Re: [PATCH v3 03/34] arm: mm: Add p?d_large() definitions

From: Kirill A. Shutemov
Date: Mon Mar 04 2019 - 08:11:02 EST

On Mon, Mar 04, 2019 at 11:56:13AM +0000, Steven Price wrote:
> On 01/03/2019 21:47, Kirill A. Shutemov wrote:
> > On Wed, Feb 27, 2019 at 05:05:37PM +0000, Steven Price wrote:
> >> walk_page_range() is going to be allowed to walk page tables other than
> >> those of user space. For this it needs to know when it has reached a
> >> 'leaf' entry in the page tables. This information will be provided by the
> >> p?d_large() functions/macros.
> >>
> >> For arm, we already provide most p?d_large() macros. Add a stub for PUD
> >> as we don't have huge pages at that level.
> >
> > We do not have PUD for 2- and 3-level paging. Macros from generic header
> > should cover it, shouldn't it?
> >
> I'm not sure of the reasoning behind this, but levels are folded in a
> slightly strange way. arm/include/asm/pgtable.h defines
> __ARCH_USE_5LEVEL_HACK which means:
> PGD has 2048 (2-level) or 4 (3-level) entries which are always
> considered 'present' (pgd_present() returns 1 defined in
> asm-generic/pgtables-nop4d-hack.h).
> P4D has 1 entry which is always present (see asm-generic/5level-fixup.h)
> PUD has 1 entry (see asm-generic/pgtable-nop4d-hack.h). This is always
> present for 2-level, and present only if the first level of real page
> table is present with a 3-level.
> PMD/PTE are as you might expect.
> So in terms of tables which are more than one entry you have PGD,
> (optionally) PMD, PTE. But the levels which actually read the table
> entries are PUD, PMD, PTE.
> This means that the corresponding p?d_large() macros are needed for
> PUD/PMD as that is where the actual entries are read. The asm-generic
> files provide the definitions for PGD/P4D.

Makes sense.

Only additional thing worth nothing that ARM in 2-level paging case folds
PMD manually without help from generic headres.

I'm partly responsible for the mess with folding. Sorry that you need to
explain this to :P

Kirill A. Shutemov