From: Pedro Alves <palves@redhat.com>
Date: Wed, 14 Jan 2009 15:00:19 +0000 (+0000)
Subject: 	* remote.c (extended_remote_mourn_1): Invalidate our notion of
X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=dc1981d76992dfb9e56a6425baff2caf7eefc86e;p=binutils-gdb.git

	* remote.c (extended_remote_mourn_1): Invalidate our notion of
	current general thread.
---

diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index caff8a8ee54..c9c46166928 100644
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,3 +1,8 @@
+2009-01-14  Pedro Alves  <pedro@codesourcery.com>
+
+	* remote.c (extended_remote_mourn_1): Invalidate our notion of
+	current general thread.
+
 2009-01-14  Pedro Alves  <pedro@codesourcery.com>
 
 	* mi/mi-main.c (mi_cmd_execute): Also allow -list-thread-groups
diff --git a/gdb/remote.c b/gdb/remote.c
index 92d00c29034..4864dde3845 100644
--- a/gdb/remote.c
+++ b/gdb/remote.c
@@ -6548,6 +6548,31 @@ extended_remote_mourn_1 (struct target_ops *target)
   /* We're no longer interested in these events.  */
   discard_pending_stop_replies (ptid_get_pid (inferior_ptid));
 
+  /* If the current general thread belonged to the process we just
+     detached from or has exited, the remote side current general
+     thread becomes undefined.  Considering a case like this:
+
+     - We just got here due to a detach.
+     - The process that we're detaching from happens to immediately
+       report a global breakpoint being hit in non-stop mode, in the
+       same thread we had selected before.
+     - GDB attaches to this process again.
+     - This event happens to be the next event we handle.
+
+     GDB would consider that the current general thread didn't need to
+     be set on the stub side (with Hg), since for all it knew,
+     GENERAL_THREAD hadn't changed.
+
+     Notice that although in all-stop mode, the remote server always
+     sets the current thread to the thread reporting the stop event,
+     that doesn't happen in non-stop mode; in non-stop, the stub *must
+     not* change the current thread when reporting a breakpoint hit,
+     due to the decoupling of event reporting and event handling.
+
+     To keep things simple, we always invalidate our notion of the
+     current thread.  */
+  record_currthread (minus_one_ptid);
+
   /* Unlike "target remote", we do not want to unpush the target; then
      the next time the user says "run", we won't be connected.  */