Re: [PATCHv7 10/19] x86/mm: Make __PHYSICAL_MASK_SHIFT and __VIRTUAL_MASK_SHIFT dynamic

From: Ingo Molnar
Date: Thu Sep 28 2017 - 06:42:53 EST



* Kirill A. Shutemov <kirill@xxxxxxxxxxxxx> wrote:

> On Thu, Sep 28, 2017 at 10:28:13AM +0200, Ingo Molnar wrote:
> >
> > * Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx> wrote:
> >
> > > --- a/arch/x86/mm/dump_pagetables.c
> > > +++ b/arch/x86/mm/dump_pagetables.c
> > > @@ -82,8 +82,8 @@ static struct addr_marker address_markers[] = {
> > > { 0/* VMALLOC_START */, "vmalloc() Area" },
> > > { 0/* VMEMMAP_START */, "Vmemmap" },
> > > #ifdef CONFIG_KASAN
> > > - { KASAN_SHADOW_START, "KASAN shadow" },
> > > - { KASAN_SHADOW_END, "KASAN shadow end" },
> > > + { 0/* KASAN_SHADOW_START */, "KASAN shadow" },
> > > + { 0/* KASAN_SHADOW_END */, "KASAN shadow end" },
> >
> > What's this? Looks hacky.
>
> KASAN_SHADOW_START and KASAN_SHADOW_END depend on __VIRTUAL_MASK_SHIFT,
> which is dynamic for boot-time switching case. It means we cannot
> initialize the corresponding address_markers fields compile-time, so we do
> it boot-time.

Yes, so please instead of just commenting out the values, please do something like
this:

/*
* This field gets initialized with the (dynamic) KASAN_SHADOW_END value
* in boot_fn_foo()
*/
{ 0, "KASAN shadow end" },

where boot_fn_foo() is the function where this all gets set up.

Thanks,

Ingo