1 =================================
2 Power allocator governor tunables
3 =================================
8 The governor works optimally with the following two passive trip points:
10 1. "switch on" trip point: temperature above which the governor
11 control loop starts operating. This is the first passive trip
12 point of the thermal zone.
14 2. "desired temperature" trip point: it should be higher than the
15 "switch on" trip point. This the target temperature the governor
16 is controlling for. This is the last passive trip point of the
22 The power allocator governor implements a
23 Proportional-Integral-Derivative controller (PID controller) with
24 temperature as the control input and power as the controlled output:
26 P_max = k_p * e + k_i * err_integral + k_d * diff_err + sustainable_power
29 - e = desired_temperature - current_temperature
30 - err_integral is the sum of previous errors
31 - diff_err = e - previous_error
33 It is similar to the one depicted below::
40 | +----->| diff_err |-->| X |------+
41 | | +----------+ +---+ |
43 | | k_i | | get_requested_power()
47 +---+ | +-------+ +---+ +---+ +---+ +----------+
48 | S |-----+----->| sum e |----->| X |--->| S |-->| S |-->|power |
49 +---+ | +-------+ +---+ +---+ +---+ |allocation|
53 | +------->| X |-------------------+ v v
54 | +---+ granted performance
63 An estimate of the sustainable dissipatable power (in mW) should be
64 provided while registering the thermal zone. This estimates the
65 sustained power that can be dissipated at the desired control
66 temperature. This is the maximum sustained power for allocation at
67 the desired maximum temperature. The actual sustained power can vary
68 for a number of reasons. The closed loop controller will take care of
69 variations such as environmental conditions, and some factors related
70 to the speed-grade of the silicon. `sustainable_power` is therefore
71 simply an estimate, and may be tuned to affect the aggressiveness of
72 the thermal ramp. For reference, the sustainable power of a 4" phone
73 is typically 2000mW, while on a 10" tablet is around 4500mW (may vary
74 depending on screen size).
76 If you are using device tree, do add it as a property of the
77 thermal-zone. For example::
81 polling-delay = <1000>;
82 polling-delay-passive = <100>;
83 sustainable-power = <2500>;
86 Instead, if the thermal zone is registered from the platform code, pass a
87 `thermal_zone_params` that has a `sustainable_power`. If no
88 `thermal_zone_params` were being passed, then something like below
91 static const struct thermal_zone_params tz_params = {
92 .sustainable_power = 3500,
95 and then pass `tz_params` as the 5th parameter to
96 `thermal_zone_device_register()`
101 The implementation of the PID controller in the power allocator
102 thermal governor allows the configuration of two proportional term
103 constants: `k_po` and `k_pu`. `k_po` is the proportional term
104 constant during temperature overshoot periods (current temperature is
105 above "desired temperature" trip point). Conversely, `k_pu` is the
106 proportional term constant during temperature undershoot periods
107 (current temperature below "desired temperature" trip point).
109 These controls are intended as the primary mechanism for configuring
110 the permitted thermal "ramp" of the system. For instance, a lower
111 `k_pu` value will provide a slower ramp, at the cost of capping
112 available capacity at a low temperature. On the other hand, a high
113 value of `k_pu` will result in the governor granting very high power
114 while temperature is low, and may lead to temperature overshooting.
116 The default value for `k_pu` is::
118 2 * sustainable_power / (desired_temperature - switch_on_temp)
120 This means that at `switch_on_temp` the output of the controller's
121 proportional term will be 2 * `sustainable_power`. The default value
124 sustainable_power / (desired_temperature - switch_on_temp)
126 Focusing on the proportional and feed forward values of the PID
127 controller equation we have::
129 P_max = k_p * e + sustainable_power
131 The proportional term is proportional to the difference between the
132 desired temperature and the current one. When the current temperature
133 is the desired one, then the proportional component is zero and
134 `P_max` = `sustainable_power`. That is, the system should operate in
135 thermal equilibrium under constant load. `sustainable_power` is only
136 an estimate, which is the reason for closed-loop control such as this.
