Re: [PATCH] x86/mm/pti: Move user W+X check into pti_finalize()

[Joerg Roedel]

Hi Dave,

On Wed, Aug 08, 2018 at 08:54:37AM -0700, Dave Hansen wrote:
> One bit of information missing from the changelog: Could you clarify how
> there are any entries in the user page tables for the code to complain?
> Before pti_init(), I would have expected the user page tables to be empty.

The W+X check runs at the end of mark_readonly() in x86, which is after
pti_init() already put kernel mappings into the user page-table. Problem
is that the cloned entries are still W+X mapped, which is fixed in
pti_finalize() running _after_ mark_readonly().

Regards,

	Joerg