| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <0e24a618-4a42-4fa8-b9ed-6d7db9b1a8fc@nvidia.com>
Date: Fri, 28 Nov 2025 19:59:18 +0530
From: Sumit Gupta <sumitg@...dia.com>
To: Pierre Gondois <pierre.gondois@....com>
Cc: linux-kernel@...r.kernel.org, corbet@....net,
acpica-devel@...ts.linux.dev, linux-doc@...r.kernel.org,
linux-acpi@...r.kernel.org, linux-pm@...r.kernel.org,
zhanjie9@...ilicon.com, ionela.voinescu@....com, perry.yuan@....com,
mario.limonciello@....com, ray.huang@....com, rdunlap@...radead.org,
zhenglifeng1@...wei.com, robert.moore@...el.com, lenb@...nel.org,
viresh.kumar@...aro.org, 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,
rafael@...nel.org, gautham.shenoy@....com, sumitg@...dia.com
Subject: Re: [PATCH v4 8/8] cpufreq: CPPC: add autonomous mode boot parameter
support
On 27/11/25 20:23, Pierre Gondois wrote:
> External email: Use caution opening links or attachments
>
>
> On 11/5/25 12:38, Sumit Gupta wrote:
>> Add kernel boot parameter 'cppc_cpufreq.auto_sel_mode' to enable CPPC
>> autonomous performance selection at system startup. When autonomous mode
>> is enabled, the hardware automatically adjusts CPU performance based on
>> workload demands using Energy Performance Preference (EPP) hints.
>>
>> This parameter allows to configure the autonomous mode on all CPUs
>> without requiring runtime sysfs manipulation if the 'auto_sel' register
>> is present.
>>
>> When auto_sel_mode=1:
>> - All CPUs are configured for autonomous operation during module init
>> - EPP is set to performance preference (0x0) by default
>> - Min/max performance bounds use defaults
>> - CPU frequency scaling is handled by hardware instead of OS governor
>>
>> For Documentation/:
>> Reviewed-by: Randy Dunlap<rdunlap@...radead.org>
>> Signed-off-by: Sumit Gupta<sumitg@...dia.com>
>> ---
>> .../admin-guide/kernel-parameters.txt | 12 ++
>> drivers/cpufreq/cppc_cpufreq.c | 197 +++++++++++++++---
>> 2 files changed, 182 insertions(+), 27 deletions(-)
>>
>> diff --git a/Documentation/admin-guide/kernel-parameters.txt
>> b/Documentation/admin-guide/kernel-parameters.txt
>> index b8f8f5d74093..048f84008a7e 100644
>> --- a/Documentation/admin-guide/kernel-parameters.txt
>> +++ b/Documentation/admin-guide/kernel-parameters.txt
>> @@ -929,6 +929,18 @@
>> Format:
>> <first_slot>,<last_slot>,<port>,<enum_bit>[,<debug>]
>>
>> + cppc_cpufreq.auto_sel_mode=
>> + [CPU_FREQ] Enable ACPI CPPC autonomous
>> performance selection.
>> + When enabled, hardware automatically adjusts
>> CPU frequency
>> + on all CPUs based on workload demands. In
>> Autonomous mode,
>> + Energy Performance Preference(EPP) hints guide
>> hardware
>> + toward performance(0x0) or energy efficiency
>> (0xff).
>> + Requires ACPI CPPC autonomous selection
>> register support.
>> + Format: <bool>
>> + Default: 0 (disabled)
>> + 0: use cpufreq governors
>> + 1: enable if supoorted by hardware
>> +
>> cpuidle.off=1 [CPU_IDLE]
>> disable the cpuidle sub-system
>>
>> diff --git a/drivers/cpufreq/cppc_cpufreq.c
>> b/drivers/cpufreq/cppc_cpufreq.c
>> index d1b44beaddda..0a55ab011317 100644
>> --- a/drivers/cpufreq/cppc_cpufreq.c
>> +++ b/drivers/cpufreq/cppc_cpufreq.c
>> @@ -28,8 +28,12 @@
>> #include <acpi/cppc_acpi.h>
>>
>> static struct cpufreq_driver cppc_cpufreq_driver;
>> +
>> static DEFINE_MUTEX(cppc_cpufreq_update_autosel_config_lock);
>>
>> +/* Autonomous Selection */
>> +static bool auto_sel_mode;
>> +
>> #ifdef CONFIG_ACPI_CPPC_CPUFREQ_FIE
>> static enum {
>> FIE_UNSET = -1,
>> @@ -272,8 +276,13 @@ static int cppc_cpufreq_set_target(struct
>> cpufreq_policy *policy,
>> freqs.old = policy->cur;
>> freqs.new = target_freq;
>>
>> + /*
>> + * In autonomous selection mode, hardware handles frequency
>> scaling directly
>> + * based on workload and EPP hints. So, skip the OS frequency
>> set requests.
