Re: Introducing non-RAM managed page ("device page")

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Tue, Sep 15, 2009 at 04:55:50PM +0000, Eduardo Horvath wrote:
> On Tue, 15 Sep 2009, Joerg Sonnenberger wrote:
> 
> > On Tue, Sep 15, 2009 at 04:04:15PM +0000, Eduardo Horvath wrote:
> > > We've handled mmap()ing device memory forever without getting VM 
> > > involved.  
> > > Not to imply that the current device mmap() interface could not be 
> > > improved.  What features of the VM subsystem do you think is needed to 
> > > map 
> > > device pages into processes?
> > 
> > This is *not* mmap on a device. It is mmap of a file stored on a special
> > device. In other words, read-only access to this device can be done
> > without the buffer cache and for that the VM subsystem needs glue.
> 
> So the purpose of this excercise is to allow the buffer cache to use 
> non-VM pages?  Nifty, but does the memory you gain by not needing buffer 
> cache pages offset the additional code and control structures?

Yes. E.g. if you have a statically linked binary like /rescue/*, it
avoids buffering for 4MB or so :)

Joerg