[commit] Fix another corruption in remote.c

[Daniel Jacobowitz <drow at false dot org>]

I'm still glad I switched remote.c away from using alloca - it really needed
doing - but I keep running into unforeseen problems :-(  I've done as
thorough an audit as I can, and this is the only remaining one I know of.
The problem arises in set_thread:

    xsnprintf (&buf[2], get_remote_packet_size () - 2, "-%x", -th);

The call to get_remote_packet_size fetches the architecture-specific gdbarch
data.  Because the gdbarch data mechanism isn't set up to initialize
immediately on creation of the gdbarch, but lazily at first access, which
this is after remote_open_1 is called, this line triggers an unexpected call
to init_remote_state which in turn may change rs->buf.

So, let's make sure that when we get access to rs->buf the architecture is
always set up.

Tested x86_64-pc-linux-gnu and committed.

-- 
Daniel Jacobowitz
CodeSourcery

2006-10-05  Daniel Jacobowitz  <dan@codesourcery.com>

	* remote.c (get_remote_state_raw): Renamed from get_remote_state.
	(get_remote_state): New function.
	(init_remote_state, _initialize_remote): Use get_remote_state_raw.

Index: remote.c
===================================================================
RCS file: /cvs/src/src/gdb/remote.c,v
retrieving revision 1.230
diff -u -p -r1.230 remote.c
--- remote.c	22 Sep 2006 13:50:36 -0000	1.230
+++ remote.c	5 Oct 2006 19:14:01 -0000
@@ -244,7 +244,7 @@ struct remote_state
 static struct remote_state remote_state;
 
 static struct remote_state *
-get_remote_state (void)
+get_remote_state_raw (void)
 {
   return &remote_state;
 }
@@ -294,11 +294,26 @@ get_remote_arch_state (void)
   return gdbarch_data (current_gdbarch, remote_gdbarch_data_handle);
 }
 
+/* Fetch the global remote target state.  */
+
+static struct remote_state *
+get_remote_state (void)
+{
+  /* Make sure that the remote architecture state has been
+     initialized, because doing so might reallocate rs->buf.  Any
+     function which calls getpkt also needs to be mindful of changes
+     to rs->buf, but this call limits the number of places which run
+     into trouble.  */
+  get_remote_arch_state ();
+
+  return get_remote_state_raw ();
+}
+
 static void *
 init_remote_state (struct gdbarch *gdbarch)
 {
   int regnum;
-  struct remote_state *rs = get_remote_state ();
+  struct remote_state *rs = get_remote_state_raw ();
   struct remote_arch_state *rsa;
 
   rsa = GDBARCH_OBSTACK_ZALLOC (gdbarch, struct remote_arch_state);
@@ -6141,7 +6156,7 @@ _initialize_remote (void)
      of these, not one per target.  Only one target is active at a
      time.  The default buffer size is unimportant; it will be expanded
      whenever a larger buffer is needed.  */
-  rs = get_remote_state ();
+  rs = get_remote_state_raw ();
   rs->buf_size = 400;
   rs->buf = xmalloc (rs->buf_size);