Re: [PATCH v1] mm/pagewalk: fix usage of pmd_leaf()/pud_leaf() without present check

[Jann Horn <jannh@google.com>]

On Tue, Oct 15, 2024 at 4:40 PM David Hildenbrand <david@redhat.com> wrote:
> On 15.10.24 16:32, Kirill A. Shutemov wrote:
> > On Tue, Oct 15, 2024 at 01:12:36PM +0200, David Hildenbrand wrote:
> >> pmd_leaf()/pud_leaf() only implies a pmd_present()/pud_present() check on
> >> some architectures.
> >
> > Should we clarify what behaviour we actually want from arch code?
>
> We probably should document somewhere that things like pmd_special(),
> pmd_leaf() ... should only be used when we know that the PMD is present.
>
> I wonder if we should even add ways to detect mis-use
>
> Jann also raised that recently in a private message, that it is rather
> unclear (well, and repeatedly leads to issues) when pmd_leaf() is valid
> to be called.

I think one place where that should probably be addressed is in
Documentation/mm/arch_pgtable_helpers.rst - that is supposed to be an
overview of these helper functions and what they mean, but the only
thing it currently says about pmd_leaf() is "Tests a leaf mapped PMD",
which doesn't really tell you much more than the function name. It
would be nice if that table contained information about the conditions
under which these helpers may be used.