Re: [PATCH 0/2] FLATMEM: allow memory to start at pfn != 0 [take #2]

["Franck Bui-Huu" <vagabon.xyz@gmail.com>]

On 3/12/07, peter fuerst <post@pfrst.de> wrote:
> On Mon, 12 Mar 2007, Franck Bui-Huu wrote:
> > Can you explain why the current use of pa() failed to handle all
> > kernel address with a real example ?
>
> Simply, when you convert between cached (kseg0, ckseg0, several xkphys-
> regions) and uncached (kseg1, ckseg1, several xkphys-regions) addresses
> and the other way round, you need the physical address as an intermediate
> value and __pa() or virt_to_phys() can support only one direction.
I was asking for _real_ uses. Can you point out some code where these
convertion are needed ?

I'm asking that because your uses of __pa()/virt_to_phys() to convert
cached address into uncached address and vice versa is weird. You
talked about drivers but I would think that drivers have physical
addresses of the device they control. And they get a virtual address
by using ioremap() and Co. And using __pa() on such addresses is
simply buggy whatever the implementation of __pa().
--
              Franck