Re: kernel BUG at kernel/smpboot.c:134!

[Thomas Gleixner <tglx@linutronix.de>]

On Sun, 7 Apr 2013, Borislav Petkov wrote:

> On Sun, Apr 07, 2013 at 11:20:10AM +0200, Thomas Gleixner wrote:
> > And it's even more bogus because the cpu to which we would bind in
> > kthread_create_on_cpu() is not yet online.
> 
> In case you guys are wondering about reproducibility, I saw the same
> BUG_ON yesterday on a 32-bit atom smp qemu guest too (don't ask why?
> :-))

I don't have to ask. It's caused by massive brain failure on my side.

Does the patch below fix the issue ?

Thanks,

	tglx
---
Index: linux-2.6/kernel/kthread.c
===================================================================
--- linux-2.6.orig/kernel/kthread.c
+++ linux-2.6/kernel/kthread.c
@@ -131,8 +131,14 @@ static void __kthread_parkme(struct kthr
 		schedule();
 		__set_current_state(TASK_INTERRUPTIBLE);
 	}
-	clear_bit(KTHREAD_IS_PARKED, &self->flags);
 	__set_current_state(TASK_RUNNING);
+	clear_bit(KTHREAD_IS_PARKED, &self->flags);
+
+	/* Rebind ourself to the target cpu */
+	if (test_bit(KTHREAD_IS_PER_CPU, &self->flags)) {
+		set_cpus_allowed_ptr(currrent, cpumask_of(self->cpu));
+		current->flags |= PF_THREAD_BOUND;
+	}
 }
 
 void kthread_parkme(void)
@@ -344,11 +350,8 @@ void kthread_unpark(struct task_struct *
 		 * park before that happens we'd see the IS_PARKED bit
 		 * which might be about to be cleared.
 		 */
-		if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
-			if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
-				__kthread_bind(k, kthread->cpu);
+		if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags))
 			wake_up_process(k);
-		}
 	}
 	put_task_struct(k);
 }