CINXE.COM

<?xml version="1.0" encoding="UTF-8" standalone="yes"?> <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>LKML: "Rafael J. Wysocki": [RFT][PATCH v1 5/5] cpuidle: menu: Avoid discarding useful information</title><link href="/css/message.css" rel="stylesheet" type="text/css" /><link href="/css/wrap.css" rel="alternate stylesheet" type="text/css" title="wrap" /><link href="/css/nowrap.css" rel="stylesheet" type="text/css" title="nowrap" /><link href="/favicon.ico" rel="shortcut icon" /><script src="/js/simple-calendar.js" type="text/javascript"></script><script src="/js/styleswitcher.js" type="text/javascript"></script><link rel="alternate" type="application/rss+xml" title="lkml.org : last 100 messages" href="/rss.php" /><link rel="alternate" type="application/rss+xml" title="lkml.org : last messages by "Rafael J. Wysocki"" href="/groupie.php?aid=" /><script> var _paq = window._paq = window._paq || []; /* tracker methods like "setCustomDimension" should be called before "trackPageView" */ _paq.push(["setDoNotTrack", true]); _paq.push(["disableCookies"]); _paq.push(['trackPageView']); _paq.push(['enableLinkTracking']); (function() { var u="//m.lkml.org/"; _paq.push(['setTrackerUrl', u+'matomo.php']); _paq.push(['setSiteId', '1']); var d=document, g=d.createElement('script'), s=d.getElementsByTagName('script')[0]; g.async=true; g.src=u+'matomo.js'; s.parentNode.insertBefore(g,s); })(); </script></head><body onload="es.jasper.simpleCalendar.init();" itemscope="itemscope" itemtype="http://schema.org/BlogPosting"><table border="0" cellpadding="0" cellspacing="0"><tr><td width="180" align="center"><a href="/"><img style="border:0;width:135px;height:32px" src="/images/toprowlk.gif" alt="lkml.org" /></a></td><td width="32">聽</td><td class="nb"><div><a class="nb" href="/lkml"> [lkml]</a> 聽 <a class="nb" href="/lkml/2025"> [2025]</a> 聽 <a class="nb" href="/lkml/2025/2"> [Feb]</a> 聽 <a class="nb" href="/lkml/2025/2/6"> [6]</a> 聽 <a class="nb" href="/lkml/last100"> [last100]</a> 聽 <a href="/rss.php"><img src="/images/rss-or.gif" border="0" alt="RSS Feed" /></a></div><div>Views: <a href="#" class="nowrap" onclick="setActiveStyleSheet('wrap');return false;">[wrap]</a><a href="#" class="wrap" onclick="setActiveStyleSheet('nowrap');return false;">[no wrap]</a> 聽 <a class="nb" href="/lkml/mheaders/2025/2/6/924" onclick="this.href='/lkml/headers'+'/2025/2/6/924';">[headers]</a>聽 <a href="/lkml/bounce/2025/2/6/924">[forward]</a>聽 </div></td><td width="32">聽</td></tr><tr><td valign="top"><div class="es-jasper-simpleCalendar" baseurl="/lkml/"></div><div class="threadlist">Messages in this thread</div><ul class="threadlist"><li class="root"><a href="/lkml/2025/2/6/927">First message in thread</a></li><li><a href="/lkml/2025/2/6/927">"Rafael J. Wysocki"</a><ul><li><a href="/lkml/2025/2/6/923">"Rafael J. Wysocki"</a><ul><li><a href="/lkml/2025/2/17/790">Christian Loehle</a></li></ul></li><li class="origin"><a href="/lkml/2025/2/17/866">"Rafael J. Wysocki"</a><ul><li><a href="/lkml/2025/2/17/866">Christian Loehle</a><ul><li><a href="/lkml/2025/2/17/880">"Rafael J. Wysocki"</a></li></ul></li></ul></li><li><a href="/lkml/2025/2/6/925">"Rafael J. Wysocki"</a><ul><li><a href="/lkml/2025/2/17/853">Christian Loehle</a></li></ul></li><li><a href="/lkml/2025/2/6/926">"Rafael J. Wysocki"</a><ul><li><a href="/lkml/2025/2/17/804">Christian Loehle</a></li></ul></li><li><a href="/lkml/2025/2/6/928">"Rafael J. Wysocki"</a><ul><li><a href="/lkml/2025/2/6/961">Christian Loehle</a></li></ul></li><li><a href="/lkml/2025/2/7/1027">Artem Bityutskiy</a><ul><li><a href="/lkml/2025/2/7/1103">Christian Loehle</a><ul><li><a href="/lkml/2025/2/7/1115">"Rafael J. Wysocki"</a></li></ul></li><li><a href="/lkml/2025/2/7/1157">"Rafael J. Wysocki"</a></li></ul></li><li><a href="/lkml/2025/2/10/922">Christian Loehle</a><ul><li><a href="/lkml/2025/2/10/965">"Rafael J. Wysocki"</a><ul><li><a href="/lkml/2025/2/10/972">Christian Loehle</a></li></ul></li><li><a href="/lkml/2025/2/18/1941">Christian Loehle</a><ul><li><a href="/lkml/2025/2/19/773">"Rafael J. Wysocki"</a></li></ul></li></ul></li><li><a href="/lkml/2025/2/13/1996">"Doug Smythies"</a><ul><li><a href="/lkml/2025/2/14/1548">"Rafael