Re: [PATCH v2 3/3] cpufreq: CPPC: Preserve OSPM-set registers across hotplug and unload

[Sumit Gupta <sumitg@nvidia.com>]

Hi Christian,

Thanks for the review.


On 22/07/26 15:47, Christian Loehle wrote:
> External email: Use caution opening links or attachments
>
>
> Hi Sumit
>
> On 7/16/26 16:38, Sumit Gupta wrote:
>> Values written to OSPM-set CPPC registers (via sysfs or the autonomous
>> boot parameter) can be lost in two ways:
>>
>>    - Across CPU hotplug: the platform may reset a CPU's registers while it
>>      is offline.
>>    - On driver unload: the value the driver wrote is left in the register
>>      instead of returning to its pre-driver state.
>>
>> Add a small table-driven mechanism that handles both:
>>
>>    - On init(), capture each register's firmware value before the
>>      driver programs anything.
>>    - On offline(), read back each register's current value (whatever was
>>      last set via sysfs or the boot parameter) so it can be reapplied, then
>>      restore the firmware value.
>>    - On online(), reapply the value captured at offline().
>>
>> Cover the Autonomous Selection (auto_sel), Energy Performance Preference
>> (EPP) and Autonomous Activity Window (auto_act_window) registers.
>>
>> Suggested-by: Pierre Gondois <pierre.gondois@arm.com>
>> Link: https://lore.kernel.org/all/86780f97-29ee-4a72-b311-38c89434b707@arm.com/
>> Signed-off-by: Sumit Gupta <sumitg@nvidia.com>
>> ---
>>   drivers/cpufreq/cppc_cpufreq.c | 130 +++++++++++++++++++++++++++++++++
>>   1 file changed, 130 insertions(+)
>>
>> diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c
>> index 432c6a6288a7..9c88512d635c 100644
>> --- a/drivers/cpufreq/cppc_cpufreq.c
>> +++ b/drivers/cpufreq/cppc_cpufreq.c
>> @@ -28,6 +28,123 @@
>>
>>   static struct cpufreq_driver cppc_cpufreq_driver;
>>
>> +/*
>> + * OSPM-set CPPC registers tracked for save/restore. A value set via sysfs or
>> + * the autonomous boot parameter is reapplied from online() across CPU
>> + * hotplug, and the firmware value is restored from offline().
>> + */
>> +enum cppc_saved_reg_id {
>> +     CPPC_SAVED_AUTO_SEL,
>> +     CPPC_SAVED_EPP,
>> +     CPPC_SAVED_AUTO_ACT_WINDOW,
>> +     CPPC_NR_SAVED_REGS,
>> +};
>> +
>> +struct cppc_saved_reg {
>> +     int (*get)(int cpu, u64 *val);
>> +     int (*set)(int cpu, u64 val);
>> +};
>> +
>> +static const struct cppc_saved_reg cppc_saved_regs[CPPC_NR_SAVED_REGS] = {
>> +     [CPPC_SAVED_AUTO_SEL] = {
>> +             cppc_get_auto_sel_u64, cppc_set_auto_sel_u64,
>> +     },
>> +     [CPPC_SAVED_EPP] = {
>> +             cppc_get_epp_perf, cppc_set_epp,
>> +     },
>> +     [CPPC_SAVED_AUTO_ACT_WINDOW] = {
>> +             cppc_get_auto_act_window, cppc_set_auto_act_window,
>> +     },
>> +};
>> +
>> +/*
>> + * Per-policy saved state for each register in cppc_saved_regs[]:
>> + *   firmware_val  - value before the driver touched it, captured at init()
>> + *                   and restored while the policy is offline. U64_MAX if it
>> + *                   could not be read
>> + *   requested_val - value in effect when the policy last went offline,
>> + *                   reapplied at online(). U64_MAX if none
>> + */
>> +struct cppc_saved_state {
>> +     u64 firmware_val;
>> +     u64 requested_val;
>> +};
>> +
>> +static DEFINE_PER_CPU(struct cppc_saved_state[CPPC_NR_SAVED_REGS], cppc_saved_state);
>> +
>> +/*
>> + * Return this policy's saved state. Each policy keeps a single copy, stored in
>> + * the per-CPU variable of the first CPU it manages. related_cpus (the policy's
>> + * full set of CPUs) never changes while it exists, so this CPU (unlike
>> + * policy->cpu) stays the same across CPU hotplug, and every callback reaches
>> + * the same copy.
>> + */
>> +static struct cppc_saved_state *cppc_cpufreq_policy_saved_state(struct cpufreq_policy *policy)
>> +{
>> +     const struct cpumask *policy_cpus = policy->related_cpus;
>> +
>> +     /*
>> +      * related_cpus is empty until the core fills it in after init(). Until
>> +      * then, fall back to policy->cpus, which has the same first CPU.
>> +      */
>> +     if (cpumask_empty(policy_cpus))
>> +             policy_cpus = policy->cpus;
>> +
>> +     return per_cpu(cppc_saved_state, cpumask_first(policy_cpus));
>> +}
>> +
>> +/*
>> + * Capture each register's firmware value before the driver programs anything.
>> + */
>> +static void cppc_cpufreq_save_firmware_regs(struct cpufreq_policy *policy)
>> +{
>> +     struct cppc_saved_state *st = cppc_cpufreq_policy_saved_state(policy);
>> +     unsigned int cpu = policy->cpu;
>> +     u64 val;
>> +     int i;
>> +
>> +     for (i = 0; i < CPPC_NR_SAVED_REGS; i++) {
