On 26 February 2014 07:18, Saravana Kannan [off-list ref] wrote:

On 02/25/2014 02:41 PM, Rafael J. Wysocki wrote:

quoted

And is "fully initialized" actually well defined?

The point in add dev/hot plug path after which we will no longer change
policy fields without sending further CPUFREQ_UPDATE_POLICY_CPU /
CPUFRE_NOTIFY notifiers.

Okay..

Pretty much the end of __cpufreq_add_dev() so that it's after:
- cpufreq_init_policy()
- And the update of userpolicy fields that after thie init call

No. In that case it can be considered initialized before cpufreq_init_policy().
As we do send CPUFREQ_NOTIFY after that from cpufreq_init_policy()->
cpufreq_set_policy().

There are two types of fields within policy, some are very basic: cpu/min/max/
affected_cpus/related_cpus

some are advanced: sysfs/governors/..

And as a rule you have to get policy->rwsem lock before accessing policy
members. We might not have followed it very well for small things like cpu.

And so if you are doing anything over that, please use a lock and that is
already present in cpufreq_update_policy().

With my latest patchset that I sent yesterday, locking is improved and now
a policy will be usable only after the rwsem is released. And that should be
fine. And so making it available in the per-cpu variable after all the necessary
fields are filled looks fine to me. And so I don't think we need to move it
after call to cpufreq_init_policy(maybe a better name to this function is
required)..

Ok, here's some pseudo code to explain it better:

Something like, replace all calls to cpufreq_driver->get with
__cpufreq_driver_get() with the fn being something like:

unsigned int __cpufreq_driver_get(cpufreq_policy *policy)
{
  if (policy->clk)
     return clk_get_rate(policy->clk) / 1000;
  else
     return cpufreq_driver->get(policy->cpu);

This part may still use cpufreq_cpu_get().

}

Drivers are free to have their implementation of ->get() even
if they have a valid policy->clk field..