On Fri, Dec 13 2024 at 12:20, Peter Zijlstra wrote: > On Fri, Dec 13, 2024 at 01:14:54AM +0100, Thomas Gleixner wrote: > >> --- a/kernel/sched/core.c >> +++ b/kernel/sched/core.c >> @@ -7276,7 +7276,7 @@ void rt_mutex_setprio(struct task_struct >> #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) >> int __sched __cond_resched(void) >> { >> - if (should_resched(0)) { >> + if (should_resched(0) && system_state != SYSTEM_SUSPEND) { >> preempt_schedule_common(); >> return 1; >> } > > Perhaps we should just do: > > if (should_resched(0) && !irqs_disabled()) > > That's more or less what preemptible() does too. Yes, mucking about with > IF is expensive, but it's only done if preempt_count is agreeing we > should schedule, which it shouldn't be most of the time. Makes sense. Let me write a real patch. Thanks, tglx