I-Jong Lin wrote:
> Dear Kernel-hacking types,
> 
>    The kernel that I've built for my CArrera040 accelerator works fine
> in single user mode.  Unfortunately, it seems that when the kernel
> is able to start writing to disks, I run into a problem.  
> It works for a little while (I can pack ten files out of tarball)
> and then this happens:
> 
> 	/: bad dir ino 4160 at offset 512: mangled entry
> 
> and then screws up the whole partition, making unbootable and unreadable.

perhaps something still thinks it's an '040?
 
> So, I have a couple of questions:
> 
> 1.  What could cause this problem?  (Keep in mind that I'm running
> a 90Mhz '040 and the booter could possibly be sending the wrong
> information).
> 
> 2.  What files in the source code control writes to an ffs filesystem?

well, quite a few, actually.  take a look at src/sys/ufs, i think.  also,
you'll want to look at the SCSI drivers.....in all reality, this could be
really nasty to debug :-(

> 3.  Does the code use the processor number information that the Booter
> passes in?  I didn't see any places, but I might not have been looking
> hard enough.

hmmm...don't know about this one.


[snip]
> starting nfs daemons:.
> creating runtime link editor directory cache.
> /: bad dir ino 4160 at offset 512: mangled entry
> panic: bad dir
> Stopped at 		_Debugger+0x6:	unlk a6
> db>
>  (computer freezes)

can you do a 'trace' in the debugger at this point?  it might help.

later.

-- 
Colin Wood                                 cwood@ichips.intel.com
Component Design Engineer - PMD                 Intel Corporation
-----------------------------------------------------------------
I speak only on my own behalf, not for my employer.