Disable Power Features#
For better robustness and freedom from interference by bad or malicious code, it is recommended to not enter lower power states such as clock-gating and power-gating, and CPU and Device frequency scaling.
Linux Kernel runs as part of Guest OS VM and is enabled with various power management features.
Following power management features should be disabled to better comply with Safety goals:
No. |
Power Feature |
Impact and Method to Disable this Feature |
|---|---|---|
1 |
CPU Frequency Scaling |
CPU frequency shall remain at a fixed value set at boot time, typically at maximum frequency value defined for that power profile. This can be achieved by enabling PERFORMANCE CPU frequency governor by default and disabling other CPU frequency governors from Kernel Config at Compile time. |
2 |
Device Frequency Scaling |
Device frequency shall remain at a fixed value set at boot time, typically at maximum frequency value defined for that power profile. This can be achieved by enabling PERFORMANCE Device frequency governor by default and disabling other Device frequency governors from Kernel Config at Compile time. |
3 |
Runtime Power Management |
Devices shall always remain in ON state and shall not any enter low power state such clock-gating and power-gating. This can be achieved by running a script just after Guest OS boot is complete, that iterates through all platform devices |
Note
CPU Idle should remain enabled as CPU enters WFI state when CPU is idle. Moreover, disabling this feature will impact System Suspend state (SC7).
Power Impact of disabling above 3 features in Idle state after Device Boot is 10%, while the impact in some of the Runtime use-cases involving Camera capture, Tensor RT and Encode is about 7% as measured on DRIVEOS 6.0 Linux running on NVIDIA DRIVE AGX Developer Platform. We expect similar power impact on DRIVEOS 7.0 Linux running on NVIDIA DRIVE AGX Thor Developer Platform.
These power features are disabled by default in production builds.