Re: [BUG] Fusion MPT Base Driver initialization failure with kdum p

[Vivek Goyal <vgoyal@in.ibm.com>]

On Wed, Jul 13, 2005 at 04:05:03PM +0530, Bharata B Rao wrote:
> On Tue, 2005-07-12 at 12:15 -0600, Moore, Eric Dean wrote:
> > I've seen the report. I need more info from Bharata on how
> > to reproduce. Perhaps you can send me email offline which
> > provides specific instructions to how to configure kdump,
> > how to capture the dump, and what you did to crash your system.
> 
> This is how I could reproduce this bug:
> (http://bugzilla.kernel.org/show_bug.cgi?id=4870)
> 
> - Configure the 1st kernel to take a kdump and run 4 instances of LTP on
> it (I used `runltp -p -l logfile -t 12h -x 4)
> - After a few hours, the 1st kernel crashes which results in the booting
> of the 2nd kernel (the kernel which captures kdump). While this is
> booting, mptbase driver fails during initialization leading to the panic
> of the 2nd kernel.

I had also faced this problem and I had simply used Alt-Sysrq-c to force
kexec on panic. As mentioned in the report below the problem goes away
if mpt driver is compiled with MPT_DEBUG_IRQ support. 

In kdump environment, device might not be in a reset state while driver is
initializing in second kernel. Hence we probably need two things from the 
driver to be able to successfully initialize over kdump.

- Reset the underlying device before enabling the interrupts.
- In interrupt handler, make sure the interrupt belongs to the driver 
  before processing it further.  (Shared interrupt lines) 

Thanks
Vivek