Re: frlock and barrier discussion

[Andrea Arcangeli]

On Wed, Jan 29, 2003 at 05:41:33PM -0800, Richard Henderson wrote:
> 	//begin
> 	t1 = rw->pre_sequence
> 	t1 += 1
> 	rw->pre_sequence = t1
> 	wmb()
> 
> 	//stuff
> 	xtimensec = xtime.tv_nsec
> 
> 	//end
> 	wmb()
> 	t2 = rw->post_sequence
> 	t2 += 1
> 	rw->post_sequence = t2
> 
> is
> 
> 	t1 = rw->pre_sequence
> 	t2 = rw->post_sequence
> 	xtimensec = xtime.tv_nsec
> 	t1 += 1;
> 	t2 += 2;
> 	rw->pre_sequence = t1
> 	wmb()
> 	wmb()
> 	rw->post_sequence = t2

No it's:

 	t1 = rw->pre_sequence
 	t2 = rw->post_sequence
 	t1 += 1;
 	t2 += 2;
 	rw->pre_sequence = t1
 	wmb()
 	xtimensec = xtime.tv_nsec
 	wmb()
 	rw->post_sequence = t2

you're missing xtimensec is a write.

or this if you prefer:

	spin_lock() / now xtime can't change under us

 	t1 = rw->pre_sequence
 	t2 = rw->post_sequence
	t3 = xtime.tv_nsec
 	t1 += 1;
 	t2 += 2;
 	rw->pre_sequence = t1
 	wmb()
 	xtimensec = t3
 	wmb()
 	rw->post_sequence = t2

	spin_unlock() / now xtime can change again

and the above is the optimal implementation of the write-side. We
definitely don't want to forbid those reoderings. if gcc or cpu thinks
it's worthwhile they must be allowed to optimize it since it's legal.

I believe wmb() is correct, and mb() is overkill.

Andrea