5 Under certain circumstances a SoC can reach a critical temperature
6 limit and is unable to stabilize the temperature around a temperature
7 control. When the SoC has to stabilize the temperature, the kernel can
8 act on a cooling device to mitigate the dissipated power. When the
9 critical temperature is reached, a decision must be taken to reduce
10 the temperature, that, in turn impacts performance.
12 Another situation is when the silicon temperature continues to
13 increase even after the dynamic leakage is reduced to its minimum by
14 clock gating the component. This runaway phenomenon can continue due
15 to the static leakage. The only solution is to power down the
16 component, thus dropping the dynamic and static leakage that will
17 allow the component to cool down.
19 Last but not least, the system can ask for a specific power budget but
20 because of the OPP density, we can only choose an OPP with a power
21 budget lower than the requested one and under-utilize the CPU, thus
22 losing performance. In other words, one OPP under-utilizes the CPU
23 with a power less than the requested power budget and the next OPP
24 exceeds the power budget. An intermediate OPP could have been used if
30 If we can remove the static and the dynamic leakage for a specific
31 duration in a controlled period, the SoC temperature will
32 decrease. Acting on the idle state duration or the idle cycle
33 injection period, we can mitigate the temperature by modulating the
36 The Operating Performance Point (OPP) density has a great influence on
37 the control precision of cpufreq, however different vendors have a
38 plethora of OPP density, and some have large power gap between OPPs,
39 that will result in loss of performance during thermal control and
40 loss of power in other scenarios.
42 At a specific OPP, we can assume that injecting idle cycle on all CPUs
43 belong to the same cluster, with a duration greater than the cluster
44 idle state target residency, we lead to dropping the static and the
45 dynamic leakage for this period (modulo the energy needed to enter
46 this state). So the sustainable power with idle cycles has a linear
47 relation with the OPP’s sustainable power and can be computed with a
48 coefficient similar to:
50 Power(IdleCycle) = Coef x Power(OPP)
55 The base concept of the idle injection is to force the CPU to go to an
56 idle state for a specified time each control cycle, it provides
57 another way to control CPU power and heat in addition to
58 cpufreq. Ideally, if all CPUs belonging to the same cluster, inject
59 their idle cycles synchronously, the cluster can reach its power down
60 state with a minimum power consumption and reduce the static leakage
61 to almost zero. However, these idle cycles injection will add extra
62 latencies as the CPUs will have to wakeup from a deep sleep state.
64 We use a fixed duration of idle injection that gives an acceptable
65 performance penalty and a fixed latency. Mitigation can be increased
66 or decreased by modulating the duty cycle of the idle injection.
74 |_______|_______________________|_______|___________
77 idle <---------------------->
80 <----------------------------->
84 The implementation of the cooling device bases the number of states on
85 the duty cycle percentage. When no mitigation is happening the cooling
86 device state is zero, meaning the duty cycle is 0%.
88 When the mitigation begins, depending on the governor's policy, a
89 starting state is selected. With a fixed idle duration and the duty
90 cycle (aka the cooling device state), the running duration can be
93 The governor will change the cooling device state thus the duty cycle
94 and this variation will modulate the cooling effect.
102 |_______|_______________|_______|___________
105 idle <-------------->
108 <--------------------->
116 |_______|_______|_______|___________
125 The idle injection duration value must comply with the constraints:
127 - It is less than or equal to the latency we tolerate when the
128 mitigation begins. It is platform dependent and will depend on the
129 user experience, reactivity vs performance trade off we want. This
130 value should be specified.
132 - It is greater than the idle state’s target residency we want to go
133 for thermal mitigation, otherwise we end up consuming more energy.
138 When we reach the thermal trip point, we have to sustain a specified
139 power for a specific temperature but at this time we consume:
141 Power = Capacitance x Voltage^2 x Frequency x Utilisation
143 ... which is more than the sustainable power (or there is something
144 wrong in the system setup). The ‘Capacitance’ and ‘Utilisation’ are a
145 fixed value, ‘Voltage’ and the ‘Frequency’ are fixed artificially
146 because we don’t want to change the OPP. We can group the
147 ‘Capacitance’ and the ‘Utilisation’ into a single term which is the
148 ‘Dynamic Power Coefficient (Cdyn)’ Simplifying the above, we have:
150 Pdyn = Cdyn x Voltage^2 x Frequency
152 The power allocator governor will ask us somehow to reduce our power
153 in order to target the sustainable power defined in the device
154 tree. So with the idle injection mechanism, we want an average power
155 (Ptarget) resulting in an amount of time running at full power on a
156 specific OPP and idle another amount of time. That could be put in a
159 P(opp)target = ((Trunning x (P(opp)running) + (Tidle x P(opp)idle)) /
164 Tidle = Trunning x ((P(opp)running / P(opp)target) - 1)
166 At this point if we know the running period for the CPU, that gives us
167 the idle injection we need. Alternatively if we have the idle
168 injection duration, we can compute the running duration with:
170 Trunning = Tidle / ((P(opp)running / P(opp)target) - 1)
172 Practically, if the running power is less than the targeted power, we
173 end up with a negative time value, so obviously the equation usage is
174 bound to a power reduction, hence a higher OPP is needed to have the
175 running power greater than the targeted power.
177 However, in this demonstration we ignore three aspects:
179 * The static leakage is not defined here, we can introduce it in the
180 equation but assuming it will be zero most of the time as it is
181 difficult to get the values from the SoC vendors
183 * The idle state wake up latency (or entry + exit latency) is not
184 taken into account, it must be added in the equation in order to
185 rigorously compute the idle injection
187 * The injected idle duration must be greater than the idle state
188 target residency, otherwise we end up consuming more energy and
189 potentially invert the mitigation effect
191 So the final equation is:
193 Trunning = (Tidle - Twakeup ) x
194 (((P(opp)dyn + P(opp)static ) - P(opp)target) / P(opp)target )