On 26/12/25 08:56, zhenglifeng (A) wrote:

External email: Use caution opening links or attachments


On 2025/12/23 20:13, Sumit Gupta wrote:

quoted

When autonomous selection (auto_sel) is enabled, the hardware controls
performance within min_perf/max_perf register bounds making the
scaling_min/max_freq effectively read-only.

Enforce this by setting policy limits to min/max_perf bounds in
cppc_verify_policy(). Users must use min_perf/max_perf sysfs interfaces
to change performance limits in autonomous mode.

Signed-off-by: Sumit Gupta <sumitg@nvidia.com>
---
  drivers/cpufreq/cppc_cpufreq.c | 32 +++++++++++++++++++++++++++++++-
  1 file changed, 31 insertions(+), 1 deletion(-)

diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c
index b1f570d6de34..b3da263c18b0 100644
--- a/drivers/cpufreq/cppc_cpufreq.c
+++ b/drivers/cpufreq/cppc_cpufreq.c

@@ -305,7 +305,37 @@ static unsigned int cppc_cpufreq_fast_switch(struct cpufreq_policy *policy,

  static int cppc_verify_policy(struct cpufreq_policy_data *policy)
  {
-     cpufreq_verify_within_cpu_limits(policy);
+     unsigned int min_freq = policy->cpuinfo.min_freq;
+     unsigned int max_freq = policy->cpuinfo.max_freq;
+     struct cpufreq_policy *cpu_policy;
+     struct cppc_cpudata *cpu_data;
+     struct cppc_perf_caps *caps;
+
+     cpu_policy = cpufreq_cpu_get(policy->cpu);

Better to use:

         struct cpufreq_policy *cpu_policy __free(put_cpufreq_policy) = cpufreq_cpu_get(policy->cpu);

Will use this in v6.

quoted

+     if (!cpu_policy)
+             return -ENODEV;
+
+     cpu_data = cpu_policy->driver_data;
+     caps = &cpu_data->perf_caps;

cpu_policy, cpu_data and cpas are only used in the if branch. Just put them
in it.

Can move caps inside the if branch.
cpu_policy and cpu_data can't be moved inside because we need
perf_ctrls.auto_sel to evaluate the condition itself.

quoted

+
+     if (cpu_data->perf_ctrls.auto_sel) {
+             u32 min_perf, max_perf;
+
+             /*
+              * Set policy limits to HW min/max_perf bounds. In autonomous
+              * mode, scaling_min/max_freq is effectively read-only.
+              */
+             min_perf = cpu_data->perf_ctrls.min_perf ?:
+                        caps->lowest_nonlinear_perf;
+             max_perf = cpu_data->perf_ctrls.max_perf ?: caps->nominal_perf;
+
+             policy->min = cppc_perf_to_khz(caps, min_perf);
+             policy->max = cppc_perf_to_khz(caps, max_perf);
+     } else {
+             cpufreq_verify_within_limits(policy, min_freq, max_freq);

Why not still using cpufreq_verify_within_cpu_limits()?

Will change to use it in v6.

Thank you,
Sumit Gupta

quoted

+     }
+
+     cpufreq_cpu_put(cpu_policy);
       return 0;
  }