Display specification

[Jon Smirl]

On Mon, 14 Mar 2005 13:44:09 -0500, Adam Jackson <ajax at nwnk.net> wrote:
> On Monday 14 March 2005 13:09, Jon Smirl wrote:
> > Doing something like this:
> > EGLDisplay dpy = eglGetDisplayForWidgetXXX("pci/0000:01:00.0");
> > builds a dependence in on card location. you don't want to do that.
> > <snip>
> > The basic idea is that the kernel knows about physical devices. It
> > then finds and loads the appropriate device driver. The driver then
> > exposes user space accessible /sys/class entries. All of the device
> > discovery process is removed from XGL/X.
> 
> I'm not seeing how this makes /sys/class/graphics/fb0 independent of card
> location, unless you set up those bindings by calling out to userspace.  I'm
> not insistent on being able to specify a display by bus address.  However I
> do want a deterministic way to pick a specific card independent of the
> kernel's probe order.  If that means asking for "screen/0" which is defined
> by some config file somewhere to be "the Voodoo3 as opposed to the G400",
> fine.

That's the entire reason udev was created. /etc/udev lets you write
little rules about which card will be fb0, etc. This was driven by
machines with 5,000 SCSI drives and when you add a new on they all got
renumbered. Another good use for udev is making USB peripherals appear
with a consistent name. All of this bus/slot mapping of what piece of
hardware goes to what device name is controlled by udev.

> 
> I think the idea of not needing a config file is very noble, but in practice
> it's just not going to happen for multiseat.  You cannot uniquely identify
> most HID devices by anything other than bus location, because HID devices are
> dumb and lack unique serial numbers.  You can manage this config file
> dynamically and call it a policy agent if it makes you feel better, but it's
> still userspace configuration.
> 
> Also there exist less fortunate OSes where we don't have sysfs, so we may need
> the ability to specify devices by bus address anyway.
> 
> - ajax
> 
> 
> 


-- 
Jon Smirl
jonsmirl at gmail.com