From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 10567 invoked by alias); 7 Jun 2004 12:12:44 -0000 Mailing-List: contact gdb-patches-help@sources.redhat.com; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sources.redhat.com Received: (qmail 10204 invoked from network); 7 Jun 2004 12:12:15 -0000 Received: from unknown (HELO miranda.se.axis.com) (212.209.10.220) by sourceware.org with SMTP; 7 Jun 2004 12:12:15 -0000 Received: from axis.com (ironmadien.se.axis.com [10.84.130.1]) by miranda.se.axis.com (8.12.9/8.12.9/Debian-5local0.1) with ESMTP id i57CC5dB009777; Mon, 7 Jun 2004 14:12:06 +0200 Message-ID: <40C45B95.9050309@axis.com> Date: Mon, 07 Jun 2004 12:12:00 -0000 From: Orjan Friberg Organization: Axis Communications User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.6) Gecko/20040113 MIME-Version: 1.0 To: Andrew Cagney CC: gdb-patches@sources.redhat.com Subject: Re: STEP_SKIPS_DELAY question, sort of References: <40AE38D0.7010204@axis.com> <40AE659A.90207@gnu.org> <40B1BD1B.4090300@axis.com> <40B23BB2.6070001@gnu.org> <40B33399.3090803@axis.com> <40B3B742.50007@gnu.org> <40B465E7.7050702@axis.com> In-Reply-To: <40B465E7.7050702@axis.com> Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit X-SW-Source: 2004-06/txt/msg00115.txt.bz2 Orjan Friberg wrote: > > I'll go and have a closer look at STEP_SKIPS_DELAY. Short recap: the MIPS checks if there's a breakpoint in the delay slot of the instruction where GDB is about to resume execution. If so, GDB single-steps that instruction before re-inserting the breakpoint. For CRISv32, if there's a breakpoint on an instruction which has a delay slot, we cannot re-insert the breakpoint when we're stopped in the delay slot (after making one single-step), because the previous instruction will be re-executed when returning from the delay slot. I tried implementing a fix in handle_inferior_event (the MIPS fix is in proceed). It seemed easier to step one more time when we find out that we need to, rather than to determine beforehand that we're going to have to step twice (and I couldn't determine how to pass that information). The concept patch below illustrates what I'm trying to do; by setting another_trap in the execution control state, by the time we get to keep_going we won't insert the breakpoint and we'll instead continue single-stepping. Comments? Is this the right approach at all? Index: infrun.c =================================================================== RCS file: /cvs/src/src/gdb/infrun.c,v retrieving revision 1.156 diff -u -r1.156 infrun.c --- infrun.c 11 May 2004 23:30:31 -0000 1.156 +++ infrun.c 7 Jun 2004 11:44:19 -0000 @@ -1975,7 +1975,17 @@ /* Don't even think about breakpoints if just proceeded over a breakpoint. */ if (stop_signal == TARGET_SIGNAL_TRAP && trap_expected) - bpstat_clear (&stop_bpstat); + { + bpstat_clear (&stop_bpstat); + + /* If we stepped into a delay slot, and the preceding instruction + will be re-executed when resuming, step again before re-inserting + the breakpoint. */ + if (STEP_SKIPS_IN_DELAY_P + && breakpoint_here_p (read_pc () - 2) + && STEP_SKIPS_IN_DELAY (read_pc ())) + ecs->another_trap = 1; + } else { /* See if there is a breakpoint at the current PC. */ -- Orjan Friberg Axis Communications