[PATCH RT] time/hrtimer: disable that mode check

[PATCH RT] time/hrtimer: disable that mode check

[Sebastian Andrzej Siewior]

On -RT we push all timers by default into the soft-mode. The
"start/update" code does not use that SOFT bit so we always see that
warning.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
I am going to fold it into the original patch since only v4.14 is
affected.

 kernel/time/hrtimer.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 2fff8790bc56..f4df924d4b48 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -422,8 +422,10 @@ static inline void debug_hrtimer_activate(struct hrtimer *timer,
 	 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
 	 * match, when a timer is started via__hrtimer_start_range_ns().
 	 */
+#ifndef CONFIG_PREEMPT_RT_BASE
 	if (modecheck)
-		WARN_ON_ONCE((mode & HRTIMER_MODE_SOFT) & !timer->is_soft);
+		WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
+#endif
 
 	debug_object_activate(timer, &hrtimer_debug_descr);
 }
-- 
2.15.0

Re: [PATCH RT] time/hrtimer: disable that mode check

[Peter Zijlstra]

On Wed, Dec 13, 2017 at 02:58:35PM +0100, Sebastian Andrzej Siewior wrote:
> On -RT we push all timers by default into the soft-mode. The
> "start/update" code does not use that SOFT bit so we always see that
> warning.

Not _all_ timers I hope? There's a few the scheduler and perf use that
really should stay in hardirq context.

Re: [PATCH RT] time/hrtimer: disable that mode check

[Sebastian Andrzej Siewior]

On 2017-12-13 15:19:19 [+0100], Peter Zijlstra wrote:
> On Wed, Dec 13, 2017 at 02:58:35PM +0100, Sebastian Andrzej Siewior wrote:
> > On -RT we push all timers by default into the soft-mode. The
> > "start/update" code does not use that SOFT bit so we always see that
> > warning.
> 
> Not _all_ timers I hope? There's a few the scheduler and perf use that
> really should stay in hardirq context.
That is correct, a few ones remain in the hardirq context.

Sebastian

Re: [PATCH RT] time/hrtimer: disable that mode check

[Steven Rostedt]

On Wed, 13 Dec 2017 15:22:57 +0100
Sebastian Andrzej Siewior <bigeasy@linutronix.de> wrote:

> On 2017-12-13 15:19:19 [+0100], Peter Zijlstra wrote:
> > On Wed, Dec 13, 2017 at 02:58:35PM +0100, Sebastian Andrzej Siewior wrote:  
> > > On -RT we push all timers by default into the soft-mode. The
> > > "start/update" code does not use that SOFT bit so we always see that
> > > warning.  
> > 
> > Not _all_ timers I hope? There's a few the scheduler and perf use that
> > really should stay in hardirq context.  
> That is correct, a few ones remain in the hardirq context.
> 

As Sebastian said, by default they are in SOFT mode. I'm assuming the
ones that are not have some logic to make sure they are not in SOFT
mode.

-- Steve