405GP f_dentry panics

[Frank Rowand]

Glenn Toshi Martin wrote:
>
> Hi all,
>
> We have a custom 405GP board and the "Walnut" board. We are using the
> 2.4-test2 (Version released Nov 26 2000) kernel supplied via the Montavista
> web site. Our custom board has an IDE harddisk while the Walnut board uses
> NFS. We are experiencing f_dentry = 0 problems when untarring a 750 MB file
> (This seems to lead to memory corruption and hence kernel panics). The
> problem is more pronounced when using the IDE but nevertheless occurs when
> using the Walnut board over NFS. This appears to be a known problem, but
> would like to know whether the problem can be fixed, and if so what is
> required to fix it? Or is there a better workaround to the ones currently
> available?
> We have another problem that may relate to the above. During the untar
> process of large files including the one mentioned above, we get a file
> system error "Could not write to file: bad file descriptor" every now and
> then (both NFS and IDE, more frequently on the IDE). Could anyone shed light
> on this problem?
>
> Thanks
>
> Glenn

We didn't implement the "f_dentry == 0" workarounds everywhere that the problem
could manifest itself.  We just implemented it opportunistically in places
where we saw the problem under stress testing.  We expect the problem to be
resolved when we move up to 2.4.0 (I'm doing the merge now...).  If you want
to mask the problem in the interim, you need to search out all places
where f_dentry is dereferenced and check for the value of zero before
allowing the dereference.

-Frank
--
Frank Rowand <frank_rowand at mvista.com>
MontaVista Software, Inc

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/