Re: A patch for the file kernel/fork.c

[Steven Rostedt <rostedt@goodmis.org>]

On Wed, 2005-05-04 at 23:21 +0200, Alexander Nyberg wrote:

> This patch is very crude but it is quite resistant to recursive faults
> in do_exit(), survives the LTP hammering I've given it. The problem is
> not knowing where in the previous path it broke down so I'd rather just
> leave it lying around and try a graceful reset/power off. But if anyone
> has a better suggestion than the msleep() I'm all ears but this area is
> sensitive.
> 
> Where is that anonymous patch hot-line...
> 

> +	/* We're taking recursive faults originating here in do_exit. Safest 
> +	 * is to just leave this task alone and wait for reboot. */
> +	if (tsk->flags & PF_EXITING) {
> +		printk(KERN_ALERT "\nFixing recursive fault but reboot is needed!\n");
> +		for (;;)
> +			msleep(1000 * 10);
> +	}
> +
>  	tsk->flags |= PF_EXITING;
>  

Instead of the for(;;) msleep, why not just take it permanently off the
run queue? With the following:

   set_current_state(TASK_UNINTERRUPTIBLE);
   schedule();

It basically gives the same effect, but is cleaner.

-- Steve