Re: port-amd64/39283: Kernel crash on Dell Poweredge 2950

[Mindaugas Rasiukevicius <rmind%netbsd.org@localhost>]

The following reply was made to PR port-amd64/39283; it has been noted by GNATS.

From: Mindaugas Rasiukevicius <rmind%netbsd.org@localhost>
To: fredrik%netbsd.se@localhost
Cc: gnats-bugs%NetBSD.org@localhost, gnats-admin%netbsd.org@localhost, 
netbsd-bugs%netbsd.org@localhost
Subject: Re: port-amd64/39283: Kernel crash on Dell Poweredge 2950
Date: Tue, 24 Nov 2009 20:21:24 +0000

 Hello,
 
 >  > Synopsis: 4.99.71 crashed after about 3-4 days when running MP-kernel
 >  >
 >  > State-Changed-From-To: open->feedback
 >  > State-Changed-By: dsl%NetBSD.org@localhost
 >  > State-Changed-When: Fri, 06 Nov 2009 22:49:43 +0000
 >  > State-Changed-Why:
 >  > Any further info ??
 >  
 >  No more than the info that we put in the PR and on the mailing lists:
 >  
 >  http://archive.netbsd.se/?ml=netbsd-current-users&a=2008-07&t=7775691
 >  http://archive.netbsd.se/?ml=port-amd64&a=2009-02&t=9259482
 >  
 >  The server crashed every 8'th day on SP-kernel and every 24'th hour with
 >  MP-kernel. The problem has evolved with NetBSD releases, these days it's
 >  "only" the filesystem that seems to hang and no i/o is possible without
 >  reboot, if it runs to long without reboot is eventually crashed.
 
 Seems there is not yet enough information to figure out where bug is hiding.
 Also, it the only problem report with such symptoms, but if it is happening
 consistently and crashing the same way - unlikely to be a hardware problem.
 
 http://mail-index.netbsd.org/port-amd64/2008/12/19/msg000684.html
 
 Looking at the data from your email, there is callback_run_roundrobin() which
 calls a function pointer, but it is very unlikely to be NULL.  I have added an
 assert, "just in case":
 
 http://mail-index.netbsd.org/source-changes/2009/11/24/msg003498.html
 
 Would you be able to try -current kernel, with that change in the link, and
 with the following debug options:
 
 options DIAGNOSTIC
 options DEBUG
 makeoptions DEBUG="-g -fno-omit-frame-pointer"
 
 Then repeat x/Lx and x/I dance again.  But no LOCKDEBUG option, as it might
 avoid some overhead and hopefully get stack dump a little bit more readable.
 Additionally, 'show uvm' output from DDB might be useful.
 
 Thanks.
 
 -- 
 Mindaugas