Re: include/asm-arm/memory.h changes break zaurus sl-5500 boot

[Russell King <rmk+lkml@arm.linux.org.uk>]

On Mon, Apr 03, 2006 at 09:15:04AM +0900, KAMEZAWA Hiroyuki wrote:
> On Mon, 3 Apr 2006 00:23:14 +0200
> Pavel Machek <pavel@ucw.cz> wrote:
> > > Not surprising given this gem:
> > > 
> > > > -#define arch_local_page_offset(pfn, nid) (LOCAL_MAP_NR((pfn) << PAGE_OFFSET))
> > > 
> > > PAGE_OFFSET being 3GB - that's one hell of a shift value!
> > 
> > Unfortunately this is mainline now. Is there some better fix than
> > simply reverting the offending patches?
> 
> Maybe this one will fix (against 2.6.16-mm2)
> 
> LOCAL_MAP_NR(kaddr) returns page offset in a node.

LOCAL_MAP_NR does not take a kernel virtual address.  If you look at how
it's defined (Eg):

#define LOCAL_MAP_NR(addr) \
        (((unsigned long)(addr) & 0x07ffffff) >> PAGE_SHIFT)

then you will notice that it's actually converting an offset into a node
into an index.  This means that converting a PFN to a virtual address is
just a complete and utter waste of time.

So:

> -#define arch_local_page_offset(pfn, nid) (LOCAL_MAP_NR((pfn) << PAGE_OFFSET))
> +#define arch_local_page_offset(pfn, nid) (LOCAL_MAP_NR(__va((pfn) << PAGE_SHIFT)))

this should be:

#define arch_local_page_offset(pfn, nid) LOCAL_MAP_NR((pfn) << PAGE_SHIFT)

Please also avoid the over-use of parens.  It's a disease which just makes
things harder to read (eg, the have I got enough close parens at the end
of the line to balance.)

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core