Re: [PATCH v9 07/18] x86/virt/tdx: Do TDX module per-cpu initialization

[Peter Zijlstra]

On Mon, Feb 13, 2023 at 10:07:30AM -0800, Dave Hansen wrote:
> On 2/13/23 03:59, Kai Huang wrote:
> > @@ -247,8 +395,17 @@ int tdx_enable(void)
> >  		ret = __tdx_enable();
> >  		break;
> >  	case TDX_MODULE_INITIALIZED:
> > -		/* Already initialized, great, tell the caller. */
> > -		ret = 0;
> > +		/*
> > +		 * The previous call of __tdx_enable() may only have
> > +		 * initialized part of present cpus during module
> > +		 * initialization, and new cpus may have become online
> > +		 * since then.
> > +		 *
> > +		 * To make sure all online cpus are TDX-runnable, always
> > +		 * do per-cpu initialization for all online cpus here
> > +		 * even the module has been initialized.
> > +		 */
> > +		ret = __tdx_enable_online_cpus();
> 
> I'm missing something here.  CPUs get initialized through either:
> 
>  1. __tdx_enable(), for the CPUs around at the time
>  2. tdx_cpu_online(), for hotplugged CPUs after __tdx_enable()
> 
> But, this is a third class.  CPUs that came online after #1, but which
> got missed by #2.  How can that happen?

offline CPUs, start TDX crap, online CPUs.