dvfsmgr-*

Dynamic Voltage Frequency Scaling driver

Syntax:

dvfsmgr [vwp:i:c:]

Runs on:

QNX Neutrino

Options:

Note: The -c and -v options can be used by all users. The other options should be used only by advanced users (with guidance from QNX).

c cfg_file: The path to the DVFS configuration file. The default is /etc/system/config/dvfs.conf.
-i interval: The CPU accounting interval, in milliseconds (default: 1000ms).
-p prio: The CPU accounting priority (default:51).
-v: Be verbose. Additional v options cause more verbosity.
-w: Enable the wfi workaround (default: disabled).

Description:

The dvfsmgr-* driver manages Dynamic Voltage Frequency Scaling (DVFS) for a system on a chip (Soc). DVFS allows SoCs to run at various power levels based on CPU load and/or temperature. This system-level power management could be beneficial in saving power during low CPU loads and ensuring that the chip’s thermal limits are never reached.

This driver provides some level of power and thermal management from a system level. The driver is responsible for the following tasks:

monitoring the CPU load (MPU load)
monitoring the SoC temperature (if possible)
scaling the voltage and the frequency of MPU, based on the reported CPU load and/or SoC temperature
providing an API for applications to send commands or read various information

Note: Driver-level power management, CPU core scaling, thread migration, non-MPU voltage/frequency monitoring/scaling and off-chip cooling device control are out of scope of this driver.

The dvfs_client utility provides a way for you to interact with the DVFS driver.

DVFS configuration file

You need to provide a DVFS configuration file to the driver. This file has the following columns, separated by either spaces or tabs:

Level number: The level number (0-based) of each row within the configuration file.
Up threshold: The “UP” or higher power-transition threshold. If the CPU load exceeds the value for a given level, a transition occurs (if allowed).
Down threshold: The “DOWN” or lower power transition threshold. If the CPU load drops below this value for a given level, a transition occurs.
Up temperature threshold: The temperature (in degrees Celsius) at which a certain power level is disabled and a transition to a lower level occurs (if allowed).
Down temperature threshold: The temperature (in degree Celsius) at which a certain power level is reenabled. A transition may or may not occur (if allowed, and depending on CPU load).

The DVFS driver parses the configuration file on startup and extracts all relevant information from it. A line starting with ‘#’ is considered a comment. The driver reports any errors encountered while parsing through this file. A DVFS configuration file with 3 levels of transition is shown below:

#DVFS Config file
# Level   UP_threshold   DOWN_threshold    Up_temperature   Down_temperature
    0        100            40                80              78
    1         80            20                90              88
    2         60             0                95              93

A sample configuration file is included with the driver. You can modify this file to meet your specific needs.

Note: Don't change the number of levels in the configuration file. This file gets mapped to an internal power table that's maintained by the driver, and adding or removing levels might result in unexpected behavior. For example, if the provided sample configuration file contains five levels, the modified file must also contain five levels.

Operating modes

The DVFS driver supports the following operating modes, to ensure maximum flexibility:

Automatic mode

This is the default mode of the driver. The driver automatically adjusts voltage and frequency, based on the CPU load and temperature. Multiple clients can obtain the current status of the driver, which contains information such as load, temperature, mode of operation and so on. In this mode, clients can't change the power levels, and such requests are ignored.

This mode, however, allows an application to request a minimum level of operation, below which the driver shouldn't operate (unless forced by high temperatures). The driver is in charge of notifying the clients if any change to their desired setting is made.

Semi-automatic mode

Semi-automatic mode can be activated by only one client at any given time.

The driver may be changed to this mode when the appropriate request is made from a client. When a client successfully changes the mode to semi-automatic, it's in charge of requesting power level change, based on the status of the driver. No other client is allowed to change the mode and/or power level of the driver. The driver decreases power levels due only to temperature restrictions.

The controlling client may relinquish control by setting the mode of the driver back to automatic mode. Once the driver is back in automatic mode, any client may change the mode to manual or semi-automatic.

Manual mode

Manual mode can be activated by only one client at any given time. This mode is similar to semi-automatic, with the exception that the driver doesn't modify power levels under any circumstances, and the controlling client has full control over the driver.

Reading the driver status

To get the driver's status, read from /dev/dvfs. The DVFS driver returns the status of driver, along with all significant information regarding the driver. Here's an example:

# cat /dev/dvfs
----------------------------
   Core Temp = 42
   CPU<0> Load = 91
   CPU<1> Load = 68
   Mode = Auto (0)
----------------------------
  #  |  App  |  Thermal
----------------------------
  0  |   -   |    x
 *1  |   -   |    -
  2  |   -   |    -
----------------------------

Note: The information that the driver returns depends on the platform.

In this example, the temperature of the core is 42°C, CPU loads for cores 0 and 1 are 91% and 68%, and the driver is running in automatic mode. Furthermore, three power levels are supported. Power level 0 (the highest level) is disabled by temperature, and the driver is running at power level 1 (shown with an asterisk).