Fix crash on "quit" + TUI + remote + async

[Pedro Alves <pedro at codesourcery dot com>]

I was debugging against a remote stub in async mode, using the TUI, and I
issued "quit" while still connected, in the hopes of bailing out of GDB
without much fuss, but, I ran into this SEGV:

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fcf597fb6e0 (LWP 13869)]
0x0000000000552f5e in serial_can_async_p (scb=0x0) at ../../src/gdb/serial.c:504
504       return (scb->ops->async != NULL);
(top-gdb) bt
#0  0x0000000000552f5e in serial_can_async_p (scb=0x0) at ../../src/gdb/serial.c:504
During symbol reading, unsupported tag: 'DW_TAG_const_type'.
#1  0x000000000042ffa3 in remote_can_async_p () at ../../src/gdb/remote.c:8207
During symbol reading, DW_AT_type missing from DW_TAG_subrange_type.
#2  0x000000000040d740 in execute_command (p=0x7fff618212a0 "set width 118", from_tty=0) at ../../src/gdb/top.c:382
#3  0x0000000000457d1c in tui_update_gdb_sizes () at ../../src/gdb/tui/tui-win.c:468
#4  0x0000000000452002 in tui_disable () at ../../src/gdb/tui/tui.c:455
#5  0x0000000000452581 in tui_exit () at ../../src/gdb/tui/tui-interp.c:45
#6  0x00007fcf589e0110 in exit () from /lib/libc.so.6
#7  0x000000000040edeb in quit_force (args=0x0, from_tty=1) at ../../src/gdb/top.c:1292
#8  0x0000000000443f20 in quit_command (args=0x0, from_tty=1) at ../../src/gdb/cli/cli-cmds.c:315
#9  0x000000000043e660 in do_cfunc (c=0xa344b0, args=0x0, from_tty=1) at ../../src/gdb/cli/cli-decode.c:60
#10 0x00000000004412a3 in cmd_func (cmd=0xa344b0, args=0x0, from_tty=1) at ../../src/gdb/cli/cli-decode.c:1680
#11 0x000000000040d9df in execute_command (p=0x9e81c1 "", from_tty=1) at ../../src/gdb/top.c:465
#12 0x00000000004c19ea in command_handler (command=0x9e81c0 "q") at ../../src/gdb/event-top.c:514

As you can see from from #2, disabling the tui ends up
executing a command.  At this point, the remote target was still
pushed on the stack, but it had already been closed, hence the remote
connection was closed, and remote_desc was == NULL.

There are target_can_async_p call around every command invocation,
and since the remote target was still on the stack, remote_can_async_p
was called, but it isn't expecting to be reached with remote_desc == NULL,
as most of remote's code:

 static int
 remote_can_async_p (void)
 {
   if (!remote_async_permitted)
     /* We only enable async when the user specifically asks for it.  */
     return 0;
 
   /* We're async whenever the serial device is.  */
   return remote_async_mask_value && serial_can_async_p (remote_desc);
 }

The issue is that quit_target closes the current target, but doesn't
pop it:

 static int
 quit_target (void *arg)
 {
   struct qt_args *qt = (struct qt_args *)arg;
 
   if (! ptid_equal (inferior_ptid, null_ptid) && target_has_execution)
     {
       if (attach_flag)
         target_detach (qt->args, qt->from_tty);
       else
         target_kill ();
     }

   /* UDI wants this, to kill the TIP.  */
   target_close (&current_target, 1);

   /* Save the history information if it is appropriate to do so.  */
   if (write_history_p && history_filename)
     write_history (history_filename);

   do_final_cleanups (ALL_CLEANUPS);	/* Do any final cleanups before exiting */

   return 0;
 }


The UDI target seems to be gone from the sources, but, even if it's still
here and I missed it, this is broken.  It should at best be a pop_target.  But,
while I'm at it, if the current target is of stratum > process_stratum, this
will still leave the process_stratum target open, which is much
more likelly to need closing.  So, I've come up with a new
pop_all_targets method and used it instead.

Tested against x86_64-unknown-linux-gnu native and gdbserver.

OK?

-- 
Pedro Alves

2008-07-28  Pedro Alves  <pedro@codesourcery.com>

	* target.h (pop_all_targets): Declare.
	* target.c (pop_all_targets): New.
	* top.c (quit_target): Pop all targets instead of just closing the
	current.

---
 gdb/target.c |   16 ++++++++++++++++
 gdb/target.h |    7 +++++++
 gdb/top.c    |    5 +++--
 3 files changed, 26 insertions(+), 2 deletions(-)

Index: src/gdb/target.c
===================================================================
--- src.orig/gdb/target.c	2008-07-28 16:24:06.000000000 +0100
+++ src/gdb/target.c	2008-07-28 16:30:02.000000000 +0100
@@ -820,6 +820,22 @@ pop_target (void)
   internal_error (__FILE__, __LINE__, _("failed internal consistency check"));
 }
 
+void
+pop_all_targets (int quitting)
+{
+  while ((int) (current_target.to_stratum) > (int) dummy_stratum)
+    {
+      target_close (&current_target, quitting);
+      if (unpush_target (target_stack) == 1)
+	return;
+    }
+
+  fprintf_unfiltered (gdb_stderr,
+		      "pop_all_targets couldn't find target %s\n",
+		      current_target.to_shortname);
+  internal_error (__FILE__, __LINE__, _("failed internal consistency check"));
+}
+
 /* Using the objfile specified in OBJFILE, find the address for the
    current thread's thread-local storage with offset OFFSET.  */
 CORE_ADDR
Index: src/gdb/target.h
===================================================================
--- src.orig/gdb/target.h	2008-07-28 16:24:06.000000000 +0100
+++ src/gdb/target.h	2008-07-28 16:30:02.000000000 +0100
@@ -1172,6 +1172,13 @@ extern void target_preopen (int);
 
 extern void pop_target (void);
 
+/* Does whatever cleanup is required to get rid of all pushed targets.
+   QUITTING is propagated to target_close; it indicates that GDB is
+   exiting and should not get hung on an error (otherwise it is
+   important to perform clean termination, even if it takes a
+   while).  */
+extern void pop_all_targets (int quitting);
+
 extern CORE_ADDR target_translate_tls_address (struct objfile *objfile,
 					       CORE_ADDR offset);
 
Index: src/gdb/top.c
===================================================================
--- src.orig/gdb/top.c	2008-07-28 16:24:06.000000000 +0100
+++ src/gdb/top.c	2008-07-28 16:30:48.000000000 +0100
@@ -1229,8 +1229,9 @@ quit_target (void *arg)
         target_kill ();
     }
 
-  /* UDI wants this, to kill the TIP.  */
-  target_close (&current_target, 1);
+  /* Give all pushed targets a chance to do minimal cleanup, and pop
+     them all out.  */
+  pop_all_targets (1);
 
   /* Save the history information if it is appropriate to do so.  */
   if (write_history_p && history_filename)