of_perf_domain_get_sharing_cpumask currently assumes a 1-argument
phandle format, and directly returns the argument. Generalize this to
return the full of_phandle_args, so it can be used by drivers which use
other phandle styles (e.g. separate nodes). This also requires changing
the CPU sharing match to compare the full args structure.

Also, make sure to of_node_put(args.np) (the original code was leaking a
reference).
Signed-off-by: Hector Martin <marcan@marcan.st>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>