From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 1854 invoked by alias); 19 Mar 2018 14:29:02 -0000 Mailing-List: contact gdb-patches-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sourceware.org Received: (qmail 1833 invoked by uid 89); 19 Mar 2018 14:29:01 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-24.3 required=5.0 tests=AWL,BAYES_00,GIT_PATCH_0,GIT_PATCH_1,GIT_PATCH_2,GIT_PATCH_3,KAM_LAZY_DOMAIN_SECURITY,T_RP_MATCHES_RCVD autolearn=ham version=3.3.2 spammy= X-HELO: mx1.redhat.com Received: from mx3-rdu2.redhat.com (HELO mx1.redhat.com) (66.187.233.73) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with ESMTP; Mon, 19 Mar 2018 14:29:00 +0000 Received: from smtp.corp.redhat.com (int-mx03.intmail.prod.int.rdu2.redhat.com [10.11.54.3]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id 63565406E970; Mon, 19 Mar 2018 14:28:58 +0000 (UTC) Received: from [127.0.0.1] (ovpn04.gateway.prod.ext.ams2.redhat.com [10.39.146.4]) by smtp.corp.redhat.com (Postfix) with ESMTP id A762910B2B50; Mon, 19 Mar 2018 14:28:57 +0000 (UTC) Subject: Re: [RFA] (x86) Fix watchpoint using hardware breakpoint for some distro To: Xavier Roirand , gdb-patches@sourceware.org References: <1521209212-11264-1-git-send-email-roirand@adacore.com> Cc: brobecker@adacore.com From: Pedro Alves Message-ID: Date: Mon, 19 Mar 2018 14:29:00 -0000 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.6.0 MIME-Version: 1.0 In-Reply-To: <1521209212-11264-1-git-send-email-roirand@adacore.com> Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit X-SW-Source: 2018-03/txt/msg00354.txt.bz2 Hi Xavier, On 03/16/2018 02:06 PM, Xavier Roirand wrote: > Running watch*.exp tests in gdb.base shows this: > > on x86_64/Ubuntu 16.04: > > # of expected passes 2631 > # of unexpected failures 0 > > on x86_64/Ubuntu CentOS 5.11: > > # of expected passes 2535 > # of unexpected failures 96 > > The problem can be easily shown in a debug session: > > (gdb) watch val > Hardware watchpoint 2: val > (gdb) c > Continuing. > Program received signal SIGTRAP, Trace/breakpoint trap. > ... > > Whereas it should be: > > (gdb) watch val > Hardware watchpoint 2: val > (gdb) c > Continuing. > val before change = 0 > > Hardware watchpoint 2: val > > Old value = ... > New value = ... > > The Linux target and gdbserver now check the siginfo si_code > reported on a SIGTRAP to detect whether the trap indicates > a hardware breakpoint was hit. > > Unfortunately, on some distro (CentOS 5, Suse 11) the returned > si_code value is not equal to TRAP_HWBKPT when a hardware breakpoint > is hit thus the hardware breakpoint is not handled as it should > be, which is also the case for watchpoint when based on hardware > breakpoint. A few things are missing here: #1 - kernel versions where this was observed. #2 - If it's not equal to TRAP_HWBKPT, then what's it equal to? I assume zero? Take a look at the big comment and table in nat/linux-ptrace.h -- is this is the only case that is different on these kernels? I think that we should update the table a bit here, at least something like: - | hardware breakpoints/watchpoints | TRAP_HWBKPT | + | hardware breakpoints/watchpoints | TRAP_HWBKPT (*) | (*) - Kernels x.y.z (CentOS 5, Suse 11) leave this as zero. If other cases are different, then that might affect how to best address this. > > This patch adds an additional check when the inferior reported > a SIGTRAP in order to detect this case. > > No test have been created since all the existing ones are enough > to validate the fix. BTW, with this fix, the tests results for > the watchpoint tests are (for CentOS 5.11): > > # of expected passes 2630 > # of unexpected failures 1 > > The remaining failure is located in watch-vfork test which explicitly > disable the use of hardware breakpoint which is out of scope here. > > gdb/ChangeLog: > > * linux-nat.c (save_stop_reason): Add an additional check > to detect hardware watchpoint. > > gdbserver/ChangeLog: > > * linux-low.c (save_stop_reason): Add an additional check > to detect hardware watchpoint. > > For R309-004 > > Test: x86_64/gdb /ubuntu 16.04 > x86_64/gdbserver/ubuntu 16.04 > x86 /gdbs /ubuntu 16.04 > x86 /gdbserver/ubuntu 16.04 > x86_64/gdb /centos 5.11 > x86_64/gdbserver/centos 5.11 > x86 /gdb /centos 5.11 > x86 /gdbserver/centos 5.11 > > Change-Id: I2546aca9827d9ae12ab86deb7aa4acc60c82b4b4 > --- > gdb/gdbserver/linux-low.c | 7 +++++++ > gdb/linux-nat.c | 7 +++++++ > 2 files changed, 14 insertions(+) > > diff --git a/gdb/gdbserver/linux-low.c b/gdb/gdbserver/linux-low.c > index 2e5e19d..fe61026 100644 > --- a/gdb/gdbserver/linux-low.c > +++ b/gdb/gdbserver/linux-low.c > @@ -866,6 +866,13 @@ save_stop_reason (struct lwp_info *lwp) > if (!check_stopped_by_watchpoint (lwp)) > lwp->stop_reason = TARGET_STOPPED_BY_SINGLE_STEP; > } > + else > + { > + /* The si_code is not TRAP_HWBKPT whereas it should > + on some distro (CentOS 5, Suse 11) so still check > + if stopped due to watchpoint. */ > + check_stopped_by_watchpoint (lwp); > + } > } > } > #else > diff --git a/gdb/linux-nat.c b/gdb/linux-nat.c > index 1bbad7b..0d97aa1 100644 > --- a/gdb/linux-nat.c > +++ b/gdb/linux-nat.c > @@ -2798,6 +2798,13 @@ save_stop_reason (struct lwp_info *lp) > the debug registers separately. */ > check_stopped_by_watchpoint (lp); > } > + else > + { > + /* The si_code is not TRAP_HWBKPT whereas it should > + on some distro (CentOS 5, Suse 11) so still check > + if stopped due to watchpoint. */ This comment only makes complete sense with the context in the git log in mind: - This code is run by all architectures, so the comment should mention x86. - The comment reads a bit backwards to me -- talks about what it should be before talking about watchpoints. I'd suggest something like this: /* On some kernels (such as x86-64 x.y.z/CentOS 5, x.y.z/Suse 11), when we continue into a watchpoint, si_code indicates 0 instead of TRAP_HWBKPT so we need to check debug registers separately. */ Does the step-into-watchpoint case result in TRAP_TRACE, or does that result in 0 too? That affects the "continue" in the comment above. Thanks, Pedro Alves