LKML: Oleg Nesterov: Re: [PATCH] Fix RCU race in access of nohz_cpu

<?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: Oleg Nesterov: Re: [PATCH] Fix RCU race in access of nohz_cpu_mask</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 Oleg Nesterov" href="/groupie.php?aid=4105" /><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/2005/12/8"> [8]</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/2005/12/8/155" onclick="this.href='/lkml/headers'+'/2005/12/8/155';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/8/155">[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/2005/12/8/155">First message in thread</a></li><li class="origin"><a href="/lkml/2005/12/8/308">Oleg Nesterov</a><ul><li><a href="/lkml/2005/12/8/308">Srivatsa Vaddagiri</a><ul><li><a href="/lkml/2005/12/9/156">Oleg Nesterov</a><ul><li><a href="/lkml/2005/12/10/68">Srivatsa Vaddagiri</a><ul><li><a href="/lkml/2005/12/10/79">Oleg Nesterov</a></li></ul></li></ul></li></ul></li><li><a href="/lkml/2005/12/8/310">Srivatsa Vaddagiri</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">Date</td><td class="rp" itemprop="datePublished">Thu, 08 Dec 2005 22:31:06 +0300</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Oleg Nesterov <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: [PATCH] Fix RCU race in access of nohz_cpu_mask</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">Srivatsa Vaddagiri wrote:<br />><br />> Accessing nohz_cpu_mask before incrementing rcp->cur is racy. It can<br />> cause tickless idle CPUs to be included in rsp->cpumask, which will<br />> extend graceperiods unnecessarily.<br />> ...<br />> @@ -244,15 +244,15 @@ static void rcu_start_batch(struct rcu_c<br />><br />> if (rcp->next_pending &&<br />> rcp->completed == rcp->cur) {<br />> - /* Can't change, since spin lock held. */<br />> - cpus_andnot(rsp->cpumask, cpu_online_map, nohz_cpu_mask);<br />> -<br />> rcp->next_pending = 0;<br />> /* next_pending == 0 must be visible in __rcu_process_callbacks()<br />> * before it can see new value of cur.<br />> */<br />> smp_wmb();<br />> rcp->cur++;<br />> + smp_mb();<br />> + cpus_andnot(rsp->cpumask, cpu_online_map, nohz_cpu_mask);<br />> +<br /><br />Could you explain why this patch makes any difference?<br /><br />grep shows the only user of nohz_cpu_mask in arch/s390/kernel/time.c,<br />start_hz_timer() and stop_hz_timer().<br /><br />I can't see how this change can prevent idle cpus to be included in<br />->cpumask, cpu can add itself to nohz_cpu_mask right after some other<br />cpu started new grace period.<br /><br />I think cpu should call cpu_quiet() after adding itself to nohz mask<br />to eliminate this race.<br /><br />No?<br /><br />Oleg.<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: 2005-12-08 19:18 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>

CINXE.COM

LKML: Oleg Nesterov: Re: [PATCH] Fix RCU race in access of nohz_cpu_mask