > > > +	tdx->exit_reason.full = __seamcall_saved_ret(TDH_VP_ENTER, &args);
> > > 
> > > Call to __seamcall_saved_ret() leaves noinstr section.
> > > 
> > > __seamcall_saved_ret() has to be moved:
> > > 
> > > diff --git a/arch/x86/virt/vmx/tdx/seamcall.S b/arch/x86/virt/vmx/tdx/seamcall.S
> > > index e32cf82ed47e..6b434ab12db6 100644
> > > --- a/arch/x86/virt/vmx/tdx/seamcall.S
> > > +++ b/arch/x86/virt/vmx/tdx/seamcall.S
> > > @@ -44,6 +44,8 @@ SYM_FUNC_START(__seamcall_ret)
> > >  SYM_FUNC_END(__seamcall_ret)
> > >  EXPORT_SYMBOL_GPL(__seamcall_ret);
> > >  
> > > +.section .noinstr.text, "ax"
> > > +
> > >  /*
> > >   * __seamcall_saved_ret() - Host-side interface functions to SEAM software
> > >   * (the P-SEAMLDR or the TDX module), with saving output registers to the
> > 
> > Alternatively, I think we can explicitly use instrumentation_begin()/end()
> > around __seamcall_saved_ret() here.
> 
> No, that will just paper over the complaint.  Dang it, I was going to say that
> I called out earlier that tdx_vcpu_enter_exit() doesn't need to be noinstr, but
> it looks like my brain and fingers didn't connect.
> 
> So I'll say it now :-)
> 
> I don't think tdx_vcpu_enter_exit() needs to be noinstr, because the SEAMCALL is
> functionally a VM-Exit, and so all host state is saved/restored "atomically"
> across the SEAMCALL (some by hardware, some by software (TDX-module)).
> 
> The reason the VM-Enter flows for VMX and SVM need to be noinstr is they do things
> like load the guest's CR2, and handle NMI VM-Exits with NMIs blocks.  None of
> that applies to TDX.  Either that, or there are some massive bugs lurking due to
> missing code.

Ah right.  That's even better :-)

Thanks for jumping in and pointing out!