Changes between Initial Version and Version 1 of DVFS

Timestamp:

10/22/2017 05:28:45 AM (7 years ago)

Author:

trac

Comment:

--

DVFS

@@ +1 @@
+{{{#!html
+          <div id="wikipage" class="trac-content"><p>
+</p><div class="wiki-toc">
+<ol>
+  <li>
+    <a href="#DynamicVoltageandFrequencyScalingDVFS-VentanaOnly">Dynamic Voltage and Frequency Scaling (DVFS) - Ventana Only</a>
+  </li>
+  <li>
+    <a href="#LinuxKernelDVFSAPI">Linux Kernel DVFS API</a>
+    <ol>
+      <li>
+        <a href="#Governors">Governors</a>
+      </li>
+    </ol>
+  </li>
+</ol>
+</div><p>
+</p>
+<h1 id="DynamicVoltageandFrequencyScalingDVFS-VentanaOnly">Dynamic Voltage and Frequency Scaling (DVFS) - Ventana Only</h1>
+<p>
+Modern processors allow their core clocks to be scaled in order to trade-off performance vs power or simple to save power when performance is not needed.  This is referred to as 'Dynamic Frequency Scaling'.  In some cases processors also allow you to scale the core voltages down when using lower clock-speeds which is referred to as 'Dynamic Voltage Scaling'.
+</p>
+<p>
+The Gateworks Ventana family based off the Freescale i.MX6 CPU family allows DVFS
+</p>
+<h1 id="LinuxKernelDVFSAPI">Linux Kernel DVFS API</h1>
+<p>
+The linux kernel provides a DVFS framework that allows each CPU core to have a min/max frequency and a governor that governs it.  A sysfs API exists which is used by the cpufreq-utils package.
+</p>
+<h2 id="Governors">Governors</h2>
+<p>
+Governors are kernel models that can drive CPU core frequency/voltage operating points based on an algorithm.  Currently the following governors exist:
+</p>
+<ul><li>performance - sets the frequency statically to the highest available CPU frequency
+</li><li>powersave -  sets the frequency statically to the lowest available CPU frequency
+</li><li>userspace - set the frequency from a userspace program
+</li><li>ondemand - adjust based on utilization
+</li><li>conservative - adjust based on utilization but be a bit more conservative by adjusting gradually
+</li></ul><p>
+Each cpu has an entry in sysfs at /sys/devices/system/cpu/cpu&lt;n&gt;/cpufreq/ at /sys/devices/system/cpuwith the following files:
+</p>
+<ul><li>scaling_governor - get/set the governor
+</li><li>scaling_available_governors - get the available governors
+</li><li>scaling_available_frequencies - get the available frequencies
+</li><li>cpuinfo_cur_freq - get/set the current freq
+</li><li>cpuinfo_max_freq - get/set the max freq
+</li><li>cpuinfo_min_freq - get/set the min freq
+</li><li>scaling_setspeed - set the current freq (if using the userspace governor)
+</li><li>affected_cpus - which cpu's are affected together
+</li></ul><p>
+Examples:
+</p>
+<ul><li>show governor for cpu0
+<pre class="wiki">cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
+</pre></li><li>show available frequencies (in MHz) for cpu0
+<pre class="wiki">cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
+</pre></li><li>set *conservative* governor for a good mix of power-saving and performance
+<pre class="wiki">echo conservative &gt; /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
+</pre></li><li>set to lowest power mode using the *powersave* governor:
+<pre class="wiki">echo powersave &gt; /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
+</pre></li><li>set to highest performance mode using the *performance* governor:
+<pre class="wiki">echo performance &gt; /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
+</pre></li><li>set *userspace* governer because perhaps you have a script that wants to adjust it manually based on your own criteria over time:
+<pre class="wiki">echo userspace &gt; /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
+echo 396000 &gt; /sys/devices/system/cpu/cpu0/cpufreq/setspeed
+echo 792000 &gt; /sys/devices/system/cpu/cpu0/cpufreq/setspeed
+echo 996000 &gt; /sys/devices/system/cpu/cpu0/cpufreq/setspeed
+</pre></li></ul
+}}}