Based on kernel version 4.10.8. Page generated on 2017-04-01 14:42 EST.
1 CPU frequency and voltage scaling code in the Linux(TM) kernel 2 3 4 L i n u x C P U F r e q 5 6 C P U F r e q G o v e r n o r s 7 8 - information for users and developers - 9 10 11 Dominik Brodowski <linux@brodo.de> 12 some additions and corrections by Nico Golde <nico@ngolde.de> 13 14 15 16 Clock scaling allows you to change the clock speed of the CPUs on the 17 fly. This is a nice method to save battery power, because the lower 18 the clock speed, the less power the CPU consumes. 19 20 21 Contents: 22 --------- 23 1. What is a CPUFreq Governor? 24 25 2. Governors In the Linux Kernel 26 2.1 Performance 27 2.2 Powersave 28 2.3 Userspace 29 2.4 Ondemand 30 2.5 Conservative 31 32 3. The Governor Interface in the CPUfreq Core 33 34 35 36 1. What Is A CPUFreq Governor? 37 ============================== 38 39 Most cpufreq drivers (except the intel_pstate and longrun) or even most 40 cpu frequency scaling algorithms only offer the CPU to be set to one 41 frequency. In order to offer dynamic frequency scaling, the cpufreq 42 core must be able to tell these drivers of a "target frequency". So 43 these specific drivers will be transformed to offer a "->target/target_index" 44 call instead of the existing "->setpolicy" call. For "longrun", all 45 stays the same, though. 46 47 How to decide what frequency within the CPUfreq policy should be used? 48 That's done using "cpufreq governors". Two are already in this patch 49 -- they're the already existing "powersave" and "performance" which 50 set the frequency statically to the lowest or highest frequency, 51 respectively. At least two more such governors will be ready for 52 addition in the near future, but likely many more as there are various 53 different theories and models about dynamic frequency scaling 54 around. Using such a generic interface as cpufreq offers to scaling 55 governors, these can be tested extensively, and the best one can be 56 selected for each specific use. 57 58 Basically, it's the following flow graph: 59 60 CPU can be set to switch independently | CPU can only be set 61 within specific "limits" | to specific frequencies 62 63 "CPUfreq policy" 64 consists of frequency limits (policy->{min,max}) 65 and CPUfreq governor to be used 66 / \ 67 / \ 68 / the cpufreq governor decides 69 / (dynamically or statically) 70 / what target_freq to set within 71 / the limits of policy->{min,max} 72 / \ 73 / \ 74 Using the ->setpolicy call, Using the ->target/target_index call, 75 the limits and the the frequency closest 76 "policy" is set. to target_freq is set. 77 It is assured that it 78 is within policy->{min,max} 79 80 81 2. Governors In the Linux Kernel 82 ================================ 83 84 2.1 Performance 85 --------------- 86 87 The CPUfreq governor "performance" sets the CPU statically to the 88 highest frequency within the borders of scaling_min_freq and 89 scaling_max_freq. 90 91 92 2.2 Powersave 93 ------------- 94 95 The CPUfreq governor "powersave" sets the CPU statically to the 96 lowest frequency within the borders of scaling_min_freq and 97 scaling_max_freq. 98 99 100 2.3 Userspace 101 ------------- 102 103 The CPUfreq governor "userspace" allows the user, or any userspace 104 program running with UID "root", to set the CPU to a specific frequency 105 by making a sysfs file "scaling_setspeed" available in the CPU-device 106 directory. 107 108 109 2.4 Ondemand 110 ------------ 111 112 The CPUfreq governor "ondemand" sets the CPU depending on the 113 current usage. To do this the CPU must have the capability to 114 switch the frequency very quickly. There are a number of sysfs file 115 accessible parameters: 116 117 sampling_rate: measured in uS (10^-6 seconds), this is how often you 118 want the kernel to look at the CPU usage and to make decisions on 119 what to do about the frequency. Typically this is set to values of 120 around '10000' or more. It's default value is (cmp. with users-guide.txt): 121 transition_latency * 1000 122 Be aware that transition latency is in ns and sampling_rate is in us, so you 123 get the same sysfs value by default. 124 Sampling rate should always get adjusted considering the transition latency 125 To set the sampling rate 750 times as high as the transition latency 126 in the bash (as said, 1000 is default), do: 127 echo `$(($(cat cpuinfo_transition_latency) * 750 / 1000)) \ 128 >ondemand/sampling_rate 129 130 sampling_rate_min: 131 The sampling rate is limited by the HW transition latency: 132 transition_latency * 100 133 Or by kernel restrictions: 134 If CONFIG_NO_HZ_COMMON is set, the limit is 10ms fixed. 135 If CONFIG_NO_HZ_COMMON is not set or nohz=off boot parameter is used, the 136 limits depend on the CONFIG_HZ option: 137 HZ=1000: min=20000us (20ms) 138 HZ=250: min=80000us (80ms) 139 HZ=100: min=200000us (200ms) 140 The highest value of kernel and HW latency restrictions is shown and 141 used as the minimum sampling rate. 142 143 up_threshold: defines what the average CPU usage between the samplings 144 of 'sampling_rate' needs to be for the kernel to make a decision on 145 whether it should increase the frequency. For example when it is set 146 to its default value of '95' it means that between the checking 147 intervals the CPU needs to be on average more than 95% in use to then 148 decide that the CPU frequency needs to be increased. 149 150 ignore_nice_load: this parameter takes a value of '0' or '1'. When 151 set to '0' (its default), all processes are counted towards the 152 'cpu utilisation' value. When set to '1', the processes that are 153 run with a 'nice' value will not count (and thus be ignored) in the 154 overall usage calculation. This is useful if you are running a CPU 155 intensive calculation on your laptop that you do not care how long it 156 takes to complete as you can 'nice' it and prevent it from taking part 157 in the deciding process of whether to increase your CPU frequency. 