cpufreq: Avoid a bad reference count on CPU node

In the Linux kernel, the following vulnerability has been resolved:

cpufreq: Avoid a bad reference count on CPU node

In the parse_perf_domain function, if the call to

of_parse_phandle_with_args returns an error, then the reference to the

CPU device node that was acquired at the start of the function would not

be properly decremented.

Address this by declaring the variable with the __free(device_node)

cleanup attribute.