Re: scheduling while atomic & hang.

[Ingo Molnar <mingo@kernel.org>]

* Frederic Weisbecker <fweisbec@gmail.com> wrote:

> On Fri, Jul 05, 2013 at 12:27:51PM -0700, Linus Torvalds wrote:
> > On Thu, Jul 4, 2013 at 11:51 PM, Ingo Molnar <mingo@kernel.org> wrote:
> > >
> > > It would be nice to have a full-kernel-stack backtrace printout as an
> > > option, which prints entries 'above' the current RSP.
> > 
> > One problem with that is that it is likely to be mostly overwritten by 
> > the debug code that is about to print this all out - and any non-debug 
> > code may well be really old, just with a sufficiently deep stack trace 
> > to show up. Not to mention missing any irq-stack changes etc anyway.
> > 
> > So it can easily be *very* misleading - I'm not sure how useful it 
> > would be as an idea.
> > 
> >                 Linus
> 
> So if the unwider doesn't run with the same stak than the target, it 
> migh work. This can work with BUG for example But I believe that WARN 
> doesn't do a trap in x86.

Instead of switching stacks it might be better to create an early snapshot 
of the stack into a temporary area before calling anything, and to print 
that. This is debug code, so overhead is not a big issue.

But yeah, Linus is right, it gets hairy, can be misleading and has limited 
utility :-/

Thanks,

	Ingo