Re: [RFC PATCH 1/1] x86/mm: Mark CoCo VM pages invalid while moving between private and shared

[kirill.shutemov@xxxxxxxxxxxxxxx]

On Mon, Aug 28, 2023 at 11:23:55PM +0000, Michael Kelley (LINUX) wrote:
> > I guess in this case it is better to use slow_virt_to_phys(), but have a
> > comment somewhere why it is self. Maybe also add comment inside
> > slow_virt_to_phys() to indicate that we don't check present bit for a
> > reason.
> > 
> 
> OK, works for me.  I'll turn my original RFC patch into a small patch
> set that's pretty much as the RFC version is coded, along with the
> appropriate comments in slow_virt_to_phys().  Additional patches
> in the set will remove code that becomes unnecessary or that can
> be simplified.

Oh, I just realized that it conflicts with another patchset that I am
currently working on. This patchset adds kexec support for TDX. One thing
I do in this patchset is revert all shared pages to private during kexec,
so that the next kernel can use them[1]. I determine the shared status by
checking the shared bit in the kernel page tables.

If I understand correctly, your change will work with my patch because
my patch is buggy and does not check the preset bit (irony detected).

Ugh. This gets messy.

[1] https://github.com/intel/tdx/commit/4b63531fc315551c3c42023ea655206c77eef5a1
-- 
  Kiryl Shutsemau / Kirill A. Shutemov