| 1 |
======================= |
| 2 |
CPU cooling APIs How To |
| 3 |
======================= |
| 4 |
|
| 5 |
Written by Amit Daniel Kachhap <[email protected]> |
| 6 |
|
| 7 |
Updated: 6 Jan 2015 |
| 8 |
|
| 9 |
Copyright (c) 2012 Samsung Electronics Co., Ltd(http://www.samsung.com) |
| 10 |
|
| 11 |
0. Introduction |
| 12 |
=============== |
| 13 |
|
| 14 |
The generic cpu cooling(freq clipping) provides registration/unregistration APIs |
| 15 |
to the caller. The binding of the cooling devices to the trip point is left for |
| 16 |
the user. The registration APIs returns the cooling device pointer. |
| 17 |
|
| 18 |
1. cpu cooling APIs |
| 19 |
=================== |
| 20 |
|
| 21 |
1.1 cpufreq registration/unregistration APIs |
| 22 |
-------------------------------------------- |
| 23 |
|
| 24 |
:: |
| 25 |
|
| 26 |
struct thermal_cooling_device |
| 27 |
*cpufreq_cooling_register(struct cpumask *clip_cpus) |
| 28 |
|
| 29 |
This interface function registers the cpufreq cooling device with the name |
| 30 |
"thermal-cpufreq-%x". This api can support multiple instances of cpufreq |
| 31 |
cooling devices. |
| 32 |
|
| 33 |
clip_cpus: |
| 34 |
cpumask of cpus where the frequency constraints will happen. |
| 35 |
|
| 36 |
:: |
| 37 |
|
| 38 |
struct thermal_cooling_device |
| 39 |
*of_cpufreq_cooling_register(struct cpufreq_policy *policy) |
| 40 |
|
| 41 |
This interface function registers the cpufreq cooling device with |
| 42 |
the name "thermal-cpufreq-%x" linking it with a device tree node, in |
| 43 |
order to bind it via the thermal DT code. This api can support multiple |
| 44 |
instances of cpufreq cooling devices. |
| 45 |
|
| 46 |
policy: |
| 47 |
CPUFreq policy. |
| 48 |
|
| 49 |
|
| 50 |
:: |
| 51 |
|
| 52 |
void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev) |
| 53 |
|
| 54 |
This interface function unregisters the "thermal-cpufreq-%x" cooling device. |
| 55 |
|
| 56 |
cdev: Cooling device pointer which has to be unregistered. |
| 57 |
|
| 58 |
2. Power models |
| 59 |
=============== |
| 60 |
|
| 61 |
The power API registration functions provide a simple power model for |
| 62 |
CPUs. The current power is calculated as dynamic power (static power isn't |
| 63 |
supported currently). This power model requires that the operating-points of |
| 64 |
the CPUs are registered using the kernel's opp library and the |
| 65 |
`cpufreq_frequency_table` is assigned to the `struct device` of the |
| 66 |
cpu. If you are using CONFIG_CPUFREQ_DT then the |
| 67 |
`cpufreq_frequency_table` should already be assigned to the cpu |
| 68 |
device. |
| 69 |
|
| 70 |
The dynamic power consumption of a processor depends on many factors. |
| 71 |
For a given processor implementation the primary factors are: |
| 72 |
|
| 73 |
- The time the processor spends running, consuming dynamic power, as |
| 74 |
compared to the time in idle states where dynamic consumption is |
| 75 |
negligible. Herein we refer to this as 'utilisation'. |
| 76 |
- The voltage and frequency levels as a result of DVFS. The DVFS |
| 77 |
level is a dominant factor governing power consumption. |
| 78 |
- In running time the 'execution' behaviour (instruction types, memory |
| 79 |
access patterns and so forth) causes, in most cases, a second order |
| 80 |
variation. In pathological cases this variation can be significant, |
| 81 |
but typically it is of a much lesser impact than the factors above. |
| 82 |
|
| 83 |
A high level dynamic power consumption model may then be represented as:: |
| 84 |
|
| 85 |
Pdyn = f(run) * Voltage^2 * Frequency * Utilisation |
| 86 |
|
| 87 |
f(run) here represents the described execution behaviour and its |
| 88 |
result has a units of Watts/Hz/Volt^2 (this often expressed in |
| 89 |
mW/MHz/uVolt^2) |
| 90 |
|
| 91 |
The detailed behaviour for f(run) could be modelled on-line. However, |
| 92 |
in practice, such an on-line model has dependencies on a number of |
| 93 |
implementation specific processor support and characterisation |
| 94 |
factors. Therefore, in initial implementation that contribution is |
| 95 |
represented as a constant coefficient. This is a simplification |
| 96 |
consistent with the relative contribution to overall power variation. |
| 97 |
|
| 98 |
In this simplified representation our model becomes:: |
| 99 |
|
| 100 |
Pdyn = Capacitance * Voltage^2 * Frequency * Utilisation |
| 101 |
|
| 102 |
Where `capacitance` is a constant that represents an indicative |
| 103 |
running time dynamic power coefficient in fundamental units of |
| 104 |
mW/MHz/uVolt^2. Typical values for mobile CPUs might lie in range |
| 105 |
from 100 to 500. For reference, the approximate values for the SoC in |
| 106 |
ARM's Juno Development Platform are 530 for the Cortex-A57 cluster and |
| 107 |
140 for the Cortex-A53 cluster. |