On 02/22/2013 09:39 PM, Viresh Kumar wrote:
Hi Nathan,
Sorry for pointing out this so late but i still feel we are missing something
really important.
On 22 February 2013 21:54, Nathan Zimmer <nzimmer@xxxxxxx> wrote:
- read_lock_irqsave(&cpufreq_driver_lock, flags);
+ rcu_read_lock();
+ freqs->flags = rcu_dereference(cpufreq_driver)->flags;
policy = per_cpu(cpufreq_cpu_data, freqs->cpu);
- read_unlock_irqrestore(&cpufreq_driver_lock, flags);
+ rcu_read_unlock();
- write_lock_irqsave(&cpufreq_driver_lock, flags);
+ spin_lock_irqsave(&cpufreq_driver_lock, flags);
for_each_cpu(j, policy->cpus) {
per_cpu(cpufreq_cpu_data, j) = policy;
per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
}
- write_unlock_irqrestore(&cpufreq_driver_lock, flags);
+ spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
Look at how we are protecting cpufreq_cpu_data here. rcu_read_[un]lock()
only marks the start/end of critical section. How are we sure here that
cpufreq_cpu_data is not read simultaneously when we are updating it?
rcu lock/unlock only works for cpufreq_driver pointer only and not for
this data. We still need the same locking for for cpufreq_cpu_data.
What do you say?
That would include putting the lock around the __cpufreq_cpu_get.
But I do think your right.
Perhaps a better way at this point is to have one lock for
cpufreq_cpu_data, and a second with the rcu to protect cpufreq_driver.
--
To unsubscribe from this list: send the line "unsubscribe cpufreq" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
