Make scoped_restore_current_thread's cdtors exception free (RFC)

If the remote target closes while we're reading registers/memory for
restoring the selected frame in scoped_restore_current_thread's dtor,
the corresponding TARGET_CLOSE_ERROR error is swallowed by the
scoped_restore_current_thread's dtor, because letting exceptions
escape from a dtor is bad.  It isn't great to lose that errors like
that, though.  I've been thinking about how to avoid it, and I came up
with this patch.

The idea here is to make scoped_restore_current_thread's dtor do as
little as possible, to avoid any work that might throw in the first
place.  And to do that, instead of having the dtor call
restore_selected_frame, which re-finds the previously selected frame,
just record the frame_id/level of the desired selected frame, and have
get_selected_frame find the frame the next time it is called.  In
effect, this implements most of Cagney's suggestion, here:

  /* On demand, create the selected frame and then return it.  If the
     selected frame can not be created, this function prints then throws
     an error.  When MESSAGE is non-NULL, use it for the error message,
     otherwize use a generic error message.  */
  /* FIXME: cagney/2002-11-28: At present, when there is no selected
     frame, this function always returns the current (inner most) frame.
     It should instead, when a thread has previously had its frame
     selected (but not resumed) and the frame cache invalidated, find
     and then return that thread's previously selected frame.  */
  extern struct frame_info *get_selected_frame (const char *message);

The only thing missing to fully implement that would be to make
reinit_frame_cache just clear selected_frame instead of calling
select_frame(NULL), and the call select_frame(NULL) explicitly in the
places where we really wanted reinit_frame_cache to go back to the
current frame too.  That can done separately, though, I'm not
proposing to do that in this patch.

Note that this patch renames restore_selected_frame to
lookup_selected_frame, and adds a new restore_selected_frame function
that doesn't throw, to be paired with the also-new save_selected_frame
function.

lookup_selected_frame should really move from thread.c to frame.c, but
I didn't do that here, just to avoid churn in the patch while it
collects comments.  I did make it extern and declared it in frame.h
already, preparing for the move.  I will do the move as a follow up
patch if people agree with this approach.

Incidentally, this patch alone would fix the crashes fixed by the
previous patches in the series, because with this,
scoped_restore_current_thread's constructor doesn't throw either.

gdb/ChangeLog:

	* blockframe.c (block_innermost_frame): Use get_selected_frame.
	* frame.c
	(scoped_restore_selected_frame::scoped_restore_selected_frame):
	Use save_selected_frame.  Save language as well.
	(scoped_restore_selected_frame::~scoped_restore_selected_frame):
	Use restore_selected_frame, and restore language as well.
	(selected_frame_id, selected_frame_level): New.
	(selected_frame): Update comments.
	(save_selected_frame, restore_selected_frame): New.
	(get_selected_frame): Use lookup_selected_frame.
	(get_selected_frame_if_set): Delete.
	(select_frame): Record selected_frame_level and selected_frame_id.
	* frame.h (scoped_restore_selected_frame) <m_level, m_lang>: New
	fields.
	(get_selected_frame_if_set): Delete declaration.
	(select_frame): Update comments.
	(save_selected_frame, restore_selected_frame)
	(lookup_selected_frame): Declare.
	* gdbthread.h (scoped_restore_current_thread) <m_lang>: New field.
	* stack.c (select_frame_command_core, frame_command_core): Use
	get_selected_frame.
	* thread.c (restore_selected_frame): Rename to ...
	(lookup_selected_frame): ... this and make extern.  Select the
	current frame if the frame level is -1.
	(scoped_restore_current_thread::restore): Also restore the
	language.
	(scoped_restore_current_thread::~scoped_restore_current_thread):
	Don't try/catch.
	(scoped_restore_current_thread::scoped_restore_current_thread):
	Save the language as well.  Use save_selected_frame.

gdb/blockframe.c

@@ -448,14 +448,10 @@ find_gnu_ifunc_target_type (CORE_ADDR resolver_funaddr)
 struct frame_info *
 block_innermost_frame (const struct block *block)
 {
-  struct frame_info *frame;
-
   if (block == NULL)
     return NULL;
 
-  frame = get_selected_frame_if_set ();
-  if (frame == NULL)
-    frame = get_current_frame ();
+  frame_info *frame = get_selected_frame (NULL);
   while (frame != NULL)
     {
       const struct block *frame_block = get_frame_block (frame, NULL);

gdb/dwarf2/frame-tailcall.c

@@ -377,7 +377,6 @@ dwarf2_tailcall_sniffer_first (struct frame_info *this_frame,
      get_frame_address_in_block will decrease it by 1 in such case.  */
   this_pc = get_frame_address_in_block (this_frame);
 
