[Crash-utility] crash problems with linux-4.5

[Dave Anderson]

----- Original Message -----
> > > Hi,
> > >
> > > Just forcing a crash dump via 'echo c > /proc/sysrq-trigger' results in a crash
> > > dump that has no tasks saved.  I see _lots_ of these errors when loading the
> > > dump with the latest crash from the git repo:
> > >
> > > crash: page excluded: kernel virtual address: ffff88103c5a9000  type: "fill_task_struct"
> > > WARNING: active task ffff881078a04340 on cpu 3 not found in PID hash
> > >
> > > Then nothing is in the dump:
> > >
> > > crash> ps
> > >    PID    PPID  CPU       TASK        ST  %MEM     VSZ    RSS  COMM
> > > >     0      0   0  ffffffff81c0f4c0  RU   0.0       0      0  [swapper/0]
> > > crash>
> > >
> > > Any ideas?
> > >
> > > I'm using top-o-tree from Dave's git repo
> > > https://github.com/crash-utility/crash.git:
> > >
> > > commit 04ab5c560a58246e782509d99214afcaf8462b4c
> > > Author: Dave Anderson <anderson at redhat.com>
> > > Date:   Tue Mar 1 16:16:48 2016 -0500
> > >
> > >     Put 2016 copyright in initial banner.
> > >
> > >
> > > Thanks,
> > >
> > > Steve.
> > 
> > If you have configured kdump to use makedumpfile -d31, it is probably this:
> > 
> >   [Crash-utility] makedumpfile: 4.5 kernel commit breaks page filtering
> >   https://www.redhat.com/archives/crash-utility/2016-February/msg00009.html
> > 
> > > PS: I'm not on the crash-utility list, so please include my email in any
> > > replies.
> > 
> > PS: if you're using crash on bleeding edge kernels, you might want to join
> > the list, at least in digest mode.
> 
> Thanks Dave. I've joined. :)
> 
> I changed the makedumpfile line in /etc/init.d/kdump from -d 31 to -d 17,
> then removed the kdump initrd from /boot, and restarted kdump to generate
> the new initrd file.  I then rebooted, and produced a new crash dump.  But
> I'm still seeing the same issue.
> 
> Am I doing the correct procedure for forcing makedumpfile to use -d17?

I think so, yes.  And downgrading to -d17 seems to have worked, at least for 
the two reporter on this list.  You can verify that the -d configuration is as
expected by entering:

 crash> help -D | grep dump_level

If you're still seeing "page excluded" errors, you could try:

 (1) setting -d1 (zero-pages only),
 (2) removing the -d option completely, and just use "makedumpfile -c".  
 (3) take the makedumpfile "core_collector" configuration out of the picture entirely,
     and see if just an ELF vmcore works OK.

On the other hand, it's entirely possible that there's something new that's happened,
which is the risk you run when working with upstream kernels. 

Dave


> 
> Thanks,
> 
> Steve.
> 
> 
>