Re: [PATCH 3/4] kthreads: rework kthread_stop()

[ebiederm@xmission.com (Eric W. Biederman)]

Rusty Russell <rusty@rustcorp.com.au> writes:

> On Wednesday 04 February 2009 15:40:06 Eric W. Biederman wrote:
>> static struct kthread *to_kthread(struct task_struct *tsk)
>> {
>> 	void *stack = task_stack_page(tsk);
>> 	return (struct kthread *)(stack + kthread_offset);
>>         
>> }
> ...
>> It would remove the test and be simple and obviously correct.
>
> Clever? Sure.  Neat? Yes.
>
> But you are using a definition of obvious with which I was not previously
> familiar :)

Well the way you compute kthread_offset is:

	struct kthread kthread;
	void *stack = task_stack_page(current);
        kthread_offset = (void *)&kthread - stack;

Now Rusty I don't know about you but after I learned to do
addition and subtraction it has always been obvious to me that
one is the opposite of the other.

Further I think the rest of that code becomes a lot clearer if
we can remove that stupid, unnecessary conditional.  As worrying
if the process has exited implies we care about a lot of things
that we really don't and seem to make the code generally less
comprehensible.

I am slightly concerned that using task_stack_page(tsk) may be
overly clever, but compared to ACCESS_ONCE(), memory barriers,
or not letting kthread_stop be called on a thread that may exit
I think I am ahead of the game.

Eric