Re: [RFC 1/2] cpuidle: trace state of the CPU

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 01/31/2013 06:21 AM, Namhyung Kim wrote:
>> diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
>> index e1f6860..3594e0c 100644
>> --- a/drivers/cpuidle/cpuidle.c
>> +++ b/drivers/cpuidle/cpuidle.c
>> @@ -23,6 +23,7 @@
>>  #include "cpuidle.h"
>>  
>>  DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
>> +static DEFINE_PER_CPU(unsigned int, cpu_state);
> 
> What about making it 'int' as cpuidle_get_state() returns int type?

index should never get negative because it is an index. In case it does
(which should not) we end up with a large value which is considered as
a really deep state and the CPU will be ignored as long as possible.

>> @@ -40,13 +41,23 @@ void disable_cpuidle(void)
…

>>  static inline int cpuidle_enter(struct cpuidle_device *dev,
>>  				struct cpuidle_driver *drv, int index)
>>  {
>>  	struct cpuidle_state *target_state = &drv->states[index];
>> -	return target_state->enter(dev, drv, index);
>> +	int ret;
>> +
>> +	per_cpu(cpu_state, smp_processor_id()) = index + 1;
>> +	ret = target_state->enter(dev, drv, index);
>> +	per_cpu(cpu_state, smp_processor_id()) = 0;
> 
> Maybe we can use local variable 'cpu' for this duplicated
> smp_processor_id() call as it's never executed on another cpu in
> between?

This should be really cheap unless CONFIG_DEBUG_PREEMPT is on but okay,
will change.

> Thanks,
> Namhyung

Sebastian