About variable "IOPortBase"

[Michael Lorenz]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

On Mar 27, 2008, at 03:15, belcon zhao wrote:

> Finally, got this question fixed.

Good :)

> It is very clear that compiled Xorg with MIPS, IOPortBase hasn't  
> been initialized while it is used by outb() and inb(). Don't know why.

Several reasons:
- - not all graphics controllers need IOPortBase, for instance pretty  
much all ATI chips have everything you need in memory-mapped PCI  
resources
- - the address where you find your IO space isn't CPU-dependent at  
all, it has absolutely nothing to do with MIPS, in fact we have the  
same problem on PowerPC, ARM, to a lesser degree on SPARC and a bunch  
of others. It depends on your PCI host bridge and in some cases on  
your firmware, there is no generic way to find it. You need to know  
your hardware and adapt Xorg.
On NetBSD we get around this by defining a magic offset you mmap  
through /dev/tty* and let the kernel return the appropriate pages  
( it knows which bus the graphics chips sits on and that way it also  
knows which host bridge it's on and what IO range to return - nothing  
keeps you from having more than one and then you have an IO range for  
each ) so userland ( and therefore Xorg ) doesn't need to know where  
it really is. Linux might have something similar.

have fun
Michael
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (Darwin)

iQEVAwUBR+uYdMpnzkX8Yg2nAQIOXQf/UEFEG9PAdcQPM+oS/OQ5KIagrFnYSgl7
qMbVK9J2oFwXPFca5zdelzSRLUK7N2LZy0VwAB/2aE741CEzCcG1GjcmuMv5W8SL
BzBU3LrIotOiF4WS749vGWi0opq2osO0SySzTaBhkr0s4c/moTAREmo29BxH4iR9
Kl9TQwikqebEceQqCBopfRVA0ddLqAaiD2pt+0nl8IVRKgJ2vYzfokBtTqVj3iPS
QTDwRtU8FvfBMwAFtwvmEwNvC6NPtSuo9Ki1sZSJmagp04MU2aC5sFsAYV3jkSLw
/35KUUcuCVoKDgCMHNVRi1dni9d3gyhisv4oDWjvcRv8s9M1c07oOQ==
=75G7
-----END PGP SIGNATURE-----