Re: [PATCH v2 10/10] drivers base/arch_topology: inline cpu- and frequency-invariant accounting

[Dietmar Eggemann]

On 06/07/17 11:57, Viresh Kumar wrote:
> Sure this patch looks pretty useful, but ...
> 
> On 06-07-17, 10:49, Dietmar Eggemann wrote:
>> diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
>> index 63fb3f945d21..b4481cff14bf 100644
>> --- a/drivers/base/arch_topology.c
>> +++ b/drivers/base/arch_topology.c
>> @@ -22,12 +22,7 @@
>>  #include <linux/string.h>
>>  #include <linux/sched/topology.h>
>>  
>> -static DEFINE_PER_CPU(unsigned long, freq_scale) = SCHED_CAPACITY_SCALE;
>> -
>> -unsigned long topology_get_freq_scale(struct sched_domain *sd, int cpu)
>> -{
>> -	return per_cpu(freq_scale, cpu);
>> -}
>> +DEFINE_PER_CPU(unsigned long, freq_scale) = SCHED_CAPACITY_SCALE;
> 
> ... you just undo what you did earlier in this series, and that is somewhat
> discouraged.
> 
> What about making this as the first patch of the series and move only the below
> part to the header. And then you can add the above part to the right place in
> the first attempt itself?
> 
> But maybe this is all okay :)

I just wanted to show people what we gain in completely inlining FIE and
CIE on ARM64 in the scheduler hot-path. But yes, with the next version I
want to fold this inlining into the actual FIE/CIE patch.