perf events ring buffer memory barrier on powerpc

[Peter Zijlstra]

On Sun, Oct 27, 2013 at 11:00:33AM +0200, Victor Kaplansky wrote:
> Peter Zijlstra <peterz at infradead.org> wrote on 10/25/2013 07:37:49 PM:
> 
> > I would argue for:
> >
> >   READ ->data_tail         READ ->data_head
> >     smp_rmb()   (A)          smp_rmb()   (C)
> >   WRITE $data              READ $data
> >     smp_wmb()   (B)          smp_mb()   (D)
> >   STORE ->data_head        WRITE ->data_tail
> >
> > Where A pairs with D, and B pairs with C.
> 
> 1. I agree. My only concern is that architectures which do use atomic
> operations
> with memory barriers, will issue two consecutive barriers now, which is
> sub-optimal.

Yeah, although that would be fairly easy to optimize by the CPUs itself;
not sure they actually do this though.

But we don't really have much choice aside of introducing things like:

smp_wmb__after_local_$op; and I'm fairly sure people won't like adding a
ton of conditional barriers like that either.


> 2. I think the comment in "include/linux/perf_event.h" describing
> "data_head" and
> "data_tail" for user space need an update as well. Current version -

Oh, indeed. Thanks; I'll update that too!