CINXE.COM
LKML: Mike Christie: Re: allowed pages in the block later, was Re: [Ext2-devel] [PATCH] ext3: avoid sending down non-refcounted pages
<?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: Mike Christie: Re: allowed pages in the block later, was Re: [Ext2-devel] [PATCH] ext3: avoid sending down non-refcounted pages</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 Mike Christie" href="/groupie.php?aid=17355" /><!--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/2005/12/8"> [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/2005/12/8/156" onclick="this.href='/lkml/headers'+'/2005/12/8/156';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/8/156">[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/2005/12/8/81">First message in thread</a></li><li><a href="/lkml/2005/12/8/81">Christoph Hellwig</a><ul><li><a href="/lkml/2005/12/8/87">Pekka Enberg</a><ul><li><a href="/lkml/2005/12/12/132">Christoph Hellwig</a></li></ul></li><li class="origin"><a href="/lkml/2005/12/8/158">Mike Christie</a><ul><li><a href="/lkml/2005/12/8/158">Mike Christie</a><ul><li><a href="/lkml/2005/12/8/170">Pekka Enberg</a></li></ul></li></ul></li><li><a href="/lkml/2005/12/10/129">Andrew Morton</a><ul><li><a href="/lkml/2005/12/11/23">Arjan van de Ven</a></li><li><a href="/lkml/2005/12/12/131">Christoph Hellwig</a><ul><li><a href="/lkml/2005/12/12/185">Andrew Morton</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">Thu, 08 Dec 2005 12:18:12 -0600</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Mike Christie <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: allowed pages in the block later, was Re: [Ext2-devel] [PATCH] ext3: avoid sending down non-refcounted pages</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">Christoph Hellwig wrote:<br />> On Thu, Dec 08, 2005 at 03:18:33AM -0700, Andreas Dilger wrote:<br />> <br />>>What happens on 1kB or 2kB block filesystems (i.e. b_size != PAGE_SIZE)?<br />>>This will allocate a whole page for each block (which may be considerable<br />>>overhead on e.g. a 64kB PAGE_SIZE ia64 or PPC system).<br />> <br />> <br />> Yes. How often do we trigger this codepath?<br />> <br />> The problem we're trying to solve here is how do implement network block<br />> devices (nbd, iscsi) efficiently. The zero copy codepath in the networking<br />> layer does need to grab additional references to pages. So to use sendpage<br />> we need a refcountable page. pages used by the slab allocator are not<br />> normally refcounted so try to do get_page/pub_page on them will break.<br />> <br />> One way to work around that would be to detect kmalloced pages and use<br />> a slowpath for that. The major issues with that is that we don't have a<br />> reliable way to detect if a given struct page comes from the slab allocator<br />> or not. The minor problem is that even with such an indicator it means<br />> having a separate and lightly tested slowpath for this rare case.<br />> <br />> All in all I think we should document that the block layer only accepts<br />> properly refcounted pages, which is everything but kmalloced pages (even<br />> vmalloc is totally fine)<br /><br />Is it anytime kmalloc is used? For scsi when it uses scsi_execute* for <br />something like scanning (report luns result is kmallocd) would this be a <br />problem?<br /><br />If PageSlab() does work, then could we have a request queue flag that <br />bounces those pages for all block layer drivers. Pretty slow and yucky <br />but if we have to convert SCSI and maybe other parts of the block layer <br />maybe it will be easiest for now.<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: 2005-12-08 19:21 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>