From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 16052 invoked by alias); 5 May 2006 16:42:49 -0000 Received: (qmail 16030 invoked by alias); 5 May 2006 16:42:42 -0000 Date: Fri, 05 May 2006 16:42:00 -0000 Message-ID: <20060505164242.16029.qmail@sourceware.org> From: "hunt at redhat dot com" To: systemtap@sources.redhat.com In-Reply-To: <20060502201312.2637.hunt@redhat.com> References: <20060502201312.2637.hunt@redhat.com> Reply-To: sourceware-bugzilla@sourceware.org Subject: [Bug kprobes/2637] skipped probes in FC6 X-Bugzilla-Reason: AssignedTo Mailing-List: contact systemtap-help@sourceware.org; run by ezmlm Precedence: bulk List-Subscribe: List-Post: List-Help: , Sender: systemtap-owner@sourceware.org X-SW-Source: 2006-q2/txt/msg00320.txt.bz2 ------- Additional Comments From hunt at redhat dot com 2006-05-05 16:42 ------- Subject: Re: skipped probes in FC6 On Fri, 2006-05-05 at 06:56 +0000, prasanna at in dot ibm dot com wrote: > ------- Additional Comments From prasanna at in dot ibm dot com 2006-05-05 06:56 ------- > The example given by Martin, accesses the data in user address space. Most > likely the data accessed from user address space may not be present in memory. True. But such situations have in the past been extremely rare. What is it about the new kprobes that makes it now so common and predictable? We need to understand this better. > It is known that such accesses causes page_faults, where in the > fixup_exception() may not succeed most of the times, hence nmissed count goes up > (even with the current above patch). Could you explain what would cause such an event? It does not happen in my experience and I don't see how it could. If there is an entry in the fixup table, which there always should be if you use a proper userspace copy function, fixup exception will work. > Is it acceptable, if the probe handler fails most of the time, while accessing > arguments from user address space. OR do we need to find some other way to > get the agruments of the syscalls The current situation fails too often and is completely unacceptable. I cannot even run my simple demo programs. If we absolutely cannot fix the problem, then we must as a minimum change user_string() to just return "" or "" and not generate errors or warnings when copies fail. Martin -- http://sourceware.org/bugzilla/show_bug.cgi?id=2637 ------- You are receiving this mail because: ------- You are the assignee for the bug, or are watching the assignee.