Re: Fw: Re: 2.6.18-rc2-mm1

[Andi Kleen <ak@muc.de>]

On Fri, Jul 28, 2006 at 10:56:16AM +0100, Jan Beulich wrote:
> >>> Andrew Morton <akpm@osdl.org> 28.07.06 10:19 >>>
> >
> >More unwinder problems.  Seems to be specific to -mm.
> >(Followups on-list are preferred, thanks).
> 
> That's another of the lose ends - while Andi and I had decided to put
> a proper stack frame in place for the switch to the irq stack in the
> interrupt macro, we had forgotten about doing something similar for
> call_softirq() - as the code is written currently, there simply is now
> way of annotating the code properly. If Andi is agreeable to this,
> then I'll just change the code so that it'll have a proper stack frame.

Fine by me. I'll just fix it up.

> 
> The recursive traces appear to result from the fact that when we
> did the change to the interrupt macro, we neglected the fact that
> show_trace() expects to find the old stack pointer as the very first
> item on the interrupt stack, which isn't the case anymore. The value
> of where the subsequent stack access fails is a little strange, though.
> Making this assumption be true again would, however, require adding
> a push to both the interrupt macro and call_softirq. I'm afraid Andi's
> not going to be too happy about that, but on the other hand I can't
> see any other possible way that would get away without adding
> some code to these paths.

I can just add the push.

-Andi