[PATCH 45/74] SPEAr : SEV Send event to secondary CPUs

[Russell King - ARM Linux]

On Thu, Sep 02, 2010 at 04:37:37PM +0530, Shilimkar, Santosh wrote:
> May be we should have barrier version of these
> There is one already exist in mainline already in spin lock code
> "dsb_sev(void)"

I don't think so - they're not a general purpose event passing mechanism.
Their use is fairly well limited to two situations:

1. spin locks, where we already have dsb_sev as we care about ordering.
2. smp secondary cpu bring-up code, where we care a little less about the
   ordering.

I think coupling a barrier with these will just make people want to have
a version without the barrier.  At least if we don't provide the barrier,
people have the option of using mb() before sev() if they desire ordering.