G5 fan problems return moving to 2.6.15 with dual processor 2.7GHz machine

[Benjamin Herrenschmidt]

> I'm guessing I should have seen a "found U3-0", but I see a suspicious message here:
>     /u3 at 0,f8000000/i2c at f8001000: Missing interrupt or address !
> that I do not see in the working 2.6.14 boot.
> 
> I was wondering if the change from module to builtin was causing the
> problem (grasping at straws I guess) so I also tried building it as a
> module. I get the similar results:
> 
> Feb  4 12:59:26 youngmc kernel: /u3 at 0,f8000000/i2c at f8001000: Missing interrupt or address !
> Feb  4 12:59:26 youngmc kernel: Found KeyWest i2c on "mac-io", 1 channel, stepping: 4 bits

That looks like a fix in prom_init.c that is missing from 2.6.15.. I'll
have to double check... Apple indeed seems to have a "bug" in the
device-tree of some of those machines. prom_init.c has some code to fix
it up, but there have been several versions of the fix and maybe that
broke some way...

Ah... now I'm reading the rest of the message and see that you figured
that out already ;)

> Now this "/u3 at 0,f8000000/i2c at f8001000: Missing interrupt or address !"
> warning I'm seeing in both cases looked familar, in fact I was on a
> thread about it when the 2.7GHz machines first came out:
> 
>     http://patchwork.ozlabs.org/linuxppc64/patch?id=1982
> 
> The code that this patched applied to has moved to a new location
> arch/powerpc/kernel/prom_init.c, but logically it still seems like it
> should cover my case. The code says:
> 
>     if (u3_rev < 0x35 || u3_rev > 0x39)
>         return;
> 
> and my u3_rev looks to be 0x35
>     $ hexdump /proc/device-tree/u3 at 0,f8000000/device-rev
>     0000000 0000 0035
>     0000004
> 
> Unforunately it looks like I need to use prom_print to add debugging,
> which I'm guessing only comes to the console which I'm not near right
> now.
> 
> Before going further, is there something obvious that the Fedora
> 2.6.15 kernel is doing wrong, given that the 2.6.14 kernel works and
> the 2.6.15 seems to have a regression? I'm willing to do some more
> debugging or try a more up-to-date kernel to help resolve this issue.
> 
> One last note, my dual processor 2.0GHz and 2.5GHz machines are running
> fine with 2.6.15...

Might be something in that prom_init.c fix that broke... it would be
really nice if you could give a try with the console and find out what
it is ... Unfortunately, I don't have access to one of these machines
with the "problem" at the moment...

Cheers,
Ben.