PowerMac + Radeon startup crash: PCI resource problem w/ Linux 2.6 (long)

[Benjamin Herrenschmidt]

> Adding tons of debug messages, I could verify that the reason for this  
> error is the routine xf86RegisterResources() returning a non-NULL list of  
> apparently failed resources:
> 
> <snip>
> (II) Failed Resources after driver initialization for Entity: 0
>         [0] 0   0       0x00000400 - 0x000004ff (0x100) IX[B]
> (II) xf86RegisterResources() returns 0x10702590
> </snip>
> 
> which in turn makes RADEONPreInit fail.
> 
> The interesting difference in comparison to 2.4.25-ben1 is that *earlier*  
> I see a note in that X log file
 
Yah, something doesn't like the IO resource allocated by the firmware,
but on the other hand, we don't use IOs anyway so that shouldn't be a
problem.

Can you do an lspci -vv dump as root on both kernels and send me the
output ? Also send me the output of /proc/iomem and /proc/ioports

> <snip>
> (II) LoadModule: "radeon"
> (II) Reloading /usr/X11R6/lib/modules/drivers/radeon_drv.o
> (WW) ****INVALID IO ALLOCATION**** b: 0x400 e: 0x4ff correcting^G
> (II) window:
>         [0] -1  0       0x00000000 - 0x00ffffff (0x1000000) IX[B]
> </snip>
> 
> which is missing for 2.6.9, but now xf86RegisterResources() returns NULL  
> and the rest of the startup succeeds.
> 
> Both log files as well as the lspci outputs for both kernel versions are  
> attached (tar.bz2).
> 
> I have no idea if this is a kernel or a x.org issue, but I hope that least  
> in one of the lists (sorry for cross-posting...) a real guru (Ben?!?)  
> listens and has some wisdom for me.

Seems to be a PCI issue, something X doesn't like in the PCI setup on
this machine. Strange.

One day, we should probably get rid of all the PCI mucking in X on
linux, have it just use the resources as-is provided by the kernel, with
just a kernel service/driver to handle arbitration of legacy IO enables
for legacy VGA. (We need that anyway and it's beeing worked on).

Ben.