-  /* Catch any unwinding errors.  */
   try
     {
       int sp_regnum;
@@ -439,7 +438,22 @@ dwarf2_tailcall_sniffer_first (struct frame_info *this_frame,
     {
       if (entry_values_debug)
 	exception_print (gdb_stdout, except);
-      return;
+
+      switch (except.error)
+	{
+	case NO_ENTRY_VALUE_ERROR:
+	  /* Thrown by call_site_find_chain.  */
+	case MEMORY_ERROR:
+	case OPTIMIZED_OUT_ERROR:
+	case NOT_AVAILABLE_ERROR:
+	  /* These can normally happen when we try to access an
+	     optimized out or unavailable register, either in a
+	     physical register or spilled to memory.  */
+	  return;
+	}
+
+      /* Let unexpected errors propagate.  */
+      throw;
     }
 
   /* Ambiguous unwind or unambiguous unwind verified as matching.  */

gdb/frame.c

@@ -297,17 +297,15 @@ frame_stash_invalidate (void)
 /* See frame.h  */
 scoped_restore_selected_frame::scoped_restore_selected_frame ()
 {
-  m_fid = get_frame_id (get_selected_frame (NULL));
+  m_lang = current_language->la_language;
+  save_selected_frame (&m_fid, &m_level);
 }
 
 /* See frame.h  */
 scoped_restore_selected_frame::~scoped_restore_selected_frame ()
 {
-  frame_info *frame = frame_find_by_id (m_fid);
-  if (frame == NULL)
-    warning (_("Unable to restore previously selected frame."));
-  else
-    select_frame (frame);
+  restore_selected_frame (m_fid, m_level);
+  set_language (m_lang);
 }
 
 /* Flag to control debugging.  */
@@ -1641,10 +1639,51 @@ get_current_frame (void)
 }
 
 /* The "selected" stack frame is used by default for local and arg
-   access.  May be zero, for no selected frame.  */
+   access.  */
 
+/* The "single source of truth" for the selected frame is the
+   SELECTED_FRAME_ID / SELECTED_FRAME_LEVEL pair.  Frame IDs can be
+   saved/restored across reinitializing the frame cache, while
+   frame_info pointers can't (frame_info objects are invalidated).  If
+   we know the corresponding frame_info object, it is cached in
+   SELECTED_FRAME.  If SELECTED_FRAME_ID / SELECTED_FRAME_LEVEL are
+   null_ptid / -1, and the target has stack and is stopped, the
+   selected frame is the current (innermost) frame, otherwise there's
+   no selected frame.  */
+static frame_id selected_frame_id = null_frame_id;
+static int selected_frame_level = -1;
+
+/* The cached frame_info object pointing to the selected frame.
+   Looked up on demand by get_selected_frame.  */
 static struct frame_info *selected_frame;
 
+/* See frame.h.  */
+
+void
+save_selected_frame (frame_id *frame_id, int *frame_level)
+  noexcept
+{
+  *frame_id = selected_frame_id;
+  *frame_level = selected_frame_level;
+}
+
+/* See frame.h.  */
+
+void
+restore_selected_frame (frame_id a_frame_id, int frame_level)
+  noexcept
+{
+  /* save_selected_frame never returns level == 0, so we shouldn't see
+     it here either.  */
+  gdb_assert (frame_level != 0);
+
+  selected_frame_id = a_frame_id;
+  selected_frame_level = frame_level;
+
+  /* Will be looked up latter by get_seleted_frame.  */
+  selected_frame = nullptr;
+}
+
 int
 has_stack_frames (void)
 {
@@ -1682,24 +1721,14 @@ get_selected_frame (const char *message)
     {
       if (message != NULL && !has_stack_frames ())
 	error (("%s"), message);
-      /* Hey!  Don't trust this.  It should really be re-finding the
-	 last selected frame of the currently selected thread.  This,
-	 though, is better than nothing.  */
-      select_frame (get_current_frame ());
+
+      lookup_selected_frame (selected_frame_id, selected_frame_level);
     }
   /* There is always a frame.  */
   gdb_assert (selected_frame != NULL);
   return selected_frame;
 }
 
-/* If there is a selected frame, return it.  Otherwise, return NULL.  */
-
-struct frame_info *
-get_selected_frame_if_set (void)
-{
-  return selected_frame;
-}
-
 /* This is a variant of get_selected_frame() which can be called when
    the inferior does not have a frame; in that case it will return
    NULL instead of calling error().  */
@@ -1712,12 +1741,42 @@ deprecated_safe_get_selected_frame (void)
   return get_selected_frame (NULL);
 }
 
