[v3,4/4] nmi_backtrace: generate one-line reports for idle cpus

Commit Message

Peter Zijlstra March 22, 2016, 5:30 p.m. UTC

On Tue, Mar 22, 2016 at 01:19:39PM -0400, Chris Metcalf wrote:
> When doing an nmi backtrace of many cores, most of which are idle,
> the output is a little overwhelming and very uninformative.  Suppress
> messages for cpus that are idling when they are interrupted and just
> emit one line, "NMI backtrace for N skipped: idling at pc 0xNNN".
> 
> We do this by grouping all the cpuidle code together into a new
> .cpuidle.text section, and then checking the address of the
> interrupted PC to see if it lies within that section.
> 
> This commit suitably tags x86, arm64, and tile idle routines,
> and only adds in the minimal framework for other architectures.
> 
> Acked-by: Peter Zijlstra (Intel) <[e-mail]>
> Tested-by: Peter Zijlstra (Intel) <[e-mail]>
> Signed-off-by: Chris Metcalf <[e-mail]>

For some reason I found a few CPUs using poll_idle().

Rafael, when and why would that ever get selected as a useful idle
state? When the predicted idle time is so short even C1 isn't worth it?

Comments

Rafael J. Wysocki March 22, 2016, 10:28 p.m. UTC | #1

On Tuesday, March 22, 2016 06:30:05 PM Peter Zijlstra wrote:
> On Tue, Mar 22, 2016 at 01:19:39PM -0400, Chris Metcalf wrote:
> > When doing an nmi backtrace of many cores, most of which are idle,
> > the output is a little overwhelming and very uninformative.  Suppress
> > messages for cpus that are idling when they are interrupted and just
> > emit one line, "NMI backtrace for N skipped: idling at pc 0xNNN".
> > 
> > We do this by grouping all the cpuidle code together into a new
> > .cpuidle.text section, and then checking the address of the
> > interrupted PC to see if it lies within that section.
> > 
> > This commit suitably tags x86, arm64, and tile idle routines,
> > and only adds in the minimal framework for other architectures.
> > 
> > Acked-by: Peter Zijlstra (Intel) <[e-mail]>
> > Tested-by: Peter Zijlstra (Intel) <[e-mail]>
> > Signed-off-by: Chris Metcalf <[e-mail]>
> 
> For some reason I found a few CPUs using poll_idle().
> 
> Rafael, when and why would that ever get selected as a useful idle
> state? When the predicted idle time is so short even C1 isn't worth it?

Yes, that's the case.

Patch

--- a/drivers/cpuidle/driver.c
+++ b/drivers/cpuidle/driver.c
@@ -14,6 +14,7 @@ 
 #include <linux/cpuidle.h>
 #include <linux/cpumask.h>
 #include <linux/tick.h>
+#include <linux/cpu.h>
 
 #include "cpuidle.h"
 
@@ -178,7 +179,7 @@  static void __cpuidle_driver_init(struct
 }
 
 #ifdef CONFIG_ARCH_HAS_CPU_RELAX
-static int poll_idle(struct cpuidle_device *dev,
+__cpuidle static int poll_idle(struct cpuidle_device *dev,
 		struct cpuidle_driver *drv, int index)
 {
 	local_irq_enable();