Re: LOR: ufs vs bufwait

From: Jeff Roberson <jroberson_at_jroberson.net>
Date: Thu, 20 May 2010 23:24:21 -1000 (HST)
On Fri, 21 May 2010, Erik Cederstrand wrote:

>
> Den 12/05/2010 kl. 22.44 skrev Jeff Roberson:
>>
>> I think Peter Holm also saw this once while we were testing SUJ and reproduced ~30 second hangs with stock sources.  At this point we need to brainstorm ideas for adding debugging instrumentation and come up with the quickest possible repro.
>
> FWIW, I get this LOR on a ClangBSD virtual machine running the stess2 test suite.
>
> I can reproduce the LOR reliably like this:
>
> # cd stress2
> #./run.sh lockf.cfg
> - press ctrl-C
> - another LOR is triggered by the ctrl-C (a dirhash/bufwait LOR described in kern/137852)
> # ./run.sh mkdir.cfg
> - LOR is triggered immediately
>
> Erik

The LOR is actually safe.  I need to bless the acquisition.  We have 
always acquired the buffers in this order.

The deadlocks people were seeing were actually livelocks due to 
softdepflush looping indefinitely.  I have committed a fix for that.

Thanks,
Jeff
Received on Fri May 21 2010 - 07:24:26 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:03 UTC