Date: Tue, 9 Dec 2003 13:50:39 +1100
From: Nathan Scott <nathans@sgi.com>
To: marat <nospam-259928-001@slb.com>
Cc: linux-xfs@oss.sgi.com
Subject: Re: kernel panic:kmem_zone_zalloc: NULL memory on KM_SLEEP request!
Message-ID: <20031209025039.GB1798@frodo>
References: <200312081317.26603.nospam-259928-001@slb.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <200312081317.26603.nospam-259928-001@slb.com>
User-Agent: Mutt/1.5.3i
Sender: linux-xfs-bounce@oss.sgi.com
Errors-to: linux-xfs-bounce@oss.sgi.com
Precedence: bulk
Content-Length: 1191
Lines: 33

On Mon, Dec 08, 2003 at 01:17:26PM +0100, marat wrote:
> Experts,
> 
> Recently, I had one of the xfs filesystem failed and the only  message was  :
> kernel panic:kmem_zone_zalloc: NULL memory on KM_SLEEP request!  
> 
> I had to reboot machine and run xfs_repair to make files system available again.
> 
> I am using kernel 2.4.20 SMP and xfs version 1.3.0.
> 
> Any ideas, suggestions how I can avoid this problem to happen again ?
> 

Well, whats happened here is XFS is allocating memory from
a context where it cannot fail, and retries the allocation
a number of times (6 IIRC, with delays in-between each) and
after freeing up everything that XFS can itself.  Afterward
if we still cannot get any memory allocated we give up and
call BUG().

I guess we could keep retrying indefinitely, but that doesn't
seem like a particularly good solution.  I suspect we should
be getting a bit more help from the VM here (more recent 2.4
kernels might make this case better) ... you may be able to
tweak some variables in mm/vmscan.c too - these are likely
sysctls somewhere - vm_mapped_ratio, vm_vfs_scan_ratio, etc
(ah, have a look through "vm_table" in kernel/sysctl.c).

cheers.

-- 
Nathan