158 159 sampling_down_factor: this parameter controls the rate at which the 160 kernel makes a decision on when to decrease the frequency while running 161 at top speed. When set to 1 (the default) decisions to reevaluate load 162 are made at the same interval regardless of current clock speed. But 163 when set to greater than 1 (e.g. 100) it acts as a multiplier for the 164 scheduling interval for reevaluating load when the CPU is at its top 165 speed due to high load. This improves performance by reducing the overhead 166 of load evaluation and helping the CPU stay at its top speed when truly 167 busy, rather than shifting back and forth in speed. This tunable has no 168 effect on behavior at lower speeds/lower CPU loads. 169 170 powersave_bias: this parameter takes a value between 0 to 1000. It 171 defines the percentage (times 10) value of the target frequency that 172 will be shaved off of the target. For example, when set to 100 -- 10%, 173 when ondemand governor would have targeted 1000 MHz, it will target 174 1000 MHz - (10% of 1000 MHz) = 900 MHz instead. This is set to 0 175 (disabled) by default. 176 When AMD frequency sensitivity powersave bias driver -- 177 drivers/cpufreq/amd_freq_sensitivity.c is loaded, this parameter 178 defines the workload frequency sensitivity threshold in which a lower 179 frequency is chosen instead of ondemand governor's original target. 180 The frequency sensitivity is a hardware reported (on AMD Family 16h 181 Processors and above) value between 0 to 100% that tells software how 182 the performance of the workload running on a CPU will change when 183 frequency changes. A workload with sensitivity of 0% (memory/IO-bound) 184 will not perform any better on higher core frequency, whereas a 185 workload with sensitivity of 100% (CPU-bound) will perform better 186 higher the frequency. When the driver is loaded, this is set to 400 187 by default -- for CPUs running workloads with sensitivity value below 188 40%, a lower frequency is chosen. Unloading the driver or writing 0 189 will disable this feature. 190 191 192 2.5 Conservative 193 ---------------- 194 195 The CPUfreq governor "conservative", much like the "ondemand" 196 governor, sets the CPU depending on the current usage. It differs in 197 behaviour in that it gracefully increases and decreases the CPU speed 198 rather than jumping to max speed the moment there is any load on the 199 CPU. This behaviour more suitable in a battery powered environment. 200 The governor is tweaked in the same manner as the "ondemand" governor 201 through sysfs with the addition of: 202 203 freq_step: this describes what percentage steps the cpu freq should be 204 increased and decreased smoothly by. By default the cpu frequency will 205 increase in 5% chunks of your maximum cpu frequency. You can change this 206 value to anywhere between 0 and 100 where '0' will effectively lock your 207 CPU at a speed regardless of its load whilst '100' will, in theory, make 208 it behave identically to the "ondemand" governor. 209 210 down_threshold: same as the 'up_threshold' found for the "ondemand" 211 governor but for the opposite direction. For example when set to its 212 default value of '20' it means that if the CPU usage needs to be below 213 20% between samples to have the frequency decreased. 214 215 sampling_down_factor: similar functionality as in "ondemand" governor. 216 But in "conservative", it controls the rate at which the kernel makes 217 a decision on when to decrease the frequency while running in any 218 speed. Load for frequency increase is still evaluated every 219 sampling rate. 220 221 3. The Governor Interface in the CPUfreq Core 222 ============================================= 223 224 A new governor must register itself with the CPUfreq core using 225 "cpufreq_register_governor". The struct cpufreq_governor, which has to 226 be passed to that function, must contain the following values: 227 228 governor->name - A unique name for this governor 229 governor->governor - The governor callback function 230 governor->owner - .THIS_MODULE for the governor module (if 231 appropriate) 232 233 The governor->governor callback is called with the current (or to-be-set) 234 cpufreq_policy struct for that CPU, and an unsigned int event. The 235 following events are currently defined: 236 237 CPUFREQ_GOV_START: This governor shall start its duty for the CPU 238 policy->cpu 239 CPUFREQ_GOV_STOP: This governor shall end its duty for the CPU 240 policy->cpu 241 CPUFREQ_GOV_LIMITS: The limits for CPU policy->cpu have changed to 242 policy->min and policy->max. 243 244 If you need other "events" externally of your driver, _only_ use the 245 cpufreq_governor_l(unsigned int cpu, unsigned int event) call to the 246 CPUfreq core to ensure proper locking. 247 248 249 The CPUfreq governor may call the CPU processor driver using one of 250 these two functions: 251 252 int cpufreq_driver_target(struct cpufreq_policy *policy, 253 unsigned int target_freq, 254 unsigned int relation); 255 256 int __cpufreq_driver_target(struct cpufreq_policy *policy, 257 unsigned int target_freq, 258 unsigned int relation); 259 260 target_freq must be within policy->min and policy->max, of course. 261 What's the difference between these two functions? When your governor 262 still is in a direct code path of a call to governor->governor, the 263 per-CPU cpufreq lock is still held in the cpufreq core, and there's 264 no need to lock it again (in fact, this would cause a deadlock). So 265 use __cpufreq_driver_target only in these cases. In all other cases 266 (for example, when there's a "daemonized" function that wakes up 267 every second), use cpufreq_driver_target to lock the cpufreq per-CPU 268 lock before the command is passed to the cpufreq processor driver.