2016-10-13 17:42:18

by Hoan Tran

[permalink] [raw]
Subject: [PATCH v2] cpufreq: CPPC: Correct desired_perf calculation

The desired_perf is an abstract performance number. Its value should
be in the range of [lowest perf, highest perf] of CPPC.
The correct calculation is
desired_perf = freq * cppc_highest_perf / cppc_dmi_max_khz

And cppc_cpufreq_set_target() returns if desired_perf is exactly
the same with the old perf.

Signed-off-by: Hoan Tran <[email protected]>
---
v2
* Return if desired_perf is the same with the old one (Prashanth's suggestion)


drivers/cpufreq/cppc_cpufreq.c | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c
index 1b2f28f..4852d9e 100644
--- a/drivers/cpufreq/cppc_cpufreq.c
+++ b/drivers/cpufreq/cppc_cpufreq.c
@@ -80,11 +80,17 @@ static int cppc_cpufreq_set_target(struct cpufreq_policy *policy,
{
struct cppc_cpudata *cpu;
struct cpufreq_freqs freqs;
+ u32 desired_perf;
int ret = 0;

cpu = all_cpu_data[policy->cpu];

- cpu->perf_ctrls.desired_perf = (u64)target_freq * policy->max / cppc_dmi_max_khz;
+ desired_perf = (u64)target_freq * cpu->perf_caps.highest_perf / cppc_dmi_max_khz;
+ /* Return if it is exactly the same perf */
+ if (desired_perf == cpu->perf_ctrls.desired_perf)
+ return ret;
+
+ cpu->perf_ctrls.desired_perf = desired_perf;
freqs.old = policy->cur;
freqs.new = target_freq;

--
1.9.1


2016-10-13 19:47:03

by Prakash, Prashanth

[permalink] [raw]
Subject: Re: [PATCH v2] cpufreq: CPPC: Correct desired_perf calculation

On 10/13/2016 11:33 AM, Hoan Tran wrote:
> The desired_perf is an abstract performance number. Its value should
> be in the range of [lowest perf, highest perf] of CPPC.
> The correct calculation is
> desired_perf = freq * cppc_highest_perf / cppc_dmi_max_khz
>
> And cppc_cpufreq_set_target() returns if desired_perf is exactly
> the same with the old perf.
>
> Signed-off-by: Hoan Tran <[email protected]>
Reviewed-by: Prashanth Prakash <[email protected]>

Hi Rafael,
Can you please pick this one for 4.9, as it is bug fix?

--
Thanks,
Prashanth

2016-10-13 21:12:21

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: [PATCH v2] cpufreq: CPPC: Correct desired_perf calculation

On Thu, Oct 13, 2016 at 9:37 PM, Prakash, Prashanth
<[email protected]> wrote:
> On 10/13/2016 11:33 AM, Hoan Tran wrote:
>> The desired_perf is an abstract performance number. Its value should
>> be in the range of [lowest perf, highest perf] of CPPC.
>> The correct calculation is
>> desired_perf = freq * cppc_highest_perf / cppc_dmi_max_khz
>>
>> And cppc_cpufreq_set_target() returns if desired_perf is exactly
>> the same with the old perf.
>>
>> Signed-off-by: Hoan Tran <[email protected]>
> Reviewed-by: Prashanth Prakash <[email protected]>
>
> Hi Rafael,
> Can you please pick this one for 4.9, as it is bug fix?

Sure, I will queue it up.

Thanks,
Rafael

2016-10-13 21:22:14

by Hoan Tran

[permalink] [raw]
Subject: Re: [PATCH v2] cpufreq: CPPC: Correct desired_perf calculation

On Thu, Oct 13, 2016 at 2:12 PM, Rafael J. Wysocki <[email protected]> wrote:
> On Thu, Oct 13, 2016 at 9:37 PM, Prakash, Prashanth
> <[email protected]> wrote:
>> On 10/13/2016 11:33 AM, Hoan Tran wrote:
>>> The desired_perf is an abstract performance number. Its value should
>>> be in the range of [lowest perf, highest perf] of CPPC.
>>> The correct calculation is
>>> desired_perf = freq * cppc_highest_perf / cppc_dmi_max_khz
>>>
>>> And cppc_cpufreq_set_target() returns if desired_perf is exactly
>>> the same with the old perf.
>>>
>>> Signed-off-by: Hoan Tran <[email protected]>
>> Reviewed-by: Prashanth Prakash <[email protected]>
>>
>> Hi Rafael,
>> Can you please pick this one for 4.9, as it is bug fix?
>
> Sure, I will queue it up.
>

Thanks, Rafael and Prashanth !

Hoan

> Thanks,
> Rafael