1 CPU frequency and voltage scaling code in the Linux(TM) kernel
4 L i n u x C P U F r e q
6 C P U F r e q G o v e r n o r s
8 - information for users and developers -
11 Dominik Brodowski <linux@brodo.de>
12 some additions and corrections by Nico Golde <nico@ngolde.de>
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.
23 1. What is a CPUFreq Governor?
25 2. Governors In the Linux Kernel
33 3. The Governor Interface in the CPUfreq Core
37 1. What Is A CPUFreq Governor?
38 ==============================
40 Most cpufreq drivers (in fact, all except one, longrun) or even most
41 cpu frequency scaling algorithms only offer the CPU to be set to one
42 frequency. In order to offer dynamic frequency scaling, the cpufreq
43 core must be able to tell these drivers of a "target frequency". So
44 these specific drivers will be transformed to offer a "->target"
45 call instead of the existing "->setpolicy" call. For "longrun", all
46 stays the same, though.
48 How to decide what frequency within the CPUfreq policy should be used?
49 That's done using "cpufreq governors". Two are already in this patch
50 -- they're the already existing "powersave" and "performance" which
51 set the frequency statically to the lowest or highest frequency,
52 respectively. At least two more such governors will be ready for
53 addition in the near future, but likely many more as there are various
54 different theories and models about dynamic frequency scaling
55 around. Using such a generic interface as cpufreq offers to scaling
56 governors, these can be tested extensively, and the best one can be
57 selected for each specific use.
59 Basically, it's the following flow graph:
61 CPU can be set to switch independently | CPU can only be set
62 within specific "limits" | to specific frequencies
65 consists of frequency limits (policy->{min,max})
66 and CPUfreq governor to be used
69 / the cpufreq governor decides
70 / (dynamically or statically)
71 / what target_freq to set within
72 / the limits of policy->{min,max}
75 Using the ->setpolicy call, Using the ->target call,
76 the limits and the the frequency closest
77 "policy" is set. to target_freq is set.
79 is within policy->{min,max}
82 2. Governors In the Linux Kernel
83 ================================
88 The CPUfreq governor "performance" sets the CPU statically to the
89 highest frequency within the borders of scaling_min_freq and
96 The CPUfreq governor "powersave" sets the CPU statically to the
97 lowest frequency within the borders of scaling_min_freq and
104 The CPUfreq governor "userspace" allows the user, or any userspace
105 program running with UID "root", to set the CPU to a specific frequency
106 by making a sysfs file "scaling_setspeed" available in the CPU-device
113 The CPUfreq governor "ondemand" sets the CPU depending on the
114 current usage. To do this the CPU must have the capability to
115 switch the frequency very quickly. There are a number of sysfs file
116 accessible parameters:
118 sampling_rate: measured in uS (10^-6 seconds), this is how often you
119 want the kernel to look at the CPU usage and to make decisions on
120 what to do about the frequency. Typically this is set to values of
121 around '10000' or more. It's default value is (cmp. with users-guide.txt):
122 transition_latency * 1000
123 Be aware that transition latency is in ns and sampling_rate is in us, so you
124 get the same sysfs value by default.
125 Sampling rate should always get adjusted considering the transition latency
126 To set the sampling rate 750 times as high as the transition latency
127 in the bash (as said, 1000 is default), do:
128 echo `$(($(cat cpuinfo_transition_latency) * 750 / 1000)) \
129 >ondemand/sampling_rate
131 show_sampling_rate_min:
132 The sampling rate is limited by the HW transition latency:
133 transition_latency * 100
134 Or by kernel restrictions:
135 If CONFIG_NO_HZ is set, the limit is 10ms fixed.
136 If CONFIG_NO_HZ is not set or no_hz=off boot parameter is used, the
137 limits depend on the CONFIG_HZ option:
138 HZ=1000: min=20000us (20ms)
139 HZ=250: min=80000us (80ms)
140 HZ=100: min=200000us (200ms)
141 The highest value of kernel and HW latency restrictions is shown and
142 used as the minimum sampling rate.
144 show_sampling_rate_max: THIS INTERFACE IS DEPRECATED, DON'T USE IT.
146 up_threshold: defines what the average CPU usage between the samplings
147 of 'sampling_rate' needs to be for the kernel to make a decision on
148 whether it should increase the frequency. For example when it is set
149 to its default value of '95' it means that between the checking
150 intervals the CPU needs to be on average more than 95% in use to then
151 decide that the CPU frequency needs to be increased.
153 ignore_nice_load: this parameter takes a value of '0' or '1'. When
154 set to '0' (its default), all processes are counted towards the
155 'cpu utilisation' value. When set to '1', the processes that are
156 run with a 'nice' value will not count (and thus be ignored) in the
157 overall usage calculation. This is useful if you are running a CPU
158 intensive calculation on your laptop that you do not care how long it
159 takes to complete as you can 'nice' it and prevent it from taking part
160 in the deciding process of whether to increase your CPU frequency.
162 sampling_down_factor: this parameter controls the rate at which the
163 kernel makes a decision on when to decrease the frequency while running
164 at top speed. When set to 1 (the default) decisions to reevaluate load
165 are made at the same interval regardless of current clock speed. But
166 when set to greater than 1 (e.g. 100) it acts as a multiplier for the
167 scheduling interval for reevaluating load when the CPU is at its top
168 speed due to high load. This improves performance by reducing the overhead
169 of load evaluation and helping the CPU stay at its top speed when truly
170 busy, rather than shifting back and forth in speed. This tunable has no
171 effect on behavior at lower speeds/lower CPU loads.
