hi, Thomas Gleixner,

On Wed, Aug 06, 2025 at 12:39:09PM +0200, Thomas Gleixner wrote:
> On Wed, Aug 06 2025 at 16:35, Oliver Sang wrote:
> 
> > On Wed, Aug 06, 2025 at 10:19:36AM +0200, Thomas Gleixner wrote:
> >> Oliver!
> >> 
> >> On Wed, Aug 06 2025 at 16:01, Oliver Sang wrote:
> >> > On Thu, Jul 31, 2025 at 04:03:01PM +0200, Thomas Gleixner wrote:
> >> >> On Thu, Jul 31 2025 at 14:34, Thomas Gleixner wrote:
> >> >> > Either you make the timeout longer or try the following on the kernel
> >> >> > command line instead of 'ftrace_dump_on_cpu':
> >> >> 
> >> >>   instead of 'ftrace_dump_ooops'
> >> >
> >> > sorry for late, I just took several-day off.
> >> 
> >> I hope you had a good time!
> >> 
> >> > attached dmesg FYI.
> >> 
> >> Thanks for doing this. Now the buffer is too short and as nothing stops
> >> the tracer the interesting stuff goes out of sight.
> >> 
> >> Can you please apply the patch below and try again?
> >
> > got it!
> >
> > just want to confirm, still use below params or need some modification?
> >
> > trace_event=preemptirq:preempt_disable,preemptirq:preempt_enable ftrace_dump_on_oops=orig_cpu trace_buf_size=100K panic_on_warn=1
> 
> Those should be good. Let's see.
> 

I applied your patch upon 56180dd20c:

* cdf0aafee2bddf test patch for 56180dd20c from Thomas Gleixner
* 56180dd20c19e5 futex: Use RCU-based per-CPU reference counting instead of rcuref_t


commit cdf0aafee2bddf948281bed5eb0574bf33a09802
Author: 0day robot <lkp@intel.com>
Date:   Wed Aug 6 21:38:42 2025 +0800

    test patch for 56180dd20c from Thomas Gleixner

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index ec68fc686bd742..2e7d7b5b8e7589 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -5900,6 +5900,8 @@ void preempt_count_sub(int val)
        /*
         * Underflow?
         */
+       if (val > preempt_count())
+               tracing_off();
        if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
                return;
        /*

attached one dmesg FYI.