[Top] [All Lists]

Re: Oops while running xfstests on Power architecture

To: Chandra Seetharaman <sekharan@xxxxxxxxxx>
Subject: Re: Oops while running xfstests on Power architecture
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Fri, 4 Feb 2011 21:56:46 +1100
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <1296778668.32230.18.camel@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
References: <1296778668.32230.18.camel@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.20 (2009-06-14)
On Thu, Feb 03, 2011 at 04:17:48PM -0800, Chandra Seetharaman wrote:
> Hello All,
> I see the following Oops when I run the test 070 on power Architecture.
> Looking at the code, I do not see anything that is popping out as a
> problem.
> I do not see the same failures in x86_64, so, I thought it may be due to
> the pagesize, so changed the pagesize to be 4k, but that didn't help.
> Is this assertion a false-positive ?
> If not, Any suggestions on where to start looking at the code.
> I see the same Oops with 2.6.37, 2.6.38-rc2 and 2.6.38-rc3.

It's an assert failure, not an oops:

> Feb  3 14:01:30 mytestmachine kernel: Assertion failed: args->op_flags & 
> XFS_DA_OP_OKNOENT, file: fs/xfs/xfs_da_btree.c, line: 1428

XFS_DA_OP_OKNOENT is set when an operation on the
directory/attribute can be expected to get an ENOENT error.
If it is not set, then the attribute/dirent is expected to be
present in the tree, and there's something wrong if it is not.

e.g. if you are removing an attribute, it is expected to be in the

PowerPC used to get regular QA coverage but hasn't for the past
couple of years, so it probably is a relatively recent problem.
I'd suggest the first thing to do is to try older kernels to
determine if the problem is specific to newer kernels, and if so, a
git bisect is probably your best bet to finding the commit that
introduced the problem.


Dave Chinner

<Prev in Thread] Current Thread [Next in Thread>