xfs
[Top] [All Lists]

Re: [2.6.33.3] scheduling while atomic (inode reclaim races still?)

To: Simon Kirby <sim@xxxxxxxxxx>
Subject: Re: [2.6.33.3] scheduling while atomic (inode reclaim races still?)
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Thu, 20 May 2010 05:03:28 -0400
Cc: xfs@xxxxxxxxxxx
In-reply-to: <20100520083104.GA4723@xxxxxxxxxx>
References: <20100520083104.GA4723@xxxxxxxxxx>
User-agent: Mutt/1.5.19 (2009-01-05)
On Thu, May 20, 2010 at 01:31:04AM -0700, Simon Kirby wrote:
> This started happening on host about 10 days after upgrading to 2.6.33.3
> (in hopes that it fixed all of the reclaim issues present in 2.6.33.2). 
> I don't see any fixes in 2.6.33.4 relevant to this particular issue.
> 
> After this first error, the kernel logs just filled with repeat
> occurrences with different backtraces.
> 
> BUG: scheduling while atomic: nfsd/29671/0x00000250

The bug unfortunately doesn't tell us what caused the "atomic"
situation.  But given that XFS doesn't take any spinlock in that path
it seems like this must be caused by nfsd.

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