CINXE.COM
LKML: david singleton: Robust futex update:
<?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: david singleton: Robust futex update:</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 david singleton" href="/groupie.php?aid=30348" /><!--Matomo--><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><!--End Matomo Code--></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/2005/12/9"> [9]</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/2005/12/9/260" onclick="this.href='/lkml/headers'+'/2005/12/9/260';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/9/260">[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/2005/12/9/260">First message in thread</a></li><li class="origin"><a href="/lkml/2005/12/11/41">david singleton</a><ul><li><a href="/lkml/2005/12/11/41">Ingo Molnar</a></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">From</td><td class="rp" itemprop="author">david singleton <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Robust futex update:</td></tr><tr><td class="lp">Date</td><td class="rp" itemprop="datePublished">Fri, 9 Dec 2005 16:23:53 -0800</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody"> There is a new robust futex patch on <br /><a href="http://source.mvista.com/~dsingleton/patch-2.6.14-rt22-rf11">http://source.mvista.com/~dsingleton/patch-2.6.14-rt22-rf11</a><br /> that fixes three bugs. One of the bugs was an SMP race condition that <br />Dave Carlson had spotted<br /> and has been gracious enough to test for me.<br /><br /> The bug fixes are:<br /><br /> 1) Down_futex() must call put_task_struct() before the<br /> call to __down_interruptible so other blocking threads<br /> can find the owning thread's task struct and block on the futex.<br /><br /> If put_task_struct is called after down_interruptible returns<br /> only the first waiter can block through down_interruptible,<br /> all other waiters can not find the task struct until the<br /> first waiter is woken.<br /><br /> 2) Robust mutexes only set a timeout if the user requests one.<br /> This is mainly a performance optimization. We don't set up<br /> and tear down a timeout unless the user requested one.<br /><br /> 3) To fix an SMP race condition between wait and wake<br /> futex_wait_robust() now passes the same parameters and performs<br /> the same checks for user land race as futex_wait().<br /><br /> This change makes the code simpler and more similar to <br />futex_wait.<br /><br /> Futex_wait passes in the value that that glibc got from the<br /> pthread_mutex at compare and xchange time. Futex_wait_robust <br />now<br /> does the same check, after aquiring the correct locks, to see if<br /> the lock in user space has changed since entering the kernel.<br /><br /> (There is a corresponding patch for glibc to pass the lock<br /> value into futex_wait_robust.)<br /><br /> Futex wake also has a new race condition check for the following<br /> SMP race scenario:<br /><br /> Thread A gets the pthread_mutex via the fast path and does not<br /> enter the kernel.<br /><br /> Thread B tries to lock the same pthread mutex and discovers its<br /> already locked. It sets the waiters flag in the lock and enters<br /> the kernel to block on the futex.<br /><br /> Thread A releases the lock and sees the waiters flag. Thread A<br /> enters the kernel to unlock the futex and wake the waiters<br /> before thread B can get into the kernel and lock the futex<br /> on behalf of thread A.<br /><br /> Thread A discovers that it does not own the lock and returns<br /> EINVAL erroneously.<br /><br /> With this fix Thread A now locks the futex and unlocks the<br /> robust semaphore protecting the futex and the mmap_sem<br /> protecting the vma and goes around the loop one more time<br /> to allow the waiters a chance to actually block before the<br /> wake occurs.<br /><br /><br />David <br /><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: 2005-12-10 01:26 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>