Re: [PATCH v2 4/4] nmi_backtrace: generate one-line reports for idle cpus

[Chris Metcalf]

On 03/21/2016 11:42 AM, Peter Zijlstra wrote:
> On Wed, Mar 16, 2016 at 01:02:13PM -0400, Chris Metcalf wrote:
> > diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
> > index 9f7c21c22477..d569ae7fde37 100644
> > --- a/arch/x86/kernel/process.c
> > +++ b/arch/x86/kernel/process.c
> > @@ -298,7 +298,7 @@ void arch_cpu_idle(void)
> >   /*
> >    * We use this if we don't have any better idle routine..
> >    */
> > -void default_idle(void)
> > +void __cpuidle default_idle(void)
> >   {
> >   	trace_cpu_idle_rcuidle(1, smp_processor_id());
> >   	safe_halt();
> > @@ -413,7 +413,7 @@ static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
> >    * with interrupts enabled and no flags, which is backwards compatible with the
> >    * original MWAIT implementation.
> >    */
> > -static void mwait_idle(void)
> > +static __cpuidle void mwait_idle(void)
> >   {
> >   	if (!current_set_polling_and_test()) {
> >   		trace_cpu_idle_rcuidle(1, smp_processor_id());
> The most common idle function for x86 is: mwait_idle_with_hints(),
> trouble is, its an inline, so I'm not sure adding __cpuidle to it does
> anything.

No, you're right, it wouldn't help.  I didn't look at the drivers/cpuidle
subsystem at all in my patch, since I'm not that familiar with it,
but it seems like tagging acpi_processor_ffh_cstate_enter(), as the
only user of mwait_idle_with_hints(), will do the job.

I do see that native_play_dead() also uses mwait/monitor, but since
that's hotplug I don't think it's relevant to this patch series.

> I've yet to find the magic objdump incantation to check. Or rather
> objdump -h doesn't appear to list .cpuidle.text at all :/
>
> I'm probably doing something silly...

The easiest way to check for a given function is just to look
at the "nm -n" output and see that all the functions you expect
to reflect idle behavior are in the cpuidle begin/end range.
Or, to look at "objdump -dr" and search for monitor/mwait.

objdump -h certainly works to show .cpuidle.text if you look at
individual objects (e.g. arch/x86/kernel/process.o) but by the time
you're looking at the linked vmlinux image they have all been linked
into the giant .text section.

--
Chris Metcalf, Mellanox Technologies
http://www.mellanox.com