>> + */
>> cpufreq_freq_transition_begin(policy, &freqs);
>> - ret = cppc_set_perf(cpu, &cpu_data->perf_ctrls);
>> + if (!cpu_data->perf_caps.auto_sel)
>> + ret = cppc_set_perf(cpu, &cpu_data->perf_ctrls);
>> cpufreq_freq_transition_end(policy, &freqs, ret != 0);
>>
>> if (ret)
>> @@ -565,6 +574,12 @@ static struct cppc_cpudata
>> *cppc_cpufreq_get_cpu_data(unsigned int cpu)
>> goto free_mask;
>> }
>>
>> + ret = cppc_get_perf(cpu, &cpu_data->perf_ctrls);
>> + if (ret) {
>> + pr_debug("Err reading CPU%d perf ctrls:ret:%d\n", cpu,
>> ret);
>> + goto free_mask;
>> + }
>> +
>> return cpu_data;
>>
>> free_mask:
>> @@ -666,11 +681,81 @@ static int
>> cppc_cpufreq_update_autosel_val(struct cpufreq_policy *policy, bool a
>> return 0;
>> }
>>
>> +static int cppc_cpufreq_update_epp_val(struct cpufreq_policy
>> *policy, u32 epp)
>> +{
>> + struct cppc_cpudata *cpu_data = policy->driver_data;
>> + unsigned int cpu = policy->cpu;
>> + int ret;
>> +
>> + pr_debug("cpu%d, eppcurr:%u,new:%u\n", cpu,
>> cpu_data->perf_ctrls.energy_perf, epp);
>> +
>> + guard(mutex)(&cppc_cpufreq_update_autosel_config_lock);
> Do we need to take the mutex ? Or is it reserved to auto_sel ?
Will move this to parent function.
Explained more in reply of the previous patch '7/8'.
>> +
>> + ret = cppc_set_epp(cpu, epp);
>> + if (ret) {
>> + pr_warn("failed to set energy_perf forcpu:%d (%d)\n",
>> cpu, ret);
>> + return ret;
>> + }
>> + cpu_data->perf_ctrls.energy_perf = epp;
>> +
>> + return 0;
>> +}
>> +
>> +/**
>> + * cppc_cpufreq_update_autosel_config - Update Autonomous selection
>> configuration
>> + * @policy: cpufreq policy for the CPU
>> + * @min_perf: minimum performance value to set
>> + * @max_perf: maximum performance value to set
>> + * @auto_sel: autonomous selection mode enable/disable (also
>> controls min/max perf reg updates)
>> + * @epp_val: energy performance preference value
>> + * @update_epp: whether to update EPP register
>> + * @update_policy: whether to update policy constraints
>> + *
>> + * Return: 0 on success, negative error code on failure
>> + */
>> +static int cppc_cpufreq_update_autosel_config(struct cpufreq_policy
>> *policy,
>> + u64 min_perf, u64
>> max_perf, bool auto_sel,
>> + u32 epp_val, bool
>> update_epp, bool update_policy)
>> +{
>> + const unsigned int cpu = policy->cpu;
>> + int ret;
>> +
>> + /*
>> + * Set min/max performance registers and update policy
>> constraints.
>> + * When enabling: update both registers and policy.
>> + * When disabling: update policy only.
>> + * Continue even if min/max are not supported, as EPP and autosel
>> + * might still be supported.
>> + */
>> + ret = cppc_cpufreq_set_min_perf(policy, min_perf, auto_sel,
>> update_policy);
>> + if (ret && ret != -EOPNOTSUPP)
>> + return ret;
>> +
>> + ret = cppc_cpufreq_set_max_perf(policy, max_perf, auto_sel,
>> update_policy);
>> + if (ret && ret != -EOPNOTSUPP)
>> + return ret;
>> +
>> + if (update_epp) {
>> + ret = cppc_cpufreq_update_epp_val(policy, epp_val);
>> + if (ret)
>> + return ret;
>> + }
>> +
>> + ret = cppc_cpufreq_update_autosel_val(policy, auto_sel);
>> + if (ret)
>> + return ret;
>> +
>> + pr_debug("Updated autonomous config [%llu-%llu] for CPU%d\n",
>> min_perf, max_perf, cpu);
>> +
>> + return 0;
>> +}
>> +
>> static int cppc_cpufreq_cpu_init(struct cpufreq_policy *policy)
>> {
>> unsigned int cpu = policy->cpu;
>> struct cppc_cpudata *cpu_data;
>> struct cppc_perf_caps *caps;
>> + u64 min_perf, max_perf;
>> int ret;
>>
>> cpu_data = cppc_cpufreq_get_cpu_data(cpu);
>> @@ -734,11 +819,31 @@ 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;
>>
>> - ret = cppc_set_perf(cpu, &cpu_data->perf_ctrls);
>> - if (ret) {
>> - pr_debug("Err setting perf value:%d on CPU:%d. ret:%d\n",
>> - caps->highest_perf, cpu, ret);
>> - goto out;
>> + if (cpu_data->perf_caps.auto_sel) {
>> + ret = cppc_set_enable(cpu, true);
> The CPPC enable register is optional.
