Re: [PATCH 0/8] git pull request for tip/tracing/core

[Ingo Molnar <mingo@elte.hu>]

* Steven Rostedt <rostedt@goodmis.org> wrote:

> 
> On Wed, 11 Feb 2009, Luck, Tony wrote:
> 
> > > The bits in question is really the number of possible nested interrupts
> > > that can happen. We take the paranoid approach that we can have a max
> > > nesting of NR_IRQS. Perhaps this can be changed, and just do a max of
> > > 1<<10 nesting? And have a big warn on if it happens to be bigger, or fall
> > > to another counter if it is bigger.
> > >
> > > 1000 nested IRQs seems a bit extreme :-/
> > 
> > Ah, I see.  Then the answer is very different.  The number of nested
> > interrupts possible on a cpu is limited by the number of priority
> > classes for interrupts (See Table 5-8 on page 2:112 of the Itanium
> > software developers manual).  Effectively the max nesting depth is
> > 16.
> > 
> > 1000 nested interrupts would be certain to run us out of stack.
> 
> Ingo,
> 
> Do you think this is a good assumption to make, that no arch will have 
> over 1<<10 nested interrupts. We can add a WARN_ON if it happens. Then we 
> can make all archs have a 10 bit offset. Smaller may also be sufficient.

That's a fair assumption, yes. No need for a WARN_ON() (it slows down a critical 
path) - things will get very colorful much sooner than that, due to kernel stack 
overflow.

	Ingo