177 The CPUfreq governor "conservative", much like the "ondemand"
178 governor, sets the CPU depending on the current usage. It differs in
179 behaviour in that it gracefully increases and decreases the CPU speed
180 rather than jumping to max speed the moment there is any load on the
181 CPU. This behaviour more suitable in a battery powered environment.
182 The governor is tweaked in the same manner as the "ondemand" governor
183 through sysfs with the addition of:
185 freq_step: this describes what percentage steps the cpu freq should be
186 increased and decreased smoothly by. By default the cpu frequency will
187 increase in 5% chunks of your maximum cpu frequency. You can change this
188 value to anywhere between 0 and 100 where '0' will effectively lock your
189 CPU at a speed regardless of its load whilst '100' will, in theory, make
190 it behave identically to the "ondemand" governor.
192 down_threshold: same as the 'up_threshold' found for the "ondemand"
193 governor but for the opposite direction. For example when set to its
194 default value of '20' it means that if the CPU usage needs to be below
195 20% between samples to have the frequency decreased.
201 The CPUfreq governor "interactive" is designed for latency-sensitive,
202 interactive workloads. This governor sets the CPU speed depending on
203 usage, similar to "ondemand" and "conservative" governors. However,
204 the governor is more aggressive about scaling the CPU speed up in
205 response to CPU-intensive activity.
207 Sampling the CPU load every X ms can lead to under-powering the CPU
208 for X ms, leading to dropped frames, stuttering UI, etc. Instead of
209 sampling the cpu at a specified rate, the interactive governor will
210 check whether to scale the cpu frequency up soon after coming out of
211 idle. When the cpu comes out of idle, a timer is configured to fire
212 within 1-2 ticks. If the cpu is very busy between exiting idle and
213 when the timer fires then we assume the cpu is underpowered and ramp
216 If the cpu was not sufficiently busy to immediately ramp to MAX speed,
217 then governor evaluates the cpu load since the last speed adjustment,
218 choosing the highest value between that longer-term load or the
219 short-term load since idle exit to determine the cpu speed to ramp to.
221 The tuneable values for this governor are:
223 min_sample_time: The minimum amount of time to spend at the current
224 frequency before ramping down. This is to ensure that the governor has
225 seen enough historic cpu load data to determine the appropriate
226 workload. Default is 80000 uS.
228 go_maxspeed_load: The CPU load at which to ramp to max speed. Default
231 timer_rate: Sample rate for reevaluating cpu load when the system is
232 not idle. Default is 30000 uS.
237 The CPUfreq governor "hotplug" operates similary to "ondemand" and
238 "conservative". It's decisions are based primarily on CPU load. Like
239 "ondemand" the "hotplug" governor will ramp up to the highest frequency
240 once the run-time tunable "up_threshold" parameter is crossed. Like
241 "conservative", the "hotplug" governor exports a "down_threshold"
242 parameter that is also tunable at run-time. When the "down_threshold"
243 is crossed the CPU transitions to the next lowest frequency in the
244 CPUfreq frequency table instead of decrementing the frequency based on a
245 percentage of maximum load.
247 The main reason "hotplug" governor exists is for architectures requiring
248 that only the master CPU be online in order to hit low-power states
249 (C-states). OMAP4 is one such example of this. The "hotplug" governor
250 is also helpful in reducing thermal output in devices with tight thermal
253 Auxillary CPUs are onlined/offline based on CPU load, but the decision
254 to do so is made after averaging several sampling windows. This is to
255 reduce CPU hotplug "thrashing", which can be caused by normal system
256 entropy and leads to lots of spurious plug-in and plug-out transitions.
257 The number of sampling periods averaged together is tunable via the
258 "hotplug_in_sampling_periods" and "hotplug_out_sampling_periods"
259 run-time tunable parameters.
261 3. The Governor Interface in the CPUfreq Core
262 =============================================
264 A new governor must register itself with the CPUfreq core using
265 "cpufreq_register_governor". The struct cpufreq_governor, which has to
266 be passed to that function, must contain the following values:
268 governor->name - A unique name for this governor
269 governor->governor - The governor callback function
270 governor->owner - .THIS_MODULE for the governor module (if
273 The governor->governor callback is called with the current (or to-be-set)
274 cpufreq_policy struct for that CPU, and an unsigned int event. The
275 following events are currently defined:
277 CPUFREQ_GOV_START: This governor shall start its duty for the CPU
279 CPUFREQ_GOV_STOP: This governor shall end its duty for the CPU
281 CPUFREQ_GOV_LIMITS: The limits for CPU policy->cpu have changed to
282 policy->min and policy->max.
284 If you need other "events" externally of your driver, _only_ use the
285 cpufreq_governor_l(unsigned int cpu, unsigned int event) call to the
286 CPUfreq core to ensure proper locking.
289 The CPUfreq governor may call the CPU processor driver using one of
292 int cpufreq_driver_target(struct cpufreq_policy *policy,
293 unsigned int target_freq,
294 unsigned int relation);
296 int __cpufreq_driver_target(struct cpufreq_policy *policy,
297 unsigned int target_freq,
298 unsigned int relation);
300 target_freq must be within policy->min and policy->max, of course.
301 What's the difference between these two functions? When your governor
302 still is in a direct code path of a call to governor->governor, the
303 per-CPU cpufreq lock is still held in the cpufreq core, and there's
304 no need to lock it again (in fact, this would cause a deadlock). So
305 use __cpufreq_driver_target only in these cases. In all other cases
306 (for example, when there's a "daemonized" function that wakes up
307 every second), use cpufreq_driver_target to lock the cpufreq per-CPU
308 lock before the command is passed to the cpufreq processor driver.