1 CPU cooling APIs How To
2 ===================================
4 Written by Amit Daniel Kachhap <amit.kachhap@linaro.org>
8 Copyright (c) 2012 Samsung Electronics Co., Ltd(http://www.samsung.com)
12 The generic cpu cooling(freq clipping) provides registration/unregistration APIs
13 to the caller. The binding of the cooling devices to the trip point is left for
14 the user. The registration APIs returns the cooling device pointer.
18 1.1 cpufreq registration/unregistration APIs
19 1.1.1 struct thermal_cooling_device *cpufreq_cooling_register(
20 struct cpumask *clip_cpus)
22 This interface function registers the cpufreq cooling device with the name
23 "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
26 clip_cpus: cpumask of cpus where the frequency constraints will happen.
28 1.1.2 struct thermal_cooling_device *of_cpufreq_cooling_register(
29 struct device_node *np, const struct cpumask *clip_cpus)
31 This interface function registers the cpufreq cooling device with
32 the name "thermal-cpufreq-%x" linking it with a device tree node, in
33 order to bind it via the thermal DT code. This api can support multiple
34 instances of cpufreq cooling devices.
36 np: pointer to the cooling device device tree node
37 clip_cpus: cpumask of cpus where the frequency constraints will happen.
39 1.1.3 struct thermal_cooling_device *cpufreq_power_cooling_register(
40 const struct cpumask *clip_cpus, u32 capacitance,
41 get_static_t plat_static_func)
43 Similar to cpufreq_cooling_register, this function registers a cpufreq
44 cooling device. Using this function, the cooling device will
45 implement the power extensions by using a simple cpu power model. The
46 cpus must have registered their OPPs using the OPP library.
48 The additional parameters are needed for the power model (See 2. Power
49 models). "capacitance" is the dynamic power coefficient (See 2.1
50 Dynamic power). "plat_static_func" is a function to calculate the
51 static power consumed by these cpus (See 2.2 Static power).
53 1.1.4 struct thermal_cooling_device *of_cpufreq_power_cooling_register(
54 struct device_node *np, const struct cpumask *clip_cpus, u32 capacitance,
55 get_static_t plat_static_func)
57 Similar to cpufreq_power_cooling_register, this function register a
58 cpufreq cooling device with power extensions using the device tree
59 information supplied by the np parameter.
61 1.1.5 void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev)
63 This interface function unregisters the "thermal-cpufreq-%x" cooling device.
65 cdev: Cooling device pointer which has to be unregistered.
69 The power API registration functions provide a simple power model for
70 CPUs. The current power is calculated as dynamic + (optionally)
71 static power. This power model requires that the operating-points of
72 the CPUs are registered using the kernel's opp library and the
73 `cpufreq_frequency_table` is assigned to the `struct device` of the
74 cpu. If you are using CONFIG_CPUFREQ_DT then the
75 `cpufreq_frequency_table` should already be assigned to the cpu
78 The `plat_static_func` parameter of `cpufreq_power_cooling_register()`
79 and `of_cpufreq_power_cooling_register()` is optional. If you don't
80 provide it, only dynamic power will be considered.
84 The dynamic power consumption of a processor depends on many factors.
85 For a given processor implementation the primary factors are:
87 - The time the processor spends running, consuming dynamic power, as
88 compared to the time in idle states where dynamic consumption is
89 negligible. Herein we refer to this as 'utilisation'.
90 - The voltage and frequency levels as a result of DVFS. The DVFS
91 level is a dominant factor governing power consumption.
92 - In running time the 'execution' behaviour (instruction types, memory
93 access patterns and so forth) causes, in most cases, a second order
94 variation. In pathological cases this variation can be significant,
95 but typically it is of a much lesser impact than the factors above.
97 A high level dynamic power consumption model may then be represented as:
99 Pdyn = f(run) * Voltage^2 * Frequency * Utilisation
101 f(run) here represents the described execution behaviour and its
102 result has a units of Watts/Hz/Volt^2 (this often expressed in
105 The detailed behaviour for f(run) could be modelled on-line. However,
106 in practice, such an on-line model has dependencies on a number of
107 implementation specific processor support and characterisation
108 factors. Therefore, in initial implementation that contribution is
109 represented as a constant coefficient. This is a simplification
110 consistent with the relative contribution to overall power variation.
112 In this simplified representation our model becomes:
114 Pdyn = Capacitance * Voltage^2 * Frequency * Utilisation
116 Where `capacitance` is a constant that represents an indicative
117 running time dynamic power coefficient in fundamental units of
118 mW/MHz/uVolt^2. Typical values for mobile CPUs might lie in range
119 from 100 to 500. For reference, the approximate values for the SoC in
120 ARM's Juno Development Platform are 530 for the Cortex-A57 cluster and
121 140 for the Cortex-A53 cluster.
126 Static leakage power consumption depends on a number of factors. For a
127 given circuit implementation the primary factors are:
129 - Time the circuit spends in each 'power state'
134 The time the circuit spends in each 'power state' for a given
135 evaluation period at first order means OFF or ON. However,
136 'retention' states can also be supported that reduce power during
137 inactive periods without loss of context.
139 Note: The visibility of state entries to the OS can vary, according to
140 platform specifics, and this can then impact the accuracy of a model
141 based on OS state information alone. It might be possible in some
142 cases to extract more accurate information from system resources.
144 The temperature, operating voltage and process 'grade' (slow to fast)
145 of the circuit are all significant factors in static leakage power
146 consumption. All of these have complex relationships to static power.
148 Circuit implementation specific factors include the chosen silicon
149 process as well as the type, number and size of transistors in both
150 the logic gates and any RAM elements included.
152 The static power consumption modelling must take into account the
153 power managed regions that are implemented. Taking the example of an
154 ARM processor cluster, the modelling would take into account whether
155 each CPU can be powered OFF separately or if only a single power
156 region is implemented for the complete cluster.
158 In one view, there are others, a static power consumption model can
159 then start from a set of reference values for each power managed
160 region (e.g. CPU, Cluster/L2) in each state (e.g. ON, OFF) at an
161 arbitrary process grade, voltage and temperature point. These values
162 are then scaled for all of the following: the time in each state, the
163 process grade, the current temperature and the operating voltage.
164 However, since both implementation specific and complex relationships
165 dominate the estimate, the appropriate interface to the model from the
166 cpu cooling device is to provide a function callback that calculates
167 the static power in this platform. When registering the cpu cooling
168 device pass a function pointer that follows the `get_static_t`
171 int plat_get_static(cpumask_t *cpumask, int interval,
172 unsigned long voltage, u32 &power);
174 `cpumask` is the cpumask of the cpus involved in the calculation.
175 `voltage` is the voltage at which they are operating. The function
176 should calculate the average static power for the last `interval`
177 milliseconds. It returns 0 on success, -E* on error. If it
178 succeeds, it should store the static power in `power`. Reading the
179 temperature of the cpus described by `cpumask` is left for
180 plat_get_static() to do as the platform knows best which thermal
181 sensor is closest to the cpu.
183 If `plat_static_func` is NULL, static power is considered to be
184 negligible for this platform and only dynamic power is considered.
186 The platform specific callback can then use any combination of tables
187 and/or equations to permute the estimated value. Process grade
188 information is not passed to the model since access to such data, from
189 on-chip measurement capability or manufacture time data, is platform
192 Note: the significance of static power for CPUs in comparison to
193 dynamic power is highly dependent on implementation. Given the
194 potential complexity in implementation, the importance and accuracy of
195 its inclusion when using cpu cooling devices should be assessed on a