Re: [PATCH v11 018/113] KVM: TDX: create/destroy VM structure

[Sean Christopherson]

On Thu, Jan 19, 2023, Huang, Kai wrote:
> On Thu, 2023-01-12 at 08:31 -0800, isaku.yamahata@xxxxxxxxx wrote:
> > +static void tdx_clear_page(unsigned long page_pa)
> > +{
> > +	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
> > +	void *page = __va(page_pa);
> > +	unsigned long i;
> > +
> > +	if (!static_cpu_has(X86_FEATURE_MOVDIR64B)) {
> > +		clear_page(page);
> > +		return;
> > +	}
> 
> There might be below issues here:
> 
> 1) The kernel says static_cpu_has() should only be used in fast patch where each
> cycle is counted, otherwise use boot_cpu_has().  I don't know whether here you
> should use static_cpu_has().

That documentation is stale[*], go ahead and use cpu_feature_enabled().

https://lore.kernel.org/all/20221107211505.8572-1-bp@xxxxxxxxx

> 2) IIUC a CPU feature bit can be cleared by 'clearcpuid=xxx' kernel command

As you note below, using clearcpuid taints the kernel, i.e. any breakage due to
clearcpuid is user error.

> line, so looks you should use CPUID directly otherwise the MOVDIR64B below can
> be unintentionally skipped.  In practice w/o doing MOVDIR64B is fine since KeyID
> 0 doesn't have integrity enabled, but for the purpose you want to achieve
> checking real CPUID should be better.
> 
> But maybe you don't want to do CPUID check here each time when reclaiming a
> page.  In that case you can do CPUID during module initialization and cache
> whether MOVDIR64B is truly present.  static_key is a good fit for this purpose
> too I think.
> 
> But I am also seeing below in the kernel documentation:
> 
>         clearcpuid=X[,X...] [X86]
> 			......
>                         Note that using this option will taint your kernel.
>                         Also note that user programs calling CPUID directly
>                         or using the feature without checking anything
>                         will still see it. This just prevents it from
>                         being used by the kernel or shown in /proc/cpuinfo.
>                         Also note the kernel might malfunction if you disable
>                         some critical bits.
> 
> So the kernel is claiming using this will taint the kernel and it can even
> malfunction.  So maybe it's OK to use static_cpu_has()/boot_cpu_has().