> However this doesn't mean CPPC is not working.
Ya, changed this in v5.
>> + if (ret) {
>> + pr_err("Failed to enable CPPC on cpu%d (%d)\n",
>> cpu, ret);
>> + goto out;
>> + }
>> +
>> + min_perf = cpu_data->perf_ctrls.min_perf ?
>> + cpu_data->perf_ctrls.min_perf :
>> caps->lowest_nonlinear_perf;
>> + max_perf = cpu_data->perf_ctrls.max_perf ?
>> + cpu_data->perf_ctrls.max_perf :
>> caps->nominal_perf;
>> +
>> + ret = cppc_cpufreq_update_autosel_config(policy,
>> min_perf, max_perf, true,
>> + CPPC_EPP_PERFORMANCE_PREF, true, false);
>> + if (ret) {
>> + cppc_set_enable(cpu, false);
>> + goto out;
>> + }
>> + } else {
>> + ret = cppc_set_perf(cpu, &cpu_data->perf_ctrls);
>> + if (ret) {
>> + pr_debug("Err setting perf value:%d on CPU:%d.
>> ret:%d\n",
>> + caps->highest_perf, cpu, ret);
>> + goto out;
>> + }
>> }
>>
>> cppc_cpufreq_cpu_fie_init(policy);
>> @@ -910,7 +1015,6 @@ static int
>> cppc_cpufreq_update_auto_select(struct cpufreq_policy *policy, bool e
>> struct cppc_perf_caps *caps = &cpu_data->perf_caps;
>> u64 min_perf = caps->lowest_nonlinear_perf;
>> u64 max_perf = caps->nominal_perf;
>> - int ret;
>>
>> if (enable) {
>> if (cpu_data->perf_ctrls.min_perf)
>> @@ -919,26 +1023,8 @@ static int
>> cppc_cpufreq_update_auto_select(struct cpufreq_policy *policy, bool e
>> max_perf = cpu_data->perf_ctrls.max_perf;
>> }
>>
>> - /*
>> - * Set min/max performance registers and update policy
>> constraints.
>> - * When enabling: update both registers and policy.
>> - * When disabling: update policy only.
>> - * Continue even if min/max are not supported, as EPP and autosel
>> - * might still be supported.
>> - */
>> - ret = cppc_cpufreq_set_min_perf(policy, min_perf, enable, true);
>> - if (ret && ret != -EOPNOTSUPP)
>> - return ret;
>> -
>> - ret = cppc_cpufreq_set_max_perf(policy, max_perf, enable, true);
>> - if (ret && ret != -EOPNOTSUPP)
>> - return ret;
>> -
>> - ret = cppc_cpufreq_update_autosel_val(policy, enable);
>> - if (ret)
>> - return ret;
>> -
>> - return 0;
>> + return cppc_cpufreq_update_autosel_config(policy, min_perf,
>> max_perf, enable,
>> + 0, false, true);
>> }
>>
>> static ssize_t store_auto_select(struct cpufreq_policy *policy,
>> const char *buf, size_t count)
>> @@ -1146,13 +1232,61 @@ static struct cpufreq_driver
>> cppc_cpufreq_driver = {
>> .name = "cppc_cpufreq",
>> };
>>
>> +static int cppc_cpufreq_set_epp_autosel_allcpus(bool auto_sel, u64 epp)
>> +{
>> + int cpu, ret;
>> +
>> + for_each_present_cpu(cpu) {
>> + ret = cppc_set_epp(cpu, epp);
> Isn't the EPP optional ?
Moving this to cppc_cpufreq_cpu_init in v5. Will add handling for
EOPNOTSUPP.
> If autonomous selection is available but not EPP, we will bail out.
I couldn't find in spec that EPP is mandatory when auto_select is enabled.
>> + if (ret) {
>> + pr_warn("Failed to set EPP on CPU%d (%d)\n",
>> cpu, ret);
>> + goto disable_all;
>> + }
>> +
>> + ret = cppc_set_auto_sel(cpu, auto_sel);
>
> Also, it is possible that a platform only supports autonomous selection.
> In this case, writing to auto_sel will fail, but auto_sel is still
> relevant.
I am not sure if we will have such platform which only supports Autonomous
mode and has auto_sel as read only. Will add handling for EOPNOTSUPP if we
have such cases as the cppc_get_reg_val() will returns this error.
Thank you,
Sumit Gupta
....
Powered by blists - more mailing lists