1. Summary

When deploying OpenStack compute nodes with edpm_tuned_max_power_cstate: "c6" and the cpu-partitioning-powersave tuned profile, the C6 deep sleep state is not actually enabled at the hardware level, preventing any measurable power savings.

2. Environment

• RHEL 9.x / RHOSO 18
• Tuned profile: cpu-partitioning-powersave
• Hardware: Intel Xeon E5-2630 v4 (Broadwell)
• Deployment config:
    edpm_tuned_profile: "cpu-partitioning-powersave"
    edpm_tuned_max_power_cstate: "c6"
    edpm_tuned_isolated_cores: "2-19,22-39"

3. Problem Description

Despite configuring the powersave profile with C6 support, CPUs never enter C6 state and power consumption remains identical to baseline mode.

3.1 Expected Behavior

When edpm_tuned_max_power_cstate: "c6" is set:
1. Tuned should configure max_power_state=cstate.name:C6|140
2. C6 state should be enabled at kernel level
3. CPUs should enter C6 during idle periods
4. Power consumption should decrease compared to baseline

3.2 Actual Behavior

1. [PASS] Tuned sets max_power_state=cstate.name:C6|140 correctly
2. [FAIL] C6 remains disabled: /sys/devices/system/cpu/cpu*/cpuidle/state4/disable = 1
3. [FAIL] CPUs never enter C6 state
4. [FAIL] Power consumption identical to baseline mode

4. Reproduction Steps

4.1 Deploy with powersave profile

Configure deployment with:
  edpm_tuned_profile: "cpu-partitioning-powersave"
  edpm_tuned_max_power_cstate: "c6"

4.2 Verify tuned configuration

Check active profile:
  ssh cloud-admin@compute-node "tuned-adm active"
  Output: Current active profile: cpu-partitioning-powersave

Check tuned configuration:
  ssh cloud-admin@compute-node "grep max_power_state /etc/tuned/cpu-partitioning-powersave-variables.conf"
  Output: max_power_state=cstate.name:C6|140

4.3 Check C6 hardware state

Command:
  ssh cloud-admin@compute-node "cat /sys/devices/system/cpu/cpu0/cpuidle/state4/disable"
  Output: 1  (DISABLED) <-- BUG: Should be 0

4.4 Measure power consumption

Using Redfish API to measure actual power consumption:

Get baseline with cpu-partitioning profile:
  tuned-adm profile cpu-partitioning
  sleep 60
  Measure: ~147W per node

Switch to powersave profile:
  tuned-adm profile cpu-partitioning-powersave
  sleep 60
  Measure: ~147W per node (NO DIFFERENCE)

Power measurement via iDRAC Redfish API:
  curl -k -u root:password https://idrac-ip/redfish/v1/Chassis/System.Embedded.1/Power | jq '.PowerControl[0].PowerConsumedWatts'

4.5 Check C6 usage statistics

Check C6 time:
  ssh cloud-admin@compute-node "cat /sys/devices/system/cpu/cpu0/cpuidle/state4/time"
  Output: 0  (never entered C6)

Check C6 usage count:
  ssh cloud-admin@compute-node "cat /sys/devices/system/cpu/cpu0/cpuidle/state4/usage"
  Output: 0  (never entered C6)

5. Workaround

Manually enable C6 at kernel level:

  ssh cloud-admin@compute-node "for cpu in /sys/devices/system/cpu/cpu*/cpuidle/state4/disable; do echo 0 | sudo tee \$cpu > /dev/null; done"

After applying workaround:

• C6 state becomes enabled (disable=0)
• CPUs enter C6 during idle periods
• Power consumption reduces to ~130-135W (10-15% savings)
• C6 usage statistics show active usage

6. Verification After Workaround

Step 1 - Verify C6 is enabled:
  cat /sys/devices/system/cpu/cpu0/cpuidle/state4/disable
  Output: 0 (ENABLED)

Step 2 - Wait 60 seconds for idle time

Step 3 - Check C6 usage:
  cat /sys/devices/system/cpu/cpu0/cpuidle/state4/usage
  Output: >0 (CPU entering C6)

Step 4 - Measure power consumption:
  Result: ~130-135W (reduced from 147W)

7. Impact

• Tempest test failure: test_power_saving_tuned_profile fails because no power reduction is measured
• No power savings: Despite powersave configuration, actual power consumption unchanged
• Misleading configuration: System reports powersave mode but provides no actual power savings

8. Root Cause

The cpu-partitioning-powersave tuned profile configures the max_power_state parameter but does not write to the sysfs interface to actually enable C6 at the kernel/hardware level.

9. Proposed Fix

The tuned profile or deployment automation should enable C6 when edpm_tuned_max_power_cstate: "c6" is configured.

9.1 Option 1: Fix tuned profile

Add to /usr/lib/tuned/cpu-partitioning-powersave/script.sh:

  for cpu in /sys/devices/system/cpu/cpu*/cpuidle/state4/disable; do
      echo 0 > "$cpu"
  done

9.2 Option 2: Fix deployment automation

Add Ansible task when edpm_tuned_max_power_cstate: "c6":

  - name: Enable C6 cstate
    shell: |
      for cpu in /sys/devices/system/cpu/cpu*/cpuidle/state4/disable; do
        echo 0 > "$cpu"
      done
    when: edpm_tuned_max_power_cstate == "c6"

10. Additional Information

10.1 How to switch tuned profiles

Switch to baseline (performance):
  sudo tuned-adm profile cpu-partitioning

Switch to powersave:
  sudo tuned-adm profile cpu-partitioning-powersave

Verify active profile:
  tuned-adm active

10.2 Test case: test_power_saving_tuned_profile

The failing tempest test performs these steps:

1. Measure baseline power consumption with cpu-partitioning profile
2. Switch to cpu-partitioning-powersave profile
3. Measure powersave power consumption
4. Assert powersave < baseline (currently fails due to this bug)

The test obtains power measurements via:

• Redfish API: GET /redfish/v1/Chassis/System.Embedded.1/Power
• Field: PowerControl[0].PowerConsumedWatts
• Multiple samples averaged over measurement period