-/* Select frame FI (or NULL - to invalidate the current frame).  */
+/* Select frame FI (or NULL - to invalidate the selected frame).  */
 
 void
 select_frame (struct frame_info *fi)
 {
   selected_frame = fi;
+  selected_frame_level = frame_relative_level (fi);
+  if (selected_frame_level == 0)
+    {
+      /* Treat the current frame especially -- we want to always
+	 save/restore it without warning, even if the frame ID changes
+	 (see lookup_selected_frame).  E.g.:
+
+	  // The current frame is selected, the target had just stopped.
+	  {
+	    scoped_restore_selected_frame restore_frame;
+	    some_operation_that_changes_the_stack ();
+	  }
+	  // scoped_restore_selected_frame's dtor runs, but the
+	  // original frame_id can't be found.  No matter whether it
+	  // is found or not, we still end up with the now-current
+	  // frame selected.  Warning in lookup_selected_frame in this
+	  // case seems pointless.
+
+	 Also get_frame_id may access the target's registers/memory,
+	 and thus skipping get_frame_id optimizes the common case.
+
+	 Saving the selected frame this way makes get_selected_frame
+	 and restore_current_frame return/re-select whatever frame is
+	 the innermost (current) then.  */
+      selected_frame_level = -1;
+      selected_frame_id = null_frame_id;
+    }
+  else
+    selected_frame_id = get_frame_id (fi);
+
   /* NOTE: cagney/2002-05-04: FI can be NULL.  This occurs when the
      frame is being invalidated.  */
 

gdb/frame.h

@@ -181,8 +181,14 @@ public:
 
 private:
 
-  /* The ID of the previously selected frame.  */
+  /* The ID and level of the previously selected frame.  */
   struct frame_id m_fid;
+  int m_level;
+
+  /* Save/restore the language as well, because selecting a frame
+     changes the current language to the frame's language if "set
+     language auto".  */
+  enum language m_lang;
 };
 
 /* Methods for constructing and comparing Frame IDs.  */
@@ -329,13 +335,33 @@ extern void reinit_frame_cache (void);
    and then return that thread's previously selected frame.  */
 extern struct frame_info *get_selected_frame (const char *message);
 
-/* If there is a selected frame, return it.  Otherwise, return NULL.  */
-extern struct frame_info *get_selected_frame_if_set (void);
-
-/* Select a specific frame.  NULL, apparently implies re-select the
-   inner most frame.  */
+/* Select a specific frame.  NULL implies re-select the inner most
+   frame.  */
 extern void select_frame (struct frame_info *);
 
+/* Save the frame ID and frame level of the selected frame in FRAME_ID
+   and FRAME_LEVEL, to be restored later with restore_selected_frame.
+   This is preferred over getting the same info out of
+   get_selected_frame directly because this function does not create
+   the selected-frame's frame_info object if it hasn't been created
+   yet, and thus doesn't throw.  */
+extern void save_selected_frame (frame_id *frame_id, int *frame_level)
+  noexcept;
+
+/* Restore selected frame as saved with save_selected_frame.  Does not
+   try to find the corresponding frame_info object.  Instead the next
+   call to get_selected_frame will look it up and cache the result.
+   This function does not throw, it is designed to be safe to called
+   from the destructors of RAII types.  */
+extern void restore_selected_frame (frame_id frame_id, int frame_level)
+  noexcept;
+
+/* Lookup the frame_info object for the selected frame FRAME_ID /
+   FRAME_LEVEL and cache the result.  If FRAME_LEVEL > 0 and the
+   originally selected frame isn't found, warn and select the
+   innermost (current) frame.  */
+extern void lookup_selected_frame (frame_id frame_id, int frame_level);
+
 /* Given a FRAME, return the next (more inner, younger) or previous
    (more outer, older) frame.  */
 extern struct frame_info *get_prev_frame (struct frame_info *);

gdb/gdbthread.h

@@ -667,6 +667,10 @@ private:
   frame_id m_selected_frame_id;
   int m_selected_frame_level;
   bool m_was_stopped;
+  /* Save/restore the language as well, because selecting a frame
+     changes the current language to the frame's language if "set
+     language auto".  */
+  enum language m_lang;
 };
 
 /* Returns a pointer into the thread_info corresponding to

gdb/infrun.c

@@ -3601,23 +3601,9 @@ do_target_wait_1 (inferior *inf, ptid_t ptid,
   return event_ptid;
 }
 
-/* Returns true if INF has any resumed thread with a status
-   pending.  */
-
-static bool
-threads_are_resumed_pending_p (inferior *inf)
-{
-  for (thread_info *tp : inf->non_exited_threads ())
-    if (tp->resumed
-	&& tp->suspend.waitstatus_pending_p)
-      return true;
-
-  return false;
-}
-
 /* Wrapper for target_wait that first checks whether threads have
    pending statuses to report before actually asking the target for
-   more events. Polls for events from all inferiors/targets.  */
+   more events.  Polls for events from all inferiors/targets.  */
 
 static bool
 do_target_wait (ptid_t wait_ptid, execution_control_state *ecs, int options)
@@ -3625,20 +3611,18 @@ do_target_wait (ptid_t wait