Re: [PATCH] kthread: Prevent unpark race which puts threads on the wrong cpu

[Thomas Gleixner <tglx@linutronix.de>]

On Wed, 10 Apr 2013, Thomas Gleixner wrote:
> On Tue, 9 Apr 2013, Dave Hansen wrote:
> 
> > On 04/09/2013 12:30 PM, Thomas Gleixner wrote:
> > > On Tue, 9 Apr 2013, Thomas Gleixner wrote:
> > > Thought more about it and found, that the stupid binding only works
> > > when the task is really descheduled. So there is a small window left,
> > > which could lead to this. Revised patch below.
> > > 
> > > Anyway a trace for that issue would be appreciated nevertheless.
> > 
> > Here you go:
> > 
> > 	http://sr71.net/~dave/linux/bigbox.1365539189.txt.gz
> 
> Hmm. Unfortunately migration/146 is not in the trace.
>  
> Can you please apply the patch below? That avoids the oops, but might
> hang an online operation. Though the machine should stay up and you
> should be able to retrieve the trace.
> 
> Thanks,
> 
> 	tglx
> ---
> Index: linux-2.6/kernel/smpboot.c
> ===================================================================
> --- linux-2.6.orig/kernel/smpboot.c
> +++ linux-2.6/kernel/smpboot.c
> @@ -131,7 +131,10 @@ static int smpboot_thread_fn(void *data)
>  			continue;
>  		}
>  
> -		BUG_ON(td->cpu != smp_processor_id());
> +		if (td->cpu != smp_processor_id()) {
> +			tracing_off();
> +			schedule();

Bah, that wants a continue. Revised patch below.

> +		}
>  
>  		/* Check for state change setup */
>  		switch (td->status) {
 
Index: linux-2.6/kernel/smpboot.c
===================================================================
--- linux-2.6.orig/kernel/smpboot.c
+++ linux-2.6/kernel/smpboot.c
@@ -131,7 +131,11 @@ static int smpboot_thread_fn(void *data)
 			continue;
 		}
 
-		BUG_ON(td->cpu != smp_processor_id());
+		if (td->cpu != smp_processor_id()) {
+			tracing_off();
+			schedule();
+			continue;
+		}
 
 		/* Check for state change setup */
 		switch (td->status) {