Re: [PATCH 3/3] cpufreq: Don't use cpu removed during cpufreq_driver_unregister

On 3 January 2013 19:55, Srivatsa S. Bhat srivatsa.bhat@linux.vnet.ibm.com wrote:

I took a quick look at the problem you described above, and the cpufreq code.. If we cannot avoid calling cpufreq_add_dev() from cpufreq_remove_dev(), then I can't think of anything better than what your patch does.

Good :)

BTW, off-topic, while going through that path, I think I found a memory leak in __cpufreq_remove_dev():

    if (unlikely(cpumask_weight(data->cpus) > 1)) {
            for_each_cpu(j, data->cpus) {
                    if (j == cpu)
                            continue;
                    per_cpu(cpufreq_cpu_data, j) = NULL;
            }
    }

We are assigning NULL without freeing that memory.

Not really. All cpus in affected_cpus (data->cpus), share the same policy structure. We have already taken backup of cpufreq_cpu_data for the first cpu in "data" and are freeing it here:

kfree(data);