Nick Piggin wrote:

> What would be wrong with an array of NR_CPUS rwsems? The only
> tiny trick you would have to do AFAIKS is have up_read remember
> what rwsem down_read took, but that could be returned from
> down_read as a token.

Here's something to start with. Add initialisers / cacheline
alignment / percpu_alloc to taste.

For bonus points I guess you could change the implementation
to just a single waitqueue mechanism with NR_CPUs counters.

In fact, it probably needs quite a bit of work before it is
mainline-worthy, but as a proof of concept - do you see
anything wrong with it?

Nick

-- 
SUSE Labs, Novell Inc.