Message-ID: <2e0e7b5d-e424-4a45-9783-178a1af24ccc@nvidia.com>
Date: Thu, 8 Jan 2026 19:23:54 +0530
From: Sumit Gupta <sumitg@...dia.com>
To: "zhenglifeng (A)" <zhenglifeng1@...wei.com>, rafael@...nel.org,
 viresh.kumar@...aro.org, lenb@...nel.org, robert.moore@...el.com,
 corbet@....net, pierre.gondois@....com, rdunlap@...radead.org,
 ray.huang@....com, gautham.shenoy@....com, mario.limonciello@....com,
 perry.yuan@....com, ionela.voinescu@....com, zhanjie9@...ilicon.com,
 linux-pm@...r.kernel.org, linux-acpi@...r.kernel.org,
 linux-doc@...r.kernel.org, acpica-devel@...ts.linux.dev,
 linux-kernel@...r.kernel.org
Cc: linux-tegra@...r.kernel.org, treding@...dia.com, jonathanh@...dia.com,
 vsethi@...dia.com, ksitaraman@...dia.com, sanjayc@...dia.com,
 nhartman@...dia.com, bbasu@...dia.com, sumitg@...dia.com
Subject: Re: [PATCH v5 08/11] cpufreq: CPPC: sync policy limits when updating
 min/max_perf


On 25/12/25 19:26, zhenglifeng (A) wrote:
> External email: Use caution opening links or attachments
>
>
> On 2025/12/23 20:13, Sumit Gupta wrote:
>> When min_perf or max_perf is updated via sysfs in autonomous mode, the
>> policy frequency limits should also be updated to reflect the new
>> performance bounds.
>>
>> Add @update_policy parameter to cppc_cpufreq_set_mperf_limit() to
>> control whether policy constraints are synced with HW registers.
>> The policy is updated only when autonomous selection is enabled to
>> keep SW limits in sync with HW.
>>
>> This ensures that scaling_min_freq and scaling_max_freq values remain
>> consistent with the actual min/max_perf register values when operating
>> in autonomous mode.
>>
>> Signed-off-by: Sumit Gupta <sumitg@...dia.com>
>> ---
>>   drivers/cpufreq/cppc_cpufreq.c | 35 ++++++++++++++++++++++++++--------
>>   1 file changed, 27 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c
>> index 1f8825006940..0202c7b823e6 100644
>> --- a/drivers/cpufreq/cppc_cpufreq.c
>> +++ b/drivers/cpufreq/cppc_cpufreq.c
>> @@ -544,14 +544,20 @@ static void populate_efficiency_class(void)
>>    * cppc_cpufreq_set_mperf_limit - Set min/max performance limit
>>    * @policy: cpufreq policy
>>    * @val: performance value to set
>> + * @update_policy: whether to update policy constraints
>>    * @is_min: true for min_perf, false for max_perf
>> + *
>> + * When @update_policy is true, updates cpufreq policy frequency limits.
>> + * @update_policy is false during cpu_init when policy isn't fully set up.
>>    */
>>   static int cppc_cpufreq_set_mperf_limit(struct cpufreq_policy *policy, u64 val,
>> -                                     bool is_min)
>> +                                     bool update_policy, bool is_min)
>>   {
>>        struct cppc_cpudata *cpu_data = policy->driver_data;
>>        struct cppc_perf_caps *caps = &cpu_data->perf_caps;
>>        unsigned int cpu = policy->cpu;
>> +     struct freq_qos_request *req;
>> +     unsigned int freq;
>>        u32 perf;
>>        int ret;
>>
>> @@ -571,15 +577,26 @@ static int cppc_cpufreq_set_mperf_limit(struct cpufreq_policy *policy, u64 val,
>>        else
>>                cpu_data->perf_ctrls.max_perf = perf;
>>
>> +     if (update_policy) {
>> +             freq = cppc_perf_to_khz(caps, perf);
>> +             req = is_min ? policy->min_freq_req : policy->max_freq_req;
>> +
>> +             ret = freq_qos_update_request(req, freq);
>> +             if (ret < 0) {
>> +                     pr_warn("Failed to update %s_freq constraint for CPU%d: %d\n",
>> +                             is_min ? "min" : "max", cpu, ret);
>> +                     return ret;
>> +             }
>> +     }
>> +
> OK. Now I see the necessity of extracting this function. But why not use
> freq_khz as a input parameter and convert it to perf in this funciton,
> since you need the freq here?

That will still need cppc_perf_to_khz to be called so that policy
has what HW actually delivers. Otherwise, there could be some
asymmetry.
Also the clamping is done on perf values. So, if user provides a
very high freq value then that will get passed to freq_qos and the
HW register will have actual perf value which doesn't match with qos.

Either way the conversion chain is:
   freq_to_perf -> clamp perf -> set perf -> perf_to_freq -> set qos
It's just a matter of where we place the logic.

Thank you,
Sumit Gupta

>>        return 0;
>>   }
>>
>> -#define cppc_cpufreq_set_min_perf(policy, val) \
>> -     cppc_cpufreq_set_mperf_limit(policy, val, true)
>> -
>> -#define cppc_cpufreq_set_max_perf(policy, val) \
>> -     cppc_cpufreq_set_mperf_limit(policy, val, false)
>> +#define cppc_cpufreq_set_min_perf(policy, val, update_policy) \
>> +     cppc_cpufreq_set_mperf_limit(policy, val, update_policy, true)
>>
>> +#define cppc_cpufreq_set_max_perf(policy, val, update_policy) \
>> +     cppc_cpufreq_set_mperf_limit(policy, val, update_policy, false)
>>   static struct cppc_cpudata *cppc_cpufreq_get_cpu_data(unsigned int cpu)
>>   {
>>        struct cppc_cpudata *cpu_data;
>> @@ -988,7 +1005,8 @@ static ssize_t store_min_perf(struct cpufreq_policy *policy, const char *buf,
>>        perf = cppc_khz_to_perf(&cpu_data->perf_caps, freq_khz);
>>
>>        guard(mutex)(&cppc_cpufreq_update_autosel_config_lock);
>> -     ret = cppc_cpufreq_set_min_perf(policy, perf);
>> +     ret = cppc_cpufreq_set_min_perf(policy, perf,
>> +                                     cpu_data->perf_ctrls.auto_sel);
>>        if (ret)
>>                return ret;
>>
>> @@ -1045,7 +1063,8 @@ static ssize_t store_max_perf(struct cpufreq_policy *policy, const char *buf,
>>        perf = cppc_khz_to_perf(&cpu_data->perf_caps, freq_khz);
>>
>>        guard(mutex)(&cppc_cpufreq_update_autosel_config_lock);
>> -     ret = cppc_cpufreq_set_max_perf(policy, perf);
>> +     ret = cppc_cpufreq_set_max_perf(policy, perf,
>> +                                     cpu_data->perf_ctrls.auto_sel);
>>        if (ret)
>>                return ret;
>>

Powered by blists - more mailing lists