[Yaffs] Kernel panic in clear_inode when using yaffs as rootfs

[Andre Renaud]

Charles Manning wrote:
> On Wednesday 09 November 2005 17:06, Andre Renaud wrote:
> 
>>I'm using yaffs as the root filesystem for a unit, and it is
>>experiencing repeatible crashes - always in the same place
>>(fs/inode.c:252, clear_inode).
>>
>>This is in Linux 2.6.14, with the latest yaffs from CVS (although not
>>the latest MTD, just the MTD that 2.6.14 shipped with). Using yaffs2 in
>>yaffs1 compatibility mode (512byte NAND).
>>
>>I've attached the BUG output. Anyone have any ideas? The line that the
>>bug reports to is:
>>if (inode->i_data.nrpages)
>>	BUG();
>>so inode->i_data.nrpages should be 0 when clear_inode is called - but it
>>isn't for some reason.
> 
> 
> Curious.
> 
> Is this code new/different in 2.6.14 compared with the last kernel tree you 
> used?
> 
> What is different when being used as rootfs?
> 
> Is it 100% reproducable?

Prior to 2.6.14, we were using 2.6.11.7, but that was with the original
yaffs1 code (a version of which we'd been using for about 18 months with
no real problems).

Unfortunately I haven't tried the latest yaffs with our older kernel.
I'm not sure what hte difference is between it being a rootfs and not,
it is just the circumstances under which the fault happened.

It is 100% reproducable, but not in a nice small test case. I have a
unit here that I can boot, and it will fail 100% of the time on startup
in the same way.

Andre