> >> yeah, right... then you need an XML parser in the driver... somehow I
> >> don't think the kernel people will think it's a good idea...
> >=20
> > I agree. KISS.
> >=20
> > Peter
> 
> Why in the kernel? A perfect driver can be created from the XML
> description in userland, compile & insmod. Well optimized for the given
> hw and no overhead implementing support for something what is not in HW
> (fpga).

If this is a graphics card, and you are using it as the system console,
you need to use it to print info from the firmware, before you have
an OS, much less a userland.

If you are not using the device as the system console, then the
firmware can ignore it.

> The driver must be a module (if handles some system level functionality
> - e.g. IDE card), but i'll rather want to see it as a library on the top
> of some general OGA-PCI driver (for application level - e.g. hw
> compression on fpga - that does not need a kernel driver and such things
> should not go into kernel).
> 
> I don't think, that the kernel is ready for runtime-reconfigurable hw
> (fpga boards), so be open to new ideas ;)

Perhaps this could be handled like hotpluggable devices?  (SCSI, Firewire,
USB, ...)

There seems to be a trend towards handling USB devices with a generic
device driver and a userland library.  This has the advantage of keeping
the OS specific device driver smaller.  The userland library would be
portable across *BSD, Linux, etc.  A similar strategy could be useful,
here, if it doesn't hurt performance too much.
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to