Re: [PATCH 2/3] m68k/page_no.h: force __va argument to be unsigned long

From: Michal Hocko
Date: Tue Jul 03 2018 - 11:55:14 EST


On Tue 03-07-18 18:47:51, Mike Rapoport wrote:
> On Tue, Jul 03, 2018 at 05:14:01PM +0200, Michal Hocko wrote:
> > On Tue 03-07-18 08:05:35, Matthew Wilcox wrote:
> > > On Tue, Jul 03, 2018 at 06:03:16PM +0300, Mike Rapoport wrote:
> > > > On Tue, Jul 03, 2018 at 04:20:54PM +0200, Michal Hocko wrote:
> > > > > On Tue 03-07-18 13:29:54, Mike Rapoport wrote:
> > > > > > Add explicit casting to unsigned long to the __va() parameter
> > > > >
> > > > > Why is this needed?
> > > >
> > > > To make it consitent with other architecures and asm-generic :)
> > > >
> > > > But more importantly, __memblock_free_late() passes u64 to page_to_pfn().
> > >
> > > Why does memblock work in terms of u64 instead of phys_addr_t?
> >
> > Yes, phys_addr_t was exactly that came to my mind as well. Casting
> > physical address to unsigned long just screams for potential problems.
>
> Not sure if for m68k-nommu the physical address can really go beyond 32
> bits, but in general this is something that should be taken care of.
>
> I think adding the cast in m68k-nommu case is a viable band aid to allow
> sorting out the bootmem vs nobootmem.
>
> In any case care should be taken of all those
>
> #define __va(x) ((void *)((unsigned long)(x)))

Yeah, sounds like a good idea to me.

> all around.
>
> Regardless, I can s/u64/phys_addr_t/ in memblock.c.

Yeah, sounds like a good thing to me.

--
Michal Hocko
SUSE Labs