From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 14591 invoked by alias); 27 Jun 2008 23:11:28 -0000 Received: (qmail 14580 invoked by uid 22791); 27 Jun 2008 23:11:26 -0000 X-Spam-Check-By: sourceware.org Received: from mail.codesourcery.com (HELO mail.codesourcery.com) (65.74.133.4) by sourceware.org (qpsmtpd/0.31) with ESMTP; Fri, 27 Jun 2008 23:11:08 +0000 Received: (qmail 25957 invoked from network); 27 Jun 2008 23:11:06 -0000 Received: from unknown (HELO orlando.local) (pedro@127.0.0.2) by mail.codesourcery.com with ESMTPA; 27 Jun 2008 23:11:06 -0000 From: Pedro Alves To: gdb-patches@sourceware.org Subject: [gdbserver] Fix attaching notices Date: Fri, 27 Jun 2008 23:26:00 -0000 User-Agent: KMail/1.9.9 MIME-Version: 1.0 Content-Type: Multipart/Mixed; boundary="Boundary-00=_QOXZI/uY9NeO06t" Message-Id: <200806280011.12868.pedro@codesourcery.com> X-IsSubscribed: yes 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 X-SW-Source: 2008-06/txt/msg00523.txt.bz2 --Boundary-00=_QOXZI/uY9NeO06t Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Content-Disposition: inline Content-length: 3331 Hi, This patch fixes several issues I noticed with attaching to processes with gdbserver's --multi support, on several host/target combinations: Linux host -> Linux gdbserver There's code in gdbserver to pretend the initial SIGSTOP was a SIGTRAP, that's been put there so "gdbserver --attach PID"; "target remote", doesn't see the SIGSTOP. This smudging should not be done when handling a vAttach, otherwise, the user is presented with: (gdb) tar extended-remote :9999 Remote debugging using :9999 (gdb) attach 32762 Attached to Thread 32762 [New Thread 32762] Program received signal SIGTRAP, Trace/breakpoint trap. 0x00007fc30f5d0b30 in ?? () ... a bogus SIGTRAP. Linux host -> Windows gdbserver - On Windows, the signal we get while attaching when the inferior gets stopped is really a SIGTRAP. In this case, we do want do convert it to a SIGSTOP, so GDB suppresses it. - Due to the fact that ATTACH_NO_WAIT is a host macro (it shouldn't, and I'm posting a patch next to fix that), and it being defined in Windows hosts, the T stop reply packet gdbserver sends after attaching is just ignored on the GDB side. On most other hosts (except hurd), ATTACH_NO_WAIT is not defined, which reveals a buglet that had gotten unnoticed. (gdb) tar extended-remote sandra:9999 Remote debugging using sandra:9999 (gdb) attach 988 Attached to Thread 988 [New Thread 2232] Program received signal 0, Signal 0. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Error while mapping shared library sections: /cygdrive/c/WINDOWS/system32/ntdll.dll: No such file or directory. Error while mapping shared library sections: /cygdrive/c/WINDOWS/system32/kernel32.dll: No such file or directory. ... /usr/bin/cygiconv-2.dll: No such file or directory. Symbol file not found for /cygdrive/c/WINDOWS/system32/ntdll.dll Symbol file not found for /cygdrive/c/WINDOWS/system32/kernel32.dll ... [Switching to Thread 2232] Stopped due to shared library event ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ On initial connections, we're preventing gdbserver to report dll changes in the T stop reply packet, because GDB will always query the dll list anyway. We missed doing that on vAttach handling. The "signal 0" notice happens even though gdbserver sends T05, because the stop reason is transformed into a TARGET_WAITKIND_LOADED, and the signal number sent isn't really parsed (from remote.c): if (solibs_changed) status->kind = TARGET_WAITKIND_LOADED; else { status->kind = TARGET_WAITKIND_STOPPED; status->value.sig = (enum target_signal) (((fromhex (buf[1])) << 