cpufreq: Fix kobject memleak

Currently the error return path from kobject_init_and_add() is not
followed by a call to kobject_put() - which means we are leaking the
kobject.

Fix it by adding a call to kobject_put() in the error path of
kobject_init_and_add().

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Tobin C. Harding <tobin@kernel.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

drivers/cpufreq/cpufreq.c

@@ -1133,6 +1133,7 @@ static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
 				   cpufreq_global_kobject, "policy%u", cpu);
 	if (ret) {
 		pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
+		kobject_put(&policy->kobj);
 		goto err_free_real_cpus;
 	}
 

drivers/cpufreq/cpufreq_governor.c

@@ -459,6 +459,8 @@ int cpufreq_dbs_governor_init(struct cpufreq_policy *policy)
 	/* Failure, so roll back. */
 	pr_err("initialization failed (dbs_data kobject init error %d)\n", ret);
 
+	kobject_put(&dbs_data->attr_set.kobj);
+
 	policy->governor_data = NULL;
 
 	if (!have_governor_per_policy())