Re: Questions for CONFIG_WEAK_ORDERING and CONFIG_WEAK_REORDERING_BEYOND_LLSC

[Ralf Baechle <ralf@linux-mips.org>]

On Mon, Oct 13, 2008 at 09:01:44PM +0200, Johannes Dickgreber wrote:

> If a cpu is WEAK_ORDERING schouldn't it do a sync independent of CONFIG_SMP ?
> 
> And if it is a SMP system schouldn't it do a sync independent of 
> CONFIG_WEAK_ORDERING ?
> 
> And if a cpu has no sync with LLSC schouldn't it do a sync independent of 
> CONFIG_SMP ?
> 
> All together, is the following the right thing to do ?

A processor is always consistently ordered wrt. to itself, so uniprocessor
cores never need SYNCs even if that processor was weakly ordered in a
multiprocessor systems.

A while ago I walked through all mb(), rmb() and wmb() uses in the generic
code.  None of the ones I verified is actually needed on uniprocessor
kernels.  Ocasionally one of these functions is used to maintain I/O
ordering but again other mechanisms are prefered for that purpose.

  Ralf