The following patch is the ARM implementation for RWSEM_XCHGADD_ALGORITHM.
Based on David's rwsem benchmark from http://lwn.net/Articles/89191/ ,
I see a consistent 5-6% increase in the number of reads and writes taken as
compared to RWSEM_GENERIC_SPINLOCK on a dual core ARMv7. So far I have tested


numrd=1 numwr=1 numdg=1 do_sched=1

numrd=10 numwr=10 numdg=1 do_sched=1

numrd=20 numwr=20 numdg=1 do_sched=1


Can anyone else help to test this patch on other ARM systems ?

Cheers,
Ashwin