Re: 2.6.8-rc4-mm1

[William Lee Irwin III <wli@holomorphy.com>]

On Tue, Aug 10, 2004 at 06:39:33PM -0400, Robert Picco wrote:
> copy_thread expects a switch_stack below pt_regs  on the stack.  the 
> switch_stack would have the parent's bspstore value for computing how 
> much register backing store to copy into child.  there isn't a 
> switch_stack and the resultant bspstore size computed is enormous 
> (depends on what there is on stack).  i suspect printk has changed stack 
> and the code layout changed too.
[...]
> 	 * For SMP idle threads, fork_by_hand() calls do_fork with
> 	 * NULL regs.
> 	 */
> -	if (!regs)
> +	if (clone_flags & CLONE_IDLETASK)
> 		return 0;
> #endif

It may not be immediately obvious, but since all the checks for
CLONE_IDLETASK were ripped out in patches prior to this, respecting
CLONE_IDLETASK anywhere means that userspace can create idle threads(!).
So the alternative fixes I've been posting are very necessary.


-- wli