Quote:
First of all, emufx.c, of the emu10k1 driver, was bugged, with two cases in a switch that didn't appear to be integer (couldn't find out why... My C is fading...

). I commented them out (surely rendering bugged the whole driver) and went on.
http://svn.debian.org/wsvn/kernel/trunk ... rev=0&sc=0
This is the official fix.
Quote:
It ends up with a warning about a module, but it didn't seem of any importance.
The big problem is: at boot, the kerenel image does not inflate (when it says gunzipping, the second line of output after the boot command in OpenFirmware), and the error message has no other explanation. Only an error code -5.
What could that be??
This is discussed along with a few other Gentoo installation bugs, at Stuart Herbert's diary;
http://stu.gnqs.org/diary/archives/2004/12/15/77/
What happens is the OpenFirmware boots the small Linux CHRP boot loader, which decompresses a kernel image into a small area of memory. By default this is located at the 4MB boundary - OF variable "load-base" contains the absolute address. The kernel ends up decompressing over it's own code and the zipped image because of the way the firmware memory map.
What you can do is set your "load-base" variable such that it is set to the 32MB boundary (setenv load-base 0x2000000) so that it is mapped BEYOND the firmware memory area, and has plenty of space to decompress into (up to the limit of available memory).
The load-base is set so low in order to allow systems to boot with less than 32MB of memory. Unless you have a decompressed kernel more than 4MB in size you should never run into it.
Again the fix is to "setenv load-base 0x2000000" at the OF prompt.
--
Matt Sealey <
matt@genesi.co.uk>
Manager, Genesi, Developer Relations