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: George Anzinger: Re: [patch 00/21] hrtimer - High-resolution timer subsystem</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 George Anzinger" href="/groupie.php?aid=60" /><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/2005"> [2005]</a> 聽 <a class="nb" href="/lkml/2005/12"> [Dec]</a> 聽 <a class="nb" href="/lkml/2005/12/14"> [14]</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/2005/12/14/394" onclick="this.href='/lkml/headers'+'/2005/12/14/394';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/14/394">[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/2005/12/5/313">First message in thread</a></li><li><a href="/lkml/2005/12/14/269">Roman Zippel</a><ul><li><a href="/lkml/2005/12/14/299">Thomas Gleixner</a><ul><li class="origin"><a href="">George Anzinger</a></li><li><a href="/lkml/2005/12/15/151">Steven Rostedt</a></li><li><a href="/lkml/2005/12/19/62">Roman Zippel</a><ul><li><a href="/lkml/2005/12/19/180">Thomas Gleixner</a></li></ul></li></ul></li></ul></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">Date</td><td class="rp" itemprop="datePublished">Wed, 14 Dec 2005 16:55:12 -0800</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">George Anzinger <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: [patch 00/21] hrtimer - High-resolution timer subsystem</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">Thomas Gleixner wrote:<br />> Hi,<br />~<br /><br />> <br />> <br />>>This error is actually the expected behaviour for any timer with a <br />>>resolution different from 1 nsec. I don't want to say that we can't have <br />>>such a timer, but I'm not so sure whether this should be the default <br />>>behaviour. I actually prefer George's earlier suggestion of CLOCK_REALTIME <br />>>and CLOCK_REALTIME_HR, where one is possibly faster and the other is more <br />>>precise. Even within the kernel I would prefer to map itimer and nanosleep <br />>>to the first clock (maybe also based on arch/kconfig defaults).<br />>>OTOH if the hardware allows it, both clocks can do the same thing, but I <br />>>really would like to have the possibility to give higher (and thus <br />>>possibly more expensive) resolution only to those asking for it.<br />> <br />> <br />> Thats an rather odd approach for me. If we drag this further then we<br />> might consider that only some users (i.e. applications) of -rt patches<br />> are using the enhanced functionalities, which introduces interesting<br />> computational problems (e.g when to treat a mutex as a concurrency<br />> control which is capable of priority inversion or not). <br /><br />Er... what? This is a non-compute.<br />> <br />> I vote strongly against introducing private, special purpose APIs and I<br />> consider CLOCK_XXX_HR as such. The proposed hrtimer solution does not<br />> introduce any penalties for people who do not enable a future high<br />> resolution extension. It gives us the benefit of a clean code base which<br />> is capable to be switched simply and non intrusive to the high<br />> resolution mode. We have done extensive tests on the impact of<br />> converting all users unconditionally to high resolution mode once it is<br />> switched on and the penalty is within the noise range. <br />> <br />> You are explicitely asking for increased complexity with your approach. <br /><br />I beg to differ here. The fact that high res timers, in general, <br />require an interrupt per expiry, and that, by definition, we are <br />changing the resolution by, I would guess, a couple of orders of <br />magnitude implies a rather much larger over head. If we sum this over <br />all user timers it can IMHO get out of control. Given that only a <br />very small number of applications really need the extra resolution, I <br />think it makes a lot of sense that those applications incur the <br />overhead and others, which don't need nor want the higher resolution, <br />just use the old low resolution timers. The notion of switching this <br />at configure time implies that a given kernel is going to be used ONLY <br />one way or another for all applications, which, AFAICT is just not the <br />way most users do things.<br /><br />As to CLOCK_XXX_HR being a special purpose API, this is only half <br />true. It is a POSIX conforming extension and I do think you can find <br />it used elsewhere as well. On the other hand, it if you want to limit <br />the higher overhead timers to only those who ask, well, I guess you <br />could call that "special purpose".<br /><br />On the complexity thing, your new organization makes the added <br />"complexity" rather non-complex, in fact, you might say it is down <br />right simple, for which, thank you.<br />> <br />> <br />~<br />-- <br />George Anzinger george@mvista.com<br />HRT (High-res-timers): <a href="http://sourceforge.net/projects/high-res-timers/">http://sourceforge.net/projects/high-res-timers/</a><br />-<br />To unsubscribe from this list: send the line "unsubscribe linux-kernel" in<br />the body of a message to majordomo@vger.kernel.org<br />More majordomo info at <a href="http://vger.kernel.org/majordomo-info.html">http://vger.kernel.org/majordomo-info.html</a><br />Please read the FAQ at <a href="http://www.tux.org/lkml/">http://www.tux.org/lkml/</a><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: 2005-12-15 01:59 聽聽 [from the cache]<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>