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: (Eric W. Biederman): Re: [PATCH 01/13] [RFC] ipath basic headers</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 (Eric W. Biederman)" href="/groupie.php?aid=1272" /><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/2005/12/17"> [17]</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/2005/12/17/96" onclick="this.href='/lkml/headers'+'/2005/12/17/96';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/17/96">[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/2005/12/16/290">First message in thread</a></li><li><a href="/lkml/2005/12/16/293">Roland Dreier</a><ul><li><a href="/lkml/2005/12/16/289">Roland Dreier</a><ul><li><a href="/lkml/2005/12/16/291">Roland Dreier</a><ul><li><a href="/lkml/2005/12/16/303">Roland Dreier</a><ul><li><a href="/lkml/2005/12/16/301">Roland Dreier</a></li><li><a href="/lkml/2005/12/17/72">Andrew Morton</a></li></ul></li><li><a href="/lkml/2005/12/17/25">Pekka Enberg</a><ul><li><a href="/lkml/2005/12/17/92">Robert Walsh</a></li></ul></li><li><a href="/lkml/2005/12/17/31">Christoph Hellwig</a></li><li><a href="/lkml/2005/12/17/78">Andrew Morton</a><ul><li><a href="/lkml/2005/12/17/103">Robert Walsh</a></li></ul></li></ul></li></ul></li><li><a href="/lkml/2005/12/17/23">Pekka Enberg</a><ul><li><a href="/lkml/2005/12/17/97">Robert Walsh</a></li></ul></li><li><a href="/lkml/2005/12/17/29">Christoph Hellwig</a><ul><li class="origin"><a href="/lkml/2005/12/17/129">(Eric W. Biederman)</a><ul><li><a href="/lkml/2005/12/17/129">Andi Kleen</a><ul><li><a href="/lkml/2005/12/18/34">(Eric W. Biederman)</a></li></ul></li></ul></li><li><a href="/lkml/2005/12/17/98">Robert Walsh</a><ul><li><a href="/lkml/2005/12/17/100">Arjan van de Ven</a><ul><li><a href="/lkml/2005/12/17/104">Robert Walsh</a></li></ul></li></ul></li></ul></li><li><a href="/lkml/2005/12/17/77">Andrew Morton</a><ul><li><a href="/lkml/2005/12/17/102">Robert Walsh</a><ul><li><a href="/lkml/2005/12/17/126">Andrew Morton</a></li></ul></li><li><a href="/lkml/2005/12/19/207">Robert Walsh</a></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">Subject</td><td class="rp" itemprop="name">Re: [PATCH 01/13] [RFC] ipath basic headers</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">(Eric W. Biederman)</td></tr><tr><td class="lp">Date</td><td class="rp" itemprop="datePublished">Sat, 17 Dec 2005 14:51:19 -0700</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">Christoph Hellwig <hch@infradead.org> writes:<br /><br />> please always used fixes-size types for user communication. also please<br />> avoid ioctls like the rest of the IB codebase.<br /><br />Could someone please explain to me how the uverbs abuse of write<br />is better that ioctl? <br /><br />Every single command seems to have a __u64 response fields that is a<br />pointer into user space. When you write your commands and read your<br />responses like the netlink layer does I can see the sense of it. But<br />making write an ioctl by another name...<br /><br />One of the scarier comments I have seen lately from ib_user_verbs.h<br />/*<br /> * Make sure that all structs defined in this file remain laid out so<br /> * that they pack the same way on 32-bit and 64-bit architectures (to<br /> * avoid incompatibility between 32-bit userspace and 64-bit kernels).<br /> * Specifically:<br /> * - Do not use pointer types -- pass pointers in __u64 instead.<br /> * - Make sure that any structure larger than 4 bytes is padded to a<br /> * multiple of 8 bytes. Otherwise the structure size will be<br /> * different between 32-bit and 64-bit architectures.<br /> */<br /><br />The two points that get called out.<br />- Embedded pointers are a large part of what make ioctl a maintenance<br /> nightmare. I admit we are 15-20 years away before big machines<br /> exhaust the capability of 64bit pointers so we aren't likely<br /> to run into size issues soon. But a write that changes your address<br /> space is ugly, and unexpected.<br /><br /> What looks like a reimplementation of readv/writev using this<br /> technique is also scary.<br /><br />- 64bit compilers will not pad every structure to 8 bytes. This<br /> only will happen if you happen to have an 8 byte element in your<br /> structure that is only aligned to 32bits by a 32bit structure.<br /> Unfortunately the 32bit gcc only aligns long long to 32bits on<br /> x86, which triggers the described behavior.<br /><br /><br /><br />Eric<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: 2005-12-17 22:55 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>