LKML: Ananth N Mavinakayanahalli: Re: [(resend)patch 7/7] Kprobes - Register for page fault notify on active probes

<?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: Ananth N Mavinakayanahalli: Re: [(resend)patch 7/7] Kprobes - Register for page fault notify on active probes</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 Ananth N Mavinakayanahalli" href="/groupie.php?aid=16741" /><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/2006"> [2006]</a> 聽 <a class="nb" href="/lkml/2006/4"> [Apr]</a> 聽 <a class="nb" href="/lkml/2006/4/19"> [19]</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/2006/4/19/368" onclick="this.href='/lkml/headers'+'/2006/4/19/368';">[headers]</a>聽 <a href="/lkml/bounce/2006/4/19/368">[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/2006/4/19/324">First message in thread</a></li><li><a href="/lkml/2006/4/19/324">Anil S Keshavamurthy</a><ul><li><a href="/lkml/2006/4/19/323">Anil S Keshavamurthy</a></li><li><a href="/lkml/2006/4/19/326">Anil S Keshavamurthy</a></li><li><a href="/lkml/2006/4/19/327">Anil S Keshavamurthy</a></li><li><a href="/lkml/2006/4/19/328">Anil S Keshavamurthy</a></li><li><a href="/lkml/2006/4/19/329">Anil S Keshavamurthy</a><ul><li class="origin"><a href="/lkml/2006/4/20/6">Ananth N Mavinakayanahalli</a><ul><li><a href="/lkml/2006/4/20/6">Keshavamurthy Anil S</a></li></ul></li></ul></li><li><a href="/lkml/2006/4/19/330">Anil S Keshavamurthy</a><ul><li><a href="/lkml/2006/4/19/363">Keith Owens</a><ul><li><a href="/lkml/2006/4/20/5">Keshavamurthy Anil S</a></li></ul></li></ul></li><li><a href="/lkml/2006/4/19/331">Anil S Keshavamurthy</a></li><li><a href="/lkml/2006/4/19/354">Keith Owens</a><ul><li><a href="/lkml/2006/4/20/4">Keshavamurthy Anil S</a></li></ul></li></ul></li></ul><div class="threadlist">Patch in this message</div><ul class="threadlist"><li><a href="/lkml/diff/2006/4/19/368/1">Get diff 1</a></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, 20 Apr 2006 09:27:35 +0530</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Ananth N Mavinakayanahalli <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: [(resend)patch 7/7] Kprobes - Register for page fault notify on active probes</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">On Wed, Apr 19, 2006 at 03:14:26PM -0700, Anil S Keshavamurthy wrote: > With this patch Kprobes now registers for page fault notifications only > when their is an active probe registered. Once all the active probes are > unregistered their is no need to be notified of page faults and kprobes > unregisters itself from the page fault notifications. Hence we > will have ZERO side effects when no probes are active. This patch isn't complete yet... comments below. > Signed-off-by: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com> > --- > kernel/kprobes.c | 25 +++++++++++++++++-------- > 1 file changed, 17 insertions(+), 8 deletions(-) > > Index: linux-2.6.17-rc1-mm3/kernel/kprobes.c > =================================================================== > --- linux-2.6.17-rc1-mm3.orig/kernel/kprobes.c > +++ linux-2.6.17-rc1-mm3/kernel/kprobes.c > @@ -47,11 +47,17 @@ > > static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; > static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE]; > +static atomic_t kprobe_count; > > DEFINE_MUTEX(kprobe_mutex); /* Protects kprobe_table */ > DEFINE_SPINLOCK(kretprobe_lock); /* Protects kretprobe_inst_table */ > static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; > > +static struct notifier_block kprobe_page_fault_nb = { > + .notifier_call = kprobe_exceptions_notify, > + .priority = 0x7fffffff /* we need to notified first */ > +}; > + > #ifdef __ARCH_WANT_KPROBES_INSN_SLOT > /* > * kprobe->ainsn.insn points to the copy of the instruction to be > @@ -464,6 +470,8 @@ static int __kprobes __register_kprobe(s > old_p = get_kprobe(p->addr); > if (old_p) { > ret = register_aggr_kprobe(old_p, p); > + if (!ret) > + atomic_inc(&kprobe_count); > goto out; > } > > @@ -474,6 +482,9 @@ static int __kprobes __register_kprobe(s > hlist_add_head_rcu(&p->hlist, > &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); > > + if(atomic_add_return(1, &kprobe_count) == 2) ^^^ if (..) please, here and elsewhere This will not work as desired for i386 (i386 no longer has a kprobe on the trampoline) and sparc64 (no retprobe support). > + register_page_fault_notifier(&kprobe_page_fault_nb); > + > arch_arm_kprobe(p); > > out: > @@ -523,9 +534,13 @@ valid_p: > cleanup_p = 0; > } > > + if(atomic_add_return(-1, &kprobe_count) == 1) > + unregister_page_fault_notifier(&kprobe_page_fault_nb); Same here - i386 and sparc64 need different checks. > + else > + synchronize_rcu(); Why has this changed from synchronize_sched()? This *must* be synchronize_sched() since, by definition it'll ensure that all non-preemptive sections are completed. In contrast, synchronize_rcu() will just enure rcu_read_lock() sections are complete. As of now, synchronize_sched() is aliased to synchronize_rcu() but I am told its scheduled to change in the future. Please revert this back to synchronize_sched(). Thanks, Ananth > + > mutex_unlock(&kprobe_mutex); > > - synchronize_sched(); > if (p->mod_refcounted && > (mod = module_text_address((unsigned long)p->addr))) > module_put(mod); > @@ -544,10 +559,6 @@ static struct notifier_block kprobe_exce > .priority = 0x7fffffff /* we need to notified first */ > }; > > -static struct notifier_block kprobe_page_fault_nb = { > - .notifier_call = kprobe_exceptions_notify, > - .priority = 0x7fffffff /* we need to notified first */ > -}; > > int __kprobes register_jprobe(struct jprobe *jp) > { > @@ -654,14 +665,12 @@ static int __init init_kprobes(void) > INIT_HLIST_HEAD(&kprobe_table[i]); > INIT_HLIST_HEAD(&kretprobe_inst_table[i]); > } > + atomic_set(&kprobe_count, 0); > > err = arch_init_kprobes(); > if (!err) > err = register_die_notifier(&kprobe_exceptions_nb); > > - if (!err) > - err = register_page_fault_notifier(&kprobe_page_fault_nb); > - > return err; > } > > > -- - 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: 2006-04-20 06:01 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>

CINXE.COM

LKML: Ananth N Mavinakayanahalli: Re: [(resend)patch 7/7] Kprobes - Register for page fault notify on active probes