>> +             if (cppc_saved_regs[i].get(cpu, &val))
>> +                     val = U64_MAX;
>> +             st[i].firmware_val = val;
>> +             st[i].requested_val = U64_MAX;
>> +     }
>> +}
>> +
>> +/*
>> + * Save each register's current value so online() can later reapply it, then
>> + * restore the firmware value to leave the platform in its pre-driver state.
>> + */
>> +static void
>> +cppc_cpufreq_save_req_and_restore_firmware_regs(struct cpufreq_policy *policy)
>> +{
>> +     struct cppc_saved_state *st = cppc_cpufreq_policy_saved_state(policy);
>> +     unsigned int cpu = policy->cpu;
>> +     u64 val;
>> +     int i;
>> +
>> +     for (i = 0; i < CPPC_NR_SAVED_REGS; i++) {
>> +             if (!cppc_saved_regs[i].get(cpu, &val))
>> +                     st[i].requested_val = val;
>> +             if (st[i].firmware_val != U64_MAX)
>> +                     cppc_saved_regs[i].set(cpu, st[i].firmware_val);
>> +     }
>> +}
>> +
>> +/*
>> + * Reapply each register's requested value that offline() saved.
>> + */
>> +static void cppc_cpufreq_reapply_requested_regs(struct cpufreq_policy *policy)
>> +{
>> +     struct cppc_saved_state *st = cppc_cpufreq_policy_saved_state(policy);
>> +     unsigned int cpu = policy->cpu;
>> +     int i;
>> +
>> +     for (i = 0; i < CPPC_NR_SAVED_REGS; i++)
>> +             if (st[i].requested_val != U64_MAX)
>> +                     cppc_saved_regs[i].set(cpu, st[i].requested_val);
>> +}
>> +
>>   #ifdef CONFIG_ACPI_CPPC_CPUFREQ_FIE
>>   static enum {
>>        FIE_UNSET = -1,
>> @@ -707,6 +824,8 @@ static int cppc_cpufreq_cpu_init(struct cpufreq_policy *policy)
>>        policy->cur = cppc_perf_to_khz(caps, caps->highest_perf);
>>        cpu_data->perf_ctrls.desired_perf =  caps->highest_perf;
>>
>> +     cppc_cpufreq_save_firmware_regs(policy);
>> +
>>        ret = cppc_set_perf(cpu, &cpu_data->perf_ctrls);
>>        if (ret) {
>>                pr_debug("Err setting perf value:%d on CPU:%d. ret:%d\n",
>> @@ -725,15 +844,24 @@ static int cppc_cpufreq_cpu_init(struct cpufreq_policy *policy)
>>   /*
>>    * With offline() defined, the cpufreq core keeps the policy alive when
>>    * a CPU is hotplugged out.
>> + *
>> + * Save each register's current value so online() can reapply it, then restore
>> + * the firmware value, leaving the platform in its pre-driver state while the
>> + * policy is down (CPU hotplug or driver unload).
>>    */
>>   static int cppc_cpufreq_cpu_offline(struct cpufreq_policy *policy)
>>   {
>> +     cppc_cpufreq_save_req_and_restore_firmware_regs(policy);
>> +
>>        return 0;
>>   }
>>
>>   /*
>>    * Re-enable CPPC when the policy's CPU comes back online, since the platform
>>    * may have disabled it while the CPU was offline.
>> + *
>> + * offline() reset the registers to their firmware values, so reapply the
>> + * OSPM-set values it saved.
>>    */
>>   static int cppc_cpufreq_cpu_online(struct cpufreq_policy *policy)
>>   {
>> @@ -744,6 +872,8 @@ static int cppc_cpufreq_cpu_online(struct cpufreq_policy *policy)
>>        if (ret && ret != -EOPNOTSUPP)
>>                pr_warn("Failed to re-enable CPPC for CPU%d (%d)\n", cpu, ret);
>>
>> +     cppc_cpufreq_reapply_requested_regs(policy);
>> +
>>        return 0;
>>   }
>>
> I had a look as well and I think the saved registers aren't enough for the stated hotplug problem.
>
> CPPC also programs DESIRED_PERF, MIN_PERF and MAX_PERF. If the platform resets those controls
> while a policy is offline, online() your series reapplies AUTO_SEL_ENABLE, EPP and the activity
> window but leaves the main performance request unrestored.
> So autonomous selection may be re-enabled while MIN_PERF and MAX_PERF still contain
> reset values.
>
> I do not think these controls should simply be added to cppc_saved_regs[]:
> - MIN_PERF and MAX_PERF should be derived from the current policy, which may have changed.
> - After state loss, direct controls require safe ordering: establish [0, all-ones], restore
> a clamped DESIRED_PERF, then install the final bounds.
> PCC controls make this more complex, they should remain grouped in a coordinated transaction.
>
> Could the hotplug state instead retain the last requested DESIRED value, recompute the bounds
> during online(), restore the complete tuple, and only then reapply AUTO_SEL_ENABLE?
>

Agreed.
Will change online() to derive MIN/MAX from the current policy,
clamp the retained cpu_data->perf_ctrls.desired_perf into that range and
reprogram the tuple with cppc_set_perf() (one coordinated transaction
for PCC) before AUTO_SEL is reapplied.
With desired_perf clamped to the final [MIN, MAX] range, the temporary
[0, all-ones] widening is not needed.

Regards,
Sumit