> * Dave Hansen (
haveblue@us.ibm.com) wrote:
> > On Fri, 2007-11-16 at 09:47 -0500, Mathieu Desnoyers wrote:
> > > * Dave Hansen (
haveblue@us.ibm.com) wrote:
> > > > For most (all?) architectures, the PFN and the virtual address in the
> > > > kernel's linear are interchangeable with pretty trivial arithmetic. All
> > > > pages have a pfn, but not all have a virtual address. Thus, I suggested
> > > > using the pfn. What kind of virtual addresses are you talking about?
> > > >
> > >
> > > Hrm, in asm-generic/memory_model.h, we have various versions of
> > > __page_to_pfn. Normally they all cast the result to (unsigned long),
> > > except for :
> > >
> > >
> > > #elif defined(CONFIG_SPARSEMEM_VMEMMAP)
> > >
> > > /* memmap is virtually contigious. */
> > > #define __pfn_to_page(pfn) (vmemmap + (pfn))
> > > #define __page_to_pfn(page) ((page) - vmemmap)
> > >
> > > So I guess the result is a pointer ? Should this be expected ?
> >
> > Nope. 'pointer - pointer' is an integer. Just solve this equation for
> > integer:
> >
> > 'pointer + integer = pointer'
> >
>
> Well, using page_to_pfn turns out to be ugly in markers (and in
> printks) then. Depending on the architecture, it will result in either
> an unsigned long (x86_64) or an unsigned int (i386), which corresponds