4) + (fromhex (buf[2]))); Windows host -> Windows gdbserver Due to ATTACH_NO_WAIT being defined, the T stop reply gdbserver sends on attaches is just being ignored. That's why the "stopped by library event" notice bug went unnoticed. I had worked on the Windows port of --multi for gdbserver from a Windows host :-) I'm posting a patch next that gets rid of ATTACH_NO_WAIT, which then triggers the same "problems" as seen while attaching to a Windows gdbserver from a Linux host. Windows host -> Linux gdbserver Same rationale as above. Tested manually on all those combinations, and regression tested with local gdbservers on x86_64-unknown-linux-gnu, x86-pc-cygwin. OK? -- Pedro Alves --Boundary-00=_QOXZI/uY9NeO06t Content-Type: text/x-diff; charset="utf-8"; name="gdbserver_attach_notices.diff" Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename="gdbserver_attach_notices.diff" Content-length: 2917 2008-06-28 Pedro Alves * server.c (attach_inferior): Add vattach parameter. If attaching due to a vAttach, don't hide a SIGSTOP, and, if attaching finishes with a SIGTRAP, pretend it was a SIGSTOP. (handle_v_attach): Pass 1 as vattach parameter to attach_inferior. Don't report dll changes. (main): Pass 0 as vattach parameter to attach_inferior. --- gdb/gdbserver/server.c | 25 ++++++++++++++++++------- 1 file changed, 18 insertions(+), 7 deletions(-) Index: src/gdb/gdbserver/server.c =================================================================== --- src.orig/gdb/gdbserver/server.c +++ src/gdb/gdbserver/server.c @@ -160,7 +160,7 @@ start_inferior (char **argv, char *statu } static int -attach_inferior (int pid, char *statusptr, int *sigptr) +attach_inferior (int pid, char *statusptr, int *sigptr, int vattach) { /* myattach should return -1 if attaching is unsupported, 0 if it succeeded, and call error() otherwise. */ @@ -180,11 +180,17 @@ attach_inferior (int pid, char *statuspt *sigptr = mywait (statusptr, 0); - /* GDB knows to ignore the first SIGSTOP after attaching to a running - process using the "attach" command, but this is different; it's - just using "target remote". Pretend it's just starting up. */ - if (*statusptr == 'T' && *sigptr == TARGET_SIGNAL_STOP) + /* GDB knows to ignore the first SIGSTOP after attaching to a + running process using the "attach" command, but if we're + attaching due to --attach this is different; it's just using + "target remote". Pretend it's just starting up. */ + if (!vattach && *statusptr == 'T' && *sigptr == TARGET_SIGNAL_STOP) *sigptr = TARGET_SIGNAL_TRAP; + /* Similarly, on some targets, attaching finishes with a SIGTRAP + (e.g., on Windows it's called the initial breakpoint). Pretend + it was a SIGSTOP, so GDB hushes it. */ + else if (vattach && *statusptr == 'T' && *sigptr == TARGET_SIGNAL_TRAP) + *sigptr = TARGET_SIGNAL_STOP; return 0; } @@ -1011,8 +1017,13 @@ handle_v_attach (char *own_buf, char *st int pid; pid = strtol (own_buf + 8, NULL, 16); - if (pid != 0 && attach_inferior (pid, status, signal) == 0) + if (pid != 0 && attach_inferior (pid, status, signal, 1) == 0) { + /* Don't report shared library events after attaching, even if + some libraries are preloaded. GDB will always poll the + library list. Avoids the "stopped by shared library event" + notice on the GDB side. */ + dlls_changed = 0; prepare_resume_reply (own_buf, *status, *signal); return 1; } @@ -1336,7 +1347,7 @@ main (int argc, char *argv[]) } else if (pid != 0) { - if (attach_inferior (pid, &status, &signal) == -1) + if (attach_inferior (pid, &status, &signal, 0) == -1) error ("Attaching not supported on this target"); /* Otherwise succeeded. */ --Boundary-00=_QOXZI/uY9NeO06t--