[lug] bizarre 2.6.18 config message

[D. Stimits]

Michael J. Hammel wrote:

>Semi-educated comments follow:
>
>On Sun, 2006-10-08 at 00:35 +0000, D. Stimits wrote:
>
>  
>
>>Time: tsc clocksource has been installed.
>>Creating initial device nodes
>>Setting up hotplug.
>>Creating block device nodes.
>>Loading scsi_transport_spi.ko module
>>scsi_transport_spi: no version for "struct_module" found: kernel tainted.
>>    
>>
>
>Not sure about this.  I thought all modules got a module version by
>default that they could override with MODULE_VERSION.  In the 2.6.17.1
>source I have, the scsi_transport_spi.c doesn't seem to reference
>MODULE_VERSION internally or in any headers.  So I'd assume it was
>getting it from the default module version string (can't remember what
>that is called).  Apparently not, though.
>
>I also thought a kernel was only tainted if the MODULE_LICENSE was not
>an acceptable open source license.  I didn't know a version number
>(missing or otherwise) could cause that message.
>
>  
>
Maybe I should turn this in as a bug on kernel.org...not sure they'd 
call it a bug though.

>>Loading sym53c8xx.ko module
>>Loading jbd.ko module
>>Loading ext3.ko module
>>Creating root device.
>>Mounting root filesystem.
>>mount: could not find filesystem "/dev/root"
>>    
>>
>
>No /dev/root, which means no root filesysstem.  Without the root, you
>can't mount /dev.  This would be my guess as to what is causing all the
>other problems.  Maybe this is because the initrd doesn't have a /dev?
>
>  
>
I followed up on this, and did more research as well. One thing I did 
was to use cpio and manually unpack and inspect the initrd, and it does 
indeed have the dev subdirectory. In earlier working initrd's, there is 
no /dev/root, just /dev/<stuff>. In the non-working one, there is also 
/dev/<stuff>, and also lacks /dev/root. Research mostly indicates that 
this sort of thing is related to missing modules for LVM, but I compiled 
LVM directly into the kernel, and the same with libata. The only modules 
in the working older initrd's are directly compiled into the kernel in 
the 2.6.18 version. The partition itself is not an LVM volume, it's a 
plain vanilla ext2 partition for /boot and plain vanilla ext3 partition 
for /. ext3 is indeed present in both working and failing initrd's. 
Arguments passed in the grub.conf are exactly the same. Examining the 
ext3 lines in the initrd's show that all insmod, mkrootdev, so on, are 
exactly the same. I'm not certain though where mkrootdev is from, I 
don't see it on the system anywhere...if I grep mkrootdev though, the 
"nash" program has this in it (at least as a symbol...I don't know if it 
implements it).

Guess I'll try to package something up for a report to the kernel list, 
since it isn't a distro-specific thing.

D. Stimits, stimits AT comcast DOT net