Re: Kernel 2.4.21

[Marek S. <ma...@st...>]

Hello Geert

On 26.07.03, you wrote:

> In kernel/resoure.c:request_resource(), print the contents of the passed
> `root' and `new' resources. The routine will fail if new isn't a subset of
> root, but you want to know why it happens.

I found where is the source of this problem!!!! This is/was a very long
night ;)

In 2.4.21 (or maybe earlier, but after 2.4.18) there is a new code in PCI
drivers. This code tries to "reparent" resources, that failed to allocate
(arch/ppc/kernel/pci.c, line 343). This happens to the "bridge resource" of
PM2 bridge ("PCI: Cannot allocate resource region 1 of PCI bridge 0" line
in log from kernel 2.4.18). In 2.4.21 when allocating this resource region
fails, kernel tries to "reparent" it (whatever this means) - line "PCI:
reparented B/CVisionPPC PCI mem [e0000000..fffc0000] under B/CVisionPPC PCI
mem" line in 2.4.21 log. 

There are (IMHO) 2 ways of fixing this bug:
1. remove/comment whole code which "reparents" the resources (ugly way)
2. add some code to apus_pci.c, which will set-up the pci-bridge properly,
that it's won't cause these problems (pretty way)

Actually I don't know how to do the second. I simply commented out the line
from arch/ppc/kernel/pci.c, but that caused other errors during startup.
However the pm2fb driver and XFree with glint are working now. :)))

Regards
Ps. 
I attached dmesg log from both kernels: 2.4.{18,21} 
In 2.4.21 the DEBUG in apus_pci.c was disabled, so don't be surprised, that
there is less debug stuff. 
-- 
Marek Szyprowski .. GG:2309080 .. mailto:ma...@am... ..
...... happy AmigaOS, MacOS and Debian/Linux user ........
........... http://march.home.staszic.waw.pl/ ............