Re: [PATCHv3 09/17] x86/mm: Implement page_keyid() using page_ext

[Kirill A. Shutemov]

On Mon, Jun 18, 2018 at 12:54:29PM +0000, Dave Hansen wrote:
> On 06/18/2018 03:07 AM, Kirill A. Shutemov wrote:
> > On Wed, Jun 13, 2018 at 06:20:10PM +0000, Dave Hansen wrote:
> >>> +int page_keyid(const struct page *page)
> >>> +{
> >>> +	if (mktme_status != MKTME_ENABLED)
> >>> +		return 0;
> >>> +
> >>> +	return lookup_page_ext(page)->keyid;
> >>> +}
> >>> +EXPORT_SYMBOL(page_keyid);
> >> Please start using a proper X86_FEATURE_* flag for this.  It will give
> >> you all the fancy static patching that you are missing by doing it this way.
> > There's no MKTME CPU feature.
> 
> Right.  We have tons of synthetic features that have no basis in the
> hardware CPUID feature.

I've tried the approach, but it doesn't fit here.

We enable MKTME relatively late during boot process -- after page_ext as
page_keyid() depends on it. Enabling it earlier would make page_keyid()
return garbage.

By the time page_ext initialized, CPU features is already handled and
setup_force_cpu_cap() doesn't do anything.

I've implemented the enabling with static key instead.

-- 
 Kirill A. Shutemov