On Tue, Jul 07, 2015 at 06:24:22PM +0100, Will Deacon wrote:

When contended, architectures may be able to reduce the polling overhead
in ways which aren't expressible using a simple relax() primitive.

This patch allows architectures to override the use of
cpu_relax_lowlatency() in the qrwlock code and also implement their own
unlock macros in case explicit signalling is required to wake up a
`relaxed' CPU spinning on an unlock event.

No real objection, but could you do this _after_ you've converted
AARGH64 to use the normal qrwlock, such that you can show the benefit
with numbers?