* Kirill A. Shutemov <[email protected]> wrote:
> On Thu, Sep 28, 2017 at 10:25:14AM +0200, Ingo Molnar wrote:
> >
> > * Kirill A. Shutemov <[email protected]> wrote:
> >
> > > For boot-time switching between paging modes, we need to be able to
> > > adjust size of physical address space at runtime.
> > >
> > > As part of making physical address space size variable, we have to make
> > > X86_5LEVEL dependent on SPARSEMEM_VMEMMAP. !SPARSEMEM_VMEMMAP
> > > configuration doesn't work well with variable MAX_PHYSMEM_BITS.
> > >
> > > Affect on kernel image size:
> > >
> > > text data bss dec hex filename
> > > 10710340 4880000 860160 16450500 fb03c4 vmlinux.before
> > > 10710666 4880000 860160 16450826 fb050a vmlinux.after
> > >
> > > Signed-off-by: Kirill A. Shutemov <[email protected]>
> > > ---
> > > arch/x86/Kconfig | 1 +
> > > arch/x86/include/asm/pgtable_64_types.h | 2 +-
> > > arch/x86/include/asm/sparsemem.h | 9 ++-------
> > > arch/x86/kernel/setup.c | 5 ++---
> > > 4 files changed, 6 insertions(+), 11 deletions(-)
> > >
> > > diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> > > index 6a15297140ff..f75723d62c25 100644
> > > --- a/arch/x86/Kconfig
> > > +++ b/arch/x86/Kconfig
> > > @@ -1403,6 +1403,7 @@ config X86_PAE
> > > config X86_5LEVEL
> > > bool "Enable 5-level page tables support"
> > > depends on X86_64
> > > + depends on SPARSEMEM_VMEMMAP
> >
> > Adding a 'depends on' to random kernel internal implementational details,
> > to
> > support new hardware, sucks as an UI, as it will just randomly hide/show
> > the new
> > hardware option if certain magic Kconfig combinations are set.
> >
> > Please check how other architectures are doing it. (Hint: they are using
> > select.)
> >
> > Also, what is the real dependency here? Why don't the other memory models
> > work,
> > what's the failure mode - won't build, won't boot, or misbehaves in some
> > other
> > way?
>
> I won't build.
>
> For !SPARSEMEM_VMEMMAP SECTIONS_WIDTH depends on MAX_PHYSMEM_BITS:
>
> SECTIONS_WIDTH
> SECTIONS_SHIFT
> MAX_PHYSMEM_BITS
>
> And SECTIONS_WIDTH is used on per-processor stage, it doesn't work if it's
> dyncamic. See include/linux/page-flags-layout.h.
Ok, this would be a good addition to the changelog.
Thanks,
Ingo
