Re: [PATCH 2/5 v2] tracing: Replace the per_cpu() with this_cpu() in trace_stack.c

[Steven Rostedt]

On Fri, 7 Apr 2017 07:36:19 -0700
"Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx> wrote:

> On Fri, Apr 07, 2017 at 10:01:08AM -0400, Steven Rostedt wrote:
> > From: "Steven Rostedt (VMware)" <rostedt@xxxxxxxxxxx>
> > 
> > The updates to the trace_active per cpu variable can be updated with the
> > this_cpu_*() functions as it only gets updated on the CPU that the variable
> > is on.
> > 
> > Signed-off-by: Steven Rostedt (VMware) <rostedt@xxxxxxxxxxx>
> > ---
> >  kernel/trace/trace_stack.c | 23 +++++++----------------
> >  1 file changed, 7 insertions(+), 16 deletions(-)
> > 
> > diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c
> > index 5fb1f2c87e6b..05ad2b86461e 100644
> > --- a/kernel/trace/trace_stack.c
> > +++ b/kernel/trace/trace_stack.c
> > @@ -207,13 +207,12 @@ stack_trace_call(unsigned long ip, unsigned long parent_ip,
> >  		 struct ftrace_ops *op, struct pt_regs *pt_regs)
> >  {
> >  	unsigned long stack;
> > -	int cpu;
> > 
> >  	preempt_disable_notrace();
> > 
> > -	cpu = raw_smp_processor_id();
> >  	/* no atomic needed, we only modify this variable by this cpu */
> > -	if (per_cpu(trace_active, cpu)++ != 0)
> > +	this_cpu_inc(trace_active);  
> 
> For whatever it is worth...
> 
> I was about to complain that this_cpu_inc() only disables preemption,
> not interrupts, but then I realized that any correct interrupt handler
> would have to restore the per-CPU variable to its original value.

Yep, that's the reason for the comment about "no atomic needed". This
is a "stack modification". Any interruption in the flow will reset the
changes back to the way it was before going back to what it interrupted.

> 
> Presumably you have to sum up all the per-CPU trace_active counts,
> given that there is no guarantee that a process-level dec will happen
> on the same CPU that did the inc.

That's why we disable preemption. We guarantee that a process-level dec
*will* happen on the same CPU that did the inc.

It's also the reason for the preemption disabled check in the
stack_tracer_disable() code.

-- Steve