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: Andrew Morton: Re: [PATCH 07/13] [RFC] ipath core misc files</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 Andrew Morton" href="/groupie.php?aid=28391" /><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/2005/12/17/74" onclick="this.href='/lkml/headers'+'/2005/12/17/74';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/17/74">[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/2005/12/16/301">Roland Dreier</a><ul><li><a href="/lkml/2005/12/16/297">Roland Dreier</a><ul><li><a href="/lkml/2005/12/16/296">Roland Dreier</a><ul><li><a href="/lkml/2005/12/16/305">Roland Dreier</a><ul><li><a href="/lkml/2005/12/16/299">Roland Dreier</a></li><li><a href="/lkml/2005/12/17/75">Andrew Morton</a></li></ul></li></ul></li><li class="origin"><a href="/lkml/2005/12/17/87">Andrew Morton</a><ul><li><a href="/lkml/2005/12/17/87">Robert Walsh</a></li><li><a href="/lkml/2005/12/17/90">Robert Walsh</a><ul><li><a href="/lkml/2005/12/17/124">Andrew Morton</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">Sat, 17 Dec 2005 12:38:50 -0800</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Andrew Morton <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: [PATCH 07/13] [RFC] ipath core misc files</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">Roland Dreier <rolandd@cisco.com> wrote: > > ... > +/* > + * This isn't perfect, but it's close enough for timing work. We want this > + * to work on systems where the cycle counter isn't the same as the clock > + * frequency. The one msec spin is OK, since we execute this only once > + * when first loaded. We don't use CURRENT_TIME because on some systems > + * it only has jiffy resolution; we just assume udelay is well calibrated > + * and that we aren't likely to be rescheduled. Do it multiple times, > + * with a yield in between, to try to make sure we get the "true minimum" > + * value. > + * _ipath_pico_per_cycle isn't going to lead to completely accurate > + * conversions from timestamps to nanoseconds, but it's close enough > + * for our purposes, which is mainly to allow people to show events with > + * nsecs or usecs if desired, rather than cycles. > + */ > +void ipath_init_picotime(void) > +{ > + int i; > + u_int64_t ts, te, delta = -1ULL; > + > + for (i = 0; i < 5; i++) { > + ts = get_cycles(); > + udelay(250); > + te = get_cycles(); > + if ((te - ts) < delta) > + delta = te - ts; > + yield(); > + } > + _ipath_pico_per_cycle = 250000000 / delta; > +} hm, I hope this is debug code which is going away. If not, we should take a look at what it's trying to do here. > +/* > + * Our version of the kernel mlock function. This function is no longer > + * exposed, so we need to do it ourselves. It takes a given start page > + * (page aligned user virtual address) and pins it and the following specified > + * number of pages. > + * For now, num_pages is always 1, but that will probably change at some > + * point (because caller is doing expected sends on a single virtually > + * contiguous buffer, so we can do all pages at once). > + */ > +int ipath_mlock(unsigned long start_page, size_t num_pages, struct page **p) > +{ > + int n; > + > + _IPATH_VDBG("pin %lx pages from vaddr %lx\n", num_pages, start_page); > + down_read(&current->mm->mmap_sem); > + n = get_user_pages(current, current->mm, start_page, num_pages, 1, 1, > + p, NULL); > + up_read(&current->mm->mmap_sem); > + if (n != num_pages) { > + _IPATH_INFO > + ("get_user_pages (0x%lx pages starting at 0x%lx failed with %d\n", > + num_pages, start_page, n); > + if (n < 0) /* it's an errno */ > + return n; > + return -ENOMEM; /* no way to know actual error */ > + } > + > + return 0; > +} OK. It's perhaps not a very well named function. > +/* > + * this is similar to ipath_mlock, but it's always one page, and we mark > + * the page as locked for i/o, and shared. This is used for the user process > + * page that contains the destination address for the rcvhdrq tail update, > + * so we need to have the vma. If we don't do this, the page can be taken > + * away from us on fork, even if the child never touches it, and then > + * the user process never sees the tail register updates. > + */ > +int ipath_mlock_nocopy(unsigned long start_page, struct page **p) > +{ > + int n; > + struct vm_area_struct *vm = NULL; > + > + down_read(&current->mm->mmap_sem); > + n = get_user_pages(current, current->mm, start_page, 1, 1, 1, p, &vm); > + up_read(&current->mm->mmap_sem); > + if (n != 1) { > + _IPATH_INFO("get_user_pages for 0x%lx failed with %d\n", > + start_page, n); > + if (n < 0) /* it's an errno */ > + return n; > + return -ENOMEM; /* no way to know actual error */ > + } > + vm->vm_flags |= VM_SHM | VM_LOCKED; > + > + return 0; > +} I don't think we want to be setting the user's VMA's vm_flags in this manner. This is purely to retain the physical page across fork? > +/* > + * Our version of the kernel munlock function. This function is no longer > + * exposed, so we need to do it ourselves. It unpins the start page > + * (a page aligned full user virtual address, not a page number) > + * and pins it and the following specified number of pages. > + */ > +int ipath_munlock(size_t num_pages, struct page **p) > +{ > + int i; > + > + for (i = 0; i < num_pages; i++) { > + _IPATH_MMDBG("%u/%lu put_page %p\n", i, num_pages, p[i]); > + SetPageDirty(p[i]); > + put_page(p[i]); > + } > + return 0; > +} Nope, SetPageDirty() doesn't tell the VM that the page is dirty - it'll never get written out. Use set_page_dirty_lock(). - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at <a href="http://vger.kernel.org/majordomo-info.html">http://vger.kernel.org/majordomo-info.html</a> Please read the FAQ at <a href="http://www.tux.org/lkml/">http://www.tux.org/lkml/</a> </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 21:42 聽聽 [from the cache] 漏2003-2020 <a href="http://blog.jasper.es/">Jasper Spaans</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>