Re: [RFC v1 2/2] rcu/tree: Remove dynticks_nmi_nesting counter
From: Paul E. McKenney
Date: Thu Aug 29 2019 - 12:33:35 EST
On Thu, Aug 29, 2019 at 09:59:07AM -0400, Joel Fernandes wrote:
> On Wed, Aug 28, 2019 at 08:43:36PM -0700, Paul E. McKenney wrote:
> > On the tracing patch... That patch might be a good idea regardless,
> > but I bet that the reason that you felt the sudden need for it was due
> > to the loss of information in your eventual ->dynticks_nesting field.
> > After all, the value 0x1 might be an interrupt from idle, or it might
> > just as easily be a task running in the kernel at process level.
> True, however what really triggered me to do it was the existing code which
> does not distinguish between entry/exit from USER and IDLE.
> > The reason the patch might nevertheless be a good idea is that redundant
> > information can be helpful when debugging. Especially when debugging
> > new architecture-specific code, which is when RCU's dyntick-idle warnings
> > tend to find bugs.
> Sure, and also that it is more readable to ordinary human beings than "++="
> and "--=" :-D.
And those considerations did figure into my deciding that the tracing
change was likely a good thing in any case. ;-)