> > > 2. Speed, lots calls to ioperm()/iopl() can't be good for
> > > performance
>
> A single call to ioperm()/iopl() isn't good for portability.
Here's an idea: how about a _small_ kernel driver that mediates access to
the range of I/O ports likely to be used by the video card? Only one
device could open the driver at a time (that stops two processes from
grabbing the screen at the same time) and simple read/writes could read
and write to I/O ports in the specified range. (Maybe a write of 4 bytes
means output second word to the I/O port named by the first word.)
_Everything_ else would go into user-space. All the kernel does is provide
access to the I/O ports in question and keeps more then one program from
using them at a time.
-- Kenneth Albanowski (kjahds@kjahds.com, CIS: 70705,126)