[PATCH v3 1/2] cppc_cpufreq: Use desired perf if feedback ctrs are 0 or unchanged

Jie Zhan zhanjie9 at hisilicon.com
Thu Sep 26 01:44:36 PDT 2024



On 26/09/2024 14:07, lihuisong (C) wrote:
> 
> 在 2024/9/26 10:57, Jie Zhan 写道:
>>
>> On 25/09/2024 17:28, lihuisong (C) wrote:
>>> Hi Jie,
>>>
>>> LGTM except for some trivial,
>>> Reviewed-by: Huisong Li <lihuisong at huawei.com>
>> Thanks.
>>
>>>
>>> 在 2024/9/19 16:45, Jie Zhan 写道:
>>>> The CPPC performance feedback counters could be 0 or unchanged when the
>>>> target cpu is in a low-power idle state, e.g. power-gated or clock-gated.
>>>>
>>>> When the counters are 0, cppc_cpufreq_get_rate() returns 0 KHz, which makes
>>>> cpufreq_online() get a false error and fail to generate a cpufreq policy.
>>>>
>>>> When the counters are unchanged, the existing cppc_perf_from_fbctrs()
>>>> returns a cached desired perf, but some platforms may update the real
>>>> frequency back to the desired perf reg.
>>>>
>>>> For the above cases in cppc_cpufreq_get_rate(), get the latest desired perf
>>>> to reflect the frequency; if failed, return the cached desired perf.
>>>>
>>>> Fixes: 6a4fec4f6d30 ("cpufreq: cppc: cppc_cpufreq_get_rate() returns zero in all error cases.")
>>>> Signed-off-by: Jie Zhan <zhanjie9 at hisilicon.com>
>>>> Reviewed-by: Zeng Heng <zengheng4 at huawei.com>
>>>> Reviewed-by: Ionela Voinescu <ionela.voinescu at arm.com>
>>>> ---
>>>>    drivers/cpufreq/cppc_cpufreq.c | 49 +++++++++++++++++++++++++++-------
>>>>    1 file changed, 39 insertions(+), 10 deletions(-)
>>>>
>>>> diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c
>>>> index bafa32dd375d..e55192303a9f 100644
>>>> --- a/drivers/cpufreq/cppc_cpufreq.c
>>>> +++ b/drivers/cpufreq/cppc_cpufreq.c
>>>> @@ -118,6 +118,9 @@ static void cppc_scale_freq_workfn(struct kthread_work *work)
>>>>          perf = cppc_perf_from_fbctrs(cpu_data, &cppc_fi->prev_perf_fb_ctrs,
>>>>                         &fb_ctrs);
>>>> +    if (!perf)
>>>> +        return;
>>>> +
>>>>        cppc_fi->prev_perf_fb_ctrs = fb_ctrs;
>>>>          perf <<= SCHED_CAPACITY_SHIFT;
>>>> @@ -726,11 +729,26 @@ static int cppc_perf_from_fbctrs(struct cppc_cpudata *cpu_data,
>>>>          /* Check to avoid divide-by zero and invalid delivered_perf */
>>> Now this comment can be removed, right?
>> Didn't notice this comment, but, having a check, I think it still fits.
>> '!delta_reference' avoids divide-by zero, and '!delta_delivered' checks
>> invalid delivered_perf.
> The comment  "avoid divide-by zero" is just for the below code: "(reference_perf * delta_delivered) / delta_reference".
> So It is also useful, but I think It's obvious and it doesn't make much sense.
> 
> The comment "avoid invalid delivered_perf" is for the return value.
> Now this func return zero which can't count as a valid delivered_perf, right?

so, what about this?

/*
 * Avoid divide-by zero and unchanged feedback counters.
 * Leave it for callers to handle.
 */

>>
>> So I think we just leave it unchanged.
>>

...



More information about the linux-arm-kernel mailing list