From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 65106 invoked by alias); 9 Jun 2015 18:10:56 -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 65097 invoked by uid 89); 9 Jun 2015 18:10:55 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-2.4 required=5.0 tests=AWL,BAYES_00,RCVD_IN_DNSWL_LOW,SPF_PASS autolearn=ham version=3.3.2 X-HELO: relay1.mentorg.com Received: from relay1.mentorg.com (HELO relay1.mentorg.com) (192.94.38.131) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with ESMTP; Tue, 09 Jun 2015 18:10:54 +0000 Received: from svr-orw-fem-03.mgc.mentorg.com ([147.34.97.39]) by relay1.mentorg.com with esmtp id 1Z2Nz8-00046q-9j from Luis_Gustavo@mentor.com ; Tue, 09 Jun 2015 11:10:50 -0700 Received: from [172.30.7.56] (147.34.91.1) by svr-orw-fem-03.mgc.mentorg.com (147.34.97.39) with Microsoft SMTP Server id 14.3.224.2; Tue, 9 Jun 2015 11:10:49 -0700 Message-ID: <55772C14.5090501@codesourcery.com> Date: Tue, 09 Jun 2015 18:10:00 -0000 From: Luis Machado Reply-To: Luis Machado User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.7.0 MIME-Version: 1.0 To: Pedro Alves , Subject: Re: [PATCH] Fix problems with finishing a dummy function call on simulators. References: <1433862056-18237-1-git-send-email-lgustavo@codesourcery.com> <55772797.802@redhat.com> In-Reply-To: <55772797.802@redhat.com> Content-Type: text/plain; charset="windows-1252"; format=flowed Content-Transfer-Encoding: 7bit X-IsSubscribed: yes X-SW-Source: 2015-06/txt/msg00140.txt.bz2 On 06/09/2015 02:51 PM, Pedro Alves wrote: > On 06/09/2015 04:00 PM, Luis Machado wrote: >> This is in line with what was done by Joel's patch here: >> >> https://sourceware.org/ml/gdb-patches/2014-11/msg00478.html >> >> And it also answers Pedro's question about whether this is specific to SPARC >> QEMU or not. This indeed seems to affect multiple QEMU targets and also other >> simulators (proprietary). > > Sounds like a different issue, although related. > >> >> I ran into this weird issue of not being able to "finish" an inferior function >> call. It looks as if the program is running away, but it really is stuck >> somewhere. "finish" still works fine for regular functions not called manually >> by GDB. > > Sounds like that would fail on SPARC qemu as well. > >> >> I tracked this failure down to GDB having both a bp_call_dummy and bp_finish in >> its breakpoint list. As a result of one not being considered permanent and the >> other considered permanent, GDB will not issue a Z packet to force the insertion >> of that location's breakpoint, confusing the simulator that does not know how >> to deal properly with these permanent breakpoints that GDB inserted beforehand. >> >> The attached patch fixes this, though i'm inclined to say we could probably >> check if both bp_call_dummy and bp_finish are present and force the >> insertion of that location's breakpoint. It isn't clear to me where exactly that >> check would go or if it would be cleaner than checking that information in >> the same function Joel used. >> >> I see no regressions on x86-64 and it fixes a bunch of failures for simulator >> targets we use (MIPS and PowerPC to name two). > > If it happens that you "finish" from a normal function, and the finish > breakpoint ends up on top of a real permanent breakpoint, then this patch > will make us end up inserting a breakpoint on top of that permanent > breakpoint. I don't see what's special about finish breakpoints; > it's the address (dummy breakpoint location) that is special. It very much > sounds like that any kind of breakpoint that is placed on top of the dummy > breakpoint ends up with the same issue. E.g., if you stepi out of > the called function, with a software single-step breakpoint, sounds like > GDB will miss inserting the software step breakpoint because that's > at the same address as the dummy breakpoint. Yes, i meant breakpoints as the address themselves, so a location. It is probably the case that using permanent breakpoints and mixing them with other types of non-permanent breakpoints is causing issues, though the only well-exercised testcase is the finish-after-dummy-call one. I do recall once getting stuck with a stepi inside a dummy function call, so i may have hit what you suggested here. > > As a data point, I assume that GDB is considering the non-permanent > dummy breakpoint a duplicate of the permanent finish breakpoint and > then none ends up inserted. Is that right? That is correct. And one of them is already considered inserted. > > Not exactly sure what to do here. Maybe we should stop considering > permanent and non-permanent breakpoints at the same address as > duplicates. That should result in GDB inserting the non-permanent > one, I think. Or we could get stop marking permanent breakpoints > as always inserted, and let normal breakpoints insert on top of > permanent breakpoints normally. See also: > https://sourceware.org/ml/gdb-patches/2015-03/msg00174.html That sounds a bit hacky. Doesn't that defeat the purpose of having permanent breakpoints in the first place? It looks like non-gdbserver targets are not ready to support these tricky constructs/optimizations unfortunately. I'm afraid adding more hacks here and there will cause the code to get even more confusing without a generous amount of code comments. And i'm not even sure the bp_finish check is the best solution either. After all, there is the stepi case too. We could probably fix the simulators, but then again there are proprietary ones we cannot easily fix.