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: Matt Mackall: Re: [RFC][PATCH 0/3] TCP/IP Critical socket communication mechanism</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 Matt Mackall" href="/groupie.php?aid=3458" /><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/2005/12/15"> [15]</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/2005/12/15/2" onclick="this.href='/lkml/headers'+'/2005/12/15/2';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/15/2">[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/2005/12/14/65">First message in thread</a></li><li><a href="/lkml/2005/12/14/450">Matt Mackall</a><ul><li><a href="/lkml/2005/12/14/459">"David S. Miller"</a><ul><li class="origin"><a href="/lkml/2005/12/15/9">Matt Mackall</a><ul><li><a href="/lkml/2005/12/15/9">"David S. Miller"</a><ul><li><a href="/lkml/2005/12/15/18">Nick Piggin</a></li><li><a href="/lkml/2005/12/15/19">Stephen Hemminger</a></li><li><a href="/lkml/2005/12/15/20">Matt Mackall</a></li></ul></li></ul></li><li><a href="/lkml/2005/12/15/12">Andi Kleen</a><ul><li><a href="/lkml/2005/12/15/23">Stephen Hemminger</a></li></ul></li><li><a href="/lkml/2005/12/15/37">Sridhar Samudrala</a><ul><li><a href="/lkml/2005/12/15/45">"David S. Miller"</a><ul><li><a href="/lkml/2005/12/15/47">Arjan van de Ven</a></li><li><a href="/lkml/2005/12/15/56">Kyle Moffett</a></li><li><a href="/lkml/2005/12/15/433">Sridhar Samudrala</a></li></ul></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 21:02:50 -0800</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Matt Mackall <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: [RFC][PATCH 0/3] TCP/IP Critical socket communication mechanism</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">On Wed, Dec 14, 2005 at 08:30:23PM -0800, David S. Miller wrote:<br />> From: Matt Mackall <mpm@selenic.com><br />> Date: Wed, 14 Dec 2005 19:39:37 -0800<br />> <br />> > I think we need a global receive pool and per-socket send pools.<br />> <br />> Mind telling everyone how you plan to make use of the global receive<br />> pool when the allocation happens in the device driver and we have no<br />> idea which socket the packet is destined for? What should be done for<br />> non-local packets being routed? The device drivers allocate packets<br />> for the entire system, long before we know who the eventually received<br />> packets are for. It is fully anonymous memory, and it's easy to<br />> design cases where the whole pool can be eaten up by non-local<br />> forwarded packets.<br /><br />There needs to be two rules:<br /><br />iff global memory critical flag is set<br />- allocate from the global critical receive pool on receive<br />- return packet to global pool if not destined for a socket with an<br /> attached send mempool<br /><br />I think this will provide the desired behavior, though only<br />probabilistically. That is, we can fill the global receive pool with<br />uninteresting packets such that we're forced to drop critical ACKs,<br />but the boring packets will eventually be discarded as we walk up the<br />stack and we'll eventually have room to receive retried ACKs.<br /><br />> I truly dislike these patches being discussed because they are a<br />> complete hack, and admittedly don't even solve the problem fully. I<br />> don't have any concrete better ideas but that doesn't mean this stuff<br />> should go into the tree.<br /><br />Agreed. I'm fairly convinced a full fix is doable, if you make a<br />couple assumptions (limited fragmentation), but will unavoidably be<br />less than pretty as it needs to cross some layers.<br /><br />> I think GFP_ATOMIC memory pools are more powerful than they are given<br />> credit for. There is nothing preventing the implementation of dynamic<br />> GFP_ATOMIC watermarks, and having "critical" socket behavior "kick in"<br />> in response to hitting those water marks.<br /><br />There are two problems with GFP_ATOMIC. The first is that its users<br />don't pre-state their worst-case usage, which means sizing the pool to<br />reliably avoid deadlocks is impossible. The second is that there<br />aren't any guarantees that GFP_ATOMIC allocations are actually<br />critical in the needed-to-make-forward-VM-progress sense or will be<br />returned to the pool in a timely fashion.<br /><br />So I do think we need a distinct pool if we want to tackle this<br />problem. Though it's probably worth mentioning that Linus was rather<br />adamantly against even trying at KS.<br /><br />-- <br />Mathematics is the supreme nostalgia of our time.<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 06:12 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>