J. Wysocki"</a><ul><li><a href="/lkml/2025/2/16/278">"Doug Smythies"</a></li></ul></li></ul></li><li><a href="/lkml/2025/2/24/221">Aboorva Devarajan</a><ul><li><a href="/lkml/2025/2/24/240">Aboorva Devarajan</a></li><li><a href="/lkml/2025/2/24/1038">"Rafael J. Wysocki"</a></li><li><a href="/lkml/2025/2/26/233">Aboorva Devarajan</a><ul><li><a href="/lkml/2025/2/26/773">Christian Loehle</a></li></ul></li></ul></li></ul></li></ul><div class="threadlist">Patch in this message</div><ul class="threadlist"><li><a href="/lkml/diff/2025/2/6/924/1">Get diff 1</a></li></ul></td><td width="32" rowspan="2" class="c" valign="top"><img src="/images/icornerl.gif" width="32" height="32" alt="/" /></td><td class="c" rowspan="2" valign="top" style="padding-top: 1em"><table><tr><td><table><tr><td class="lp">From</td><td class="rp" itemprop="author">"Rafael J. Wysocki" <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">[RFT][PATCH v1 5/5] cpuidle: menu: Avoid discarding useful information</td></tr><tr><td class="lp">Date</td><td class="rp" itemprop="datePublished">Thu, 06 Feb 2025 15:29:05 +0100</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">From: Rafael J. Wysocki <rafael.j.wysocki@intel.com><br /><br />When giving up on making a high-confidence prediction,<br />get_typical_interval() always returns UINT_MAX which means that the<br />next idle interval prediction will be based entirely on the time till<br />the next timer. However, the information represented by the most<br />recent intervals may not be completely useless in those cases.<br /><br />Namely, the largest recent idle interval is an upper bound on the<br />recently observed idle duration, so it is reasonable to assume that<br />the next idle duration is unlikely to exceed it. Moreover, this is<br />still true after eliminating the suspected outliers if the sample<br />set still under consideration is at least as large as 50% of the<br />maximum sample set size.<br /><br />Accordingly, make get_typical_interval() return the current maximum<br />recent interval value in that case instead of UINT_MAX.<br /><br />Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com><br />---<br /> drivers/cpuidle/governors/menu.c | 13 ++++++++++++-<br /> 1 file changed, 12 insertions(+), 1 deletion(-)<br /><br />--- a/drivers/cpuidle/governors/menu.c<br />+++ b/drivers/cpuidle/governors/menu.c<br />@@ -190,8 +190,19 @@<br /> * This can deal with workloads that have long pauses interspersed<br /> * with sporadic activity with a bunch of short pauses.<br /> */<br />- if ((divisor * 4) <= INTERVALS * 3)<br />+ if (divisor * 4 <= INTERVALS * 3) {<br />+ /*<br />+ * If there are sufficiently many data points still under<br />+ * consideration after the outliers have been eliminated,<br />+ * returning without a prediction would be a mistake because it<br />+ * is likely that the next interval will not exceed the current<br />+ * maximum, so return the latter in that case.<br />+ */<br />+ if (divisor >= INTERVALS / 2)<br />+ return max;<br />+<br /> return UINT_MAX;<br />+ }<br /> <br /> /* Update the thresholds for the next round. */<br /> if (avg - min > max - avg)<br /><br /><br /><br /></pre></td><td width="32" rowspan="2" class="c" valign="top"><img src="/images/icornerr.gif" width="32" height="32" alt="\" /></td></tr><tr><td align="right" valign="bottom"> 聽 </td></tr><tr><td align="right" valign="bottom">聽</td><td class="c" valign="bottom" style="padding-bottom: 0px"><img src="/images/bcornerl.gif" width="32" height="32" alt="\" /></td><td class="c">聽</td><td class="c" valign="bottom" style="padding-bottom: 0px"><img src="/images/bcornerr.gif" width="32" height="32" alt="/" /></td></tr><tr><td align="right" valign="top" colspan="2"> 聽 </td><td class="lm">Last update: 2025-02-06 15:32 聽聽 [W:0.841 / U:2.155 seconds]<br />漏2003-2020 <a href="http://blog.jasper.es/"><span itemprop="editor">Jasper Spaans</span></a>|hosted at <a href="https://www.digitalocean.com/?refcode=9a8e99d24cf9">Digital Ocean</a> and my Meterkast|<a href="http://blog.jasper.es/categories.html#lkml-ref">Read the blog</a></td><td>聽</td></tr></table><script language="javascript" src="/js/styleswitcher.js" type="text/javascript"></script></body></html>