Re: [PATCH v2] sched: Enabled schedstat when schedstat tracepoints are enabled

[Peter Zijlstra]

On Thu, Apr 13, 2017 at 11:00:05AM +0200, Peter Zijlstra wrote:
> On Wed, Apr 12, 2017 at 10:56:07PM -0400, Steven Rostedt wrote:
> > From: Steven Rostedt (VMware) <rostedt@xxxxxxxxxxx>
> > 
> > During my tests, I see this in my dmesg:
> > 
> > "Scheduler tracepoints stat_sleep, stat_iowait, stat_blocked and
> > stat_runtime require the kernel parameter schedstats=enabled or
> > kernel.sched_schedstats=1"
> > 
> > And found the commit:
> > 
> > cb2517653fc ("sched/debug: Make schedstats a runtime tunable that is
> > disabled by default")
> > 
> > Which states:
> > 
> > "For tracepoints, there is a simple warning as it's not safe to activate
> >  schedstats in the context when it's known the tracepoint may be wanted
> >  but is unavailable."
> > 
> > I'm assuming that Mel did not know about the TRACE_EVENT_FN() and
> > DEFINE_EVENT_FN() that allow for callbacks for tracepoints as they are
> > enabled and disabled. I do not see any reason for not enabling
> > schedstat when one of its tracepoints are enabled.
> > 
> > The state of schedstat is saved when the first tracepoint is enabled,
> > and that state is put back when the tracepoints are disabled.
> 
> There is one additional complication with all this.
> 
> Dynamically enabling the sched_stats like this doesn't guarantee correct
> information. So you've now taken away the informational print and
> silently generate bollocks numbers.

Josh has a patch like:

  http://lkml.kernel.org/r/00e8805cc094657d5ccb20bb88e0d2ce1cbb92e1.1466184592.git.jpoimboe@xxxxxxxxxx

That tries to keep the few stats required for the tracepoint active at
all times. But I never managed to get performance overhead to 0.