138 Expanding `k_pu` we get::
140 P_max = 2 * sustainable_power * (T_set - T) / (T_set - T_on) +
145 - T_set is the desired temperature
146 - T is the current temperature
147 - T_on is the switch on temperature
149 When the current temperature is the switch_on temperature, the above
152 P_max = 2 * sustainable_power * (T_set - T_on) / (T_set - T_on) +
153 sustainable_power = 2 * sustainable_power + sustainable_power =
154 3 * sustainable_power
156 Therefore, the proportional term alone linearly decreases power from
157 3 * `sustainable_power` to `sustainable_power` as the temperature
158 rises from the switch on temperature to the desired temperature.
160 k_i and integral_cutoff
161 -----------------------
163 `k_i` configures the PID loop's integral term constant. This term
164 allows the PID controller to compensate for long term drift and for
165 the quantized nature of the output control: cooling devices can't set
166 the exact power that the governor requests. When the temperature
167 error is below `integral_cutoff`, errors are accumulated in the
168 integral term. This term is then multiplied by `k_i` and the result
169 added to the output of the controller. Typically `k_i` is set low (1
170 or 2) and `integral_cutoff` is 0.
175 `k_d` configures the PID loop's derivative term constant. It's
176 recommended to leave it as the default: 0.
178 Cooling device power API
179 ========================
181 Cooling devices controlled by this governor must supply the additional
182 "power" API in their `cooling_device_ops`. It consists on three ops:
186 int get_requested_power(struct thermal_cooling_device *cdev,
187 struct thermal_zone_device *tz, u32 *power);
191 The `struct thermal_cooling_device` pointer
193 thermal zone in which we are currently operating
195 pointer in which to store the calculated power
197 `get_requested_power()` calculates the power requested by the device
198 in milliwatts and stores it in @power . It should return 0 on
199 success, -E* on failure. This is currently used by the power
200 allocator governor to calculate how much power to give to each cooling
205 int state2power(struct thermal_cooling_device *cdev, struct
206 thermal_zone_device *tz, unsigned long state,
210 The `struct thermal_cooling_device` pointer
212 thermal zone in which we are currently operating
214 A cooling device state
216 pointer in which to store the equivalent power
218 Convert cooling device state @state into power consumption in
219 milliwatts and store it in @power. It should return 0 on success, -E*
220 on failure. This is currently used by thermal core to calculate the
221 maximum power that an actor can consume.
225 int power2state(struct thermal_cooling_device *cdev, u32 power,
226 unsigned long *state);
229 The `struct thermal_cooling_device` pointer
233 pointer in which to store the resulting state
235 Calculate a cooling device state that would make the device consume at
236 most @power mW and store it in @state. It should return 0 on success,
237 -E* on failure. This is currently used by the thermal core to convert
238 a given power set by the power allocator governor to a state that the
239 cooling device can set. It is a function because this conversion may
240 depend on external factors that may change so this function should the
241 best conversion given "current circumstances".
243 Cooling device weights
244 ----------------------
246 Weights are a mechanism to bias the allocation among cooling
247 devices. They express the relative power efficiency of different
248 cooling devices. Higher weight can be used to express higher power
249 efficiency. Weighting is relative such that if each cooling device
250 has a weight of one they are considered equal. This is particularly
251 useful in heterogeneous systems where two cooling devices may perform
252 the same kind of compute, but with different efficiency. For example,
253 a system with two different types of processors.
255 If the thermal zone is registered using
256 `thermal_zone_device_register()` (i.e., platform code), then weights
257 are passed as part of the thermal zone's `thermal_bind_parameters`.
258 If the platform is registered using device tree, then they are passed
259 as the `contribution` property of each map in the `cooling-maps` node.
261 Limitations of the power allocator governor
262 ===========================================
264 The power allocator governor's PID controller works best if there is a
265 periodic tick. If you have a driver that calls
266 `thermal_zone_device_update()` (or anything that ends up calling the
267 governor's `throttle()` function) repetitively, the governor response
268 won't be very good. Note that this is not particular to this
269 governor, step-wise will also misbehave if you call its throttle()
270 faster than the normal thermal framework tick (due to interrupts for
271 example) as it will overreact.