Re: [PATCH 7/7] sparc64: Add function graph tracer support.

[Frederic Weisbecker <fweisbec@gmail.com>]

On Wed, Apr 14, 2010 at 02:04:49AM -0700, David Miller wrote:
> From: David Miller <davem@davemloft.net>
> Date: Tue, 13 Apr 2010 18:59:31 -0700 (PDT)
> 
> > From: David Miller <davem@davemloft.net>
> > Date: Tue, 13 Apr 2010 16:56:39 -0700 (PDT)
> > 
> >> I'll dig further.
> > 
> > Ok Frederic, I think I figured it out.  We only save 128 bytes
> > of stack space in the mcount stubs, but that's not enough.
> 
> Ok, one more fix after this one.
> 
> The other problem I seem to be hitting is that we overrun the main
> kernel stack with tracing enabled.
> 
> Let's run NMIs on the hardirq stack to abate this problem for now,
> patch below which I'll also push to sparc-2.6
> 
> The only remaining issue I have is that one of the cpus can get stuck
> in the scheduler load balancer.  Whatever NOHZ cpu gets selected to
> balance idle domains can get stuck because the time it takes to
> rebalance domains with tracing enabled exceeds the interval of the
> scheduler tick, so it just balances endlessly and never makes any
> forward progress from the initial timer interrupt.


I observed the same thing yesterday, cpu 0 was stuck in the balancing
path.

Anyway, I'm testing your patches.