1 .. SPDX-License-Identifier: GPL-2.0
2
3 ================
4 CPU Idle Cooling
5 ================
6
7 Situation:
8 ----------
9
10 Under certain circumstances a SoC can reach a
11 limit and is unable to stabilize the temperatu
12 control. When the SoC has to stabilize the tem
13 act on a cooling device to mitigate the dissip
14 critical temperature is reached, a decision mu
15 the temperature, that, in turn impacts perform
16
17 Another situation is when the silicon temperat
18 increase even after the dynamic leakage is red
19 clock gating the component. This runaway pheno
20 to the static leakage. The only solution is to
21 component, thus dropping the dynamic and stati
22 allow the component to cool down.
23
24 Last but not least, the system can ask for a s
25 because of the OPP density, we can only choose
26 budget lower than the requested one and under-
27 losing performance. In other words, one OPP un
28 with a power less than the requested power bud
29 exceeds the power budget. An intermediate OPP
30 it were present.
31
32 Solutions:
33 ----------
34
35 If we can remove the static and the dynamic le
36 duration in a controlled period, the SoC tempe
37 decrease. Acting on the idle state duration or
38 injection period, we can mitigate the temperat
39 power budget.
40
41 The Operating Performance Point (OPP) density
42 the control precision of cpufreq, however diff
43 plethora of OPP density, and some have large p
44 that will result in loss of performance during
45 loss of power in other scenarios.
46
47 At a specific OPP, we can assume that injectin
48 belong to the same cluster, with a duration gr
49 idle state target residency, we lead to droppi
50 dynamic leakage for this period (modulo the en
51 this state). So the sustainable power with idl
52 relation with the OPP’s sustainable power an
53 coefficient similar to::
54
55 Power(IdleCycle) = Coef x Power(OP
56
57 Idle Injection:
58 ---------------
59
60 The base concept of the idle injection is to f
61 idle state for a specified time each control c
62 another way to control CPU power and heat in a
63 cpufreq. Ideally, if all CPUs belonging to the
64 their idle cycles synchronously, the cluster c
65 state with a minimum power consumption and red
66 to almost zero. However, these idle cycles in
67 latencies as the CPUs will have to wakeup from
68
69 We use a fixed duration of idle injection that
70 performance penalty and a fixed latency. Mitig
71 or decreased by modulating the duty cycle of t
72
73 ::
74
75 ^
76 |
77 |
78 |------- -------
79 |_______|_______________________|_______|
80
81 <------>
82 idle <---------------------->
83 running
84
85 <----------------------------->
86 duty cycle 25%
87
88
89 The implementation of the cooling device bases
90 the duty cycle percentage. When no mitigation
91 device state is zero, meaning the duty cycle i
92
93 When the mitigation begins, depending on the g
94 starting state is selected. With a fixed idle
95 cycle (aka the cooling device state), the runn
96 computed.
97
98 The governor will change the cooling device st
99 and this variation will modulate the cooling e
100
101 ::
102
103 ^
104 |
105 |
106 |------- -------
107 |_______|_______________|_______|________
108
109 <------>
110 idle <-------------->
111 running
112
113 <--------------------->
114 duty cycle 33%
115
116
117 ^
118 |
119 |
120 |------- -------
121 |_______|_______|_______|___________
122
123 <------>
124 idle <------>
125 running
126
127 <------------->
128 duty cycle 50%
129
130 The idle injection duration value must comply
131
132 - It is less than or equal to the latency we t
133 mitigation begins. It is platform dependent
134 user experience, reactivity vs performance t
135 value should be specified.
136
137 - It is greater than the idle state’s target
138 for thermal mitigation, otherwise we end up
139
140 Power considerations
141 --------------------
142
143 When we reach the thermal trip point, we have
144 power for a specific temperature but at this t
145
146 Power = Capacitance x Voltage^2 x Frequency x
147
148 ... which is more than the sustainable power (
149 wrong in the system setup). The ‘Capacitance
150 fixed value, ‘Voltage’ and the ‘Frequenc
151 because we don’t want to change the OPP. We
152 ‘Capacitance’ and the ‘Utilisation’ in
153 ‘Dynamic Power Coefficient (Cdyn)’ Simplif
154
155 Pdyn = Cdyn x Voltage^2 x Frequency
156
157 The power allocator governor will ask us someh
158 in order to target the sustainable power defin
159 tree. So with the idle injection mechanism, we
160 (Ptarget) resulting in an amount of time runni
161 specific OPP and idle another amount of time.
162 equation::
163
164 P(opp)target = ((Trunning x (P(opp)running) +
165 (Trunning + Tidle)
166
167 ...
168
169 Tidle = Trunning x ((P(opp)running / P(opp)ta
170
171 At this point if we know the running period fo
172 the idle injection we need. Alternatively if w
173 injection duration, we can compute the running
174
175 Trunning = Tidle / ((P(opp)running / P(opp)ta
176
177 Practically, if the running power is less than
178 end up with a negative time value, so obviousl
179 bound to a power reduction, hence a higher OPP
180 running power greater than the targeted power.
181
182 However, in this demonstration we ignore three
183
184 * The static leakage is not defined here, we
185 equation but assuming it will be zero most
186 difficult to get the values from the SoC ve
187
188 * The idle state wake up latency (or entry +
189 taken into account, it must be added in the
190 rigorously compute the idle injection
191
192 * The injected idle duration must be greater
193 target residency, otherwise we end up consu
194 potentially invert the mitigation effect
195
196 So the final equation is::
197
198 Trunning = (Tidle - Twakeup ) x
199 (((P(opp)dyn + P(opp)static )
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.