Re: [PATCH] x86: push old stack address on irqstack for unwinder

[Martin Hicks <mort@sgi.com>]

On Sat, Jan 31, 2009 at 01:39:21AM +0100, Ingo Molnar wrote:
> 
> * H. Peter Anvin <hpa@zytor.com> wrote:
> 
> > H. Peter Anvin wrote:
> >> Martin Hicks wrote:
> >>> Hi,
> >>>
> >>> KDB was using this information.  Could this be pushed towards 2.6.29  
> >>> please?
> >>>
> >>> This re-adds the old stack pointer to the top of the irqstack to help
> >>> with unwinding.  It was removed in commit  
> >>> d99015b1abbad743aa049b439c1e1dede6d0fa49
> >>> as part of the save_args out-of-line work.
> >>>
> >>
> >> This bothers me... why should we add even a single instruction to what  
> >> is arguably the single hottest path in the kernel to support an  
> >> out-of-tree debugger, especially if kgdb (which is in-tree) doesn't 
> >> need it?
> >>
> >> What does kgdb do differently (or is kgdb broken too)?
> >>

I was searching around, trying to find out if there was another way for
kdb to do this, and I think removing the backlink is breaking other
stuff also.  dump_trace() in dumpstack_64.S is using the same trick as
KDB to trace out of the interrupt stack:

            /*
             * We link to the next stack (which would be
             * the process stack normally) the last
             * pointer (index -1 to end) in the IRQ stack:
             */
            stack = (unsigned long *) (irqstack_end[-1]);
            irqstack_end = NULL;
            ops->stack(data, "EOI");
            continue;


mh