Re: new device driver features

Jon Smirl wrote:
> We might also want to try building a model where the server doesn't need to run
> with root priv. What would be the issues with removing the need for root?
> 

There's the longstanding bogey man of device security.  Most graphics cards 
can be convinced to write arbitary to any physical address in main memory.

Stopping them from doing this means parsing and understanding the implications 
of all the commands submitted by the client at some point before those 
commands reach the hardware (ie the kernel module.

The other option is to restrict access to the device to root priv processes, 
which can already write to any address without the benefit of insecure hardware.

Hence the two choices - ugly & probably incomplete command-stream checking 
code in the drm modules, vs. root-priv X server (and utah-glx, if you remember 
that one).

Keith