Re: [PATCH 2/2] cpufreq: don't call cpufreq_update_policy() on CPU addition

["Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>]

On 02/17/2014 02:24 PM, Viresh Kumar wrote:
> On 17 February 2014 14:13, Srivatsa S. Bhat
> <srivatsa.bhat@linux.vnet.ibm.com> wrote:
>> On 02/14/2014 04:30 PM, Viresh Kumar wrote:
>>> cpufreq_update_policy() is called from two places currently. From a workqueue
>>> handled queued from cpufreq_bp_resume() for boot CPU and from
>>> cpufreq_cpu_callback() whenever a CPU is added.
>>>
>>> The first one makes sure that boot CPU is running on the frequency present in
>>> policy->cpu. But we don't really need a call from cpufreq_cpu_callback(),
>>> because we always call cpufreq_driver->init() (which will set policy->cur
>>> correctly) whenever first CPU of any policy is added back. And so every policy
>>> structure is guaranteed to have the right frequency in policy->cur.
>>>
>>
>> This wording is slightly inaccurate. ->init() may or may not set policy->cur
>> (for example, powernowk8 driver doesn't set it in the init routine)..
> 
> Its not the wording that is wrong but this particular driver then :)
> This is what Documentation/cpu-drivers.txt says:
> 
> 1.2 Per-CPU Initialization
> Then, the driver must fill in the following values:
> 
> policy->cur The current operating frequency of
> this CPU (if appropriate)
> 
> And so it is supposed to do it.
>

Ah, I see.
 
>> But we set it for sure in __cpufreq_add_dev():
>>
>> 1117         if (cpufreq_driver->get) {
>> 1118                 policy->cur = cpufreq_driver->get(policy->cpu);
>> 1119                 if (!policy->cur) {
>> 1120                         pr_err("%s: ->get() failed\n", __func__);
>> 1121                         goto err_get_freq;
>> 1122                 }
>> 1123         }
> 
> Its just about removing that from drivers and doing it once in core :)
>

Ok..

Regards,
Srivatsa S. Bhat