Re: Does io_remap_page_range() take 5 or 6 args?

[William Lee Irwin III <wli@holomorphy.com>]

On Wed, 18 Aug 2004 15:59:15 -0700
>> Given this, will a pfn suffice?

On Wed, Aug 18, 2004 at 04:16:58PM -0700, David S. Miller wrote:
> There is an error in the calculations.  16TB "RAM", means "RAM".
> On many systems, a large chunk of the physical address space is
> taken up by I/O areas, not real memory.
> Such areas do not take up mem_map[] array space.
> Regardless, I think an "unsigned long" page frame number is sufficient
> for now.  Don't even make the new type.

Oh, virtualspace footprint of IO areas is far worse, as the convention
is to direct map them into a single address space if they're ever used.
Of course this convention is much more loosely established than e.g.
struct page is for RAM. Some analogue of kmap_atomic() for such
machines to multiplex virtualspace in interrupt context would help, but
is unrelated to physical address passing issues.


-- wli