Re: [RFC patch 08/18] cnt32_to_63 should use smp_rmb()

[Mathieu Desnoyers]

* Steven Rostedt (rostedt@xxxxxxxxxxx) wrote:
> 
> On Fri, 7 Nov 2008, Mathieu Desnoyers wrote:
> > 
> > I want to make sure
> > 
> >   __m_cnt_hi
> >  is read before
> >   mmio cnt_lo read
> 
> Hmm, let me make sure I understand why there is no wmb.
> 
> Paul, can you verify this?
> 
> Mathieu, you do the following:
> 
>    read a
>    smp_rmb
>    reab b
>    if (test b)
>       write a
> 
> So the idea is that you must read b to test it. And since we must read a 
> before reading b we can see that we write a before either?
> 
> The question remains, can the write happen before either of the reads?
> 

write a cannot happen before read a (same variable).
write a must happen after read b because it depends on the b value. It
makes sure the the side-effect of "write a" is seen by other CPUs
*after* we have read the b value.

> But since the read b is reading the hw clock, perhaps that just implies a
> wmb on the hardware side?
> 

It makes sense. The hardware clock has no cache coherency problem.. so
it could be seen as doing wmb() after each data update.

Mathieu

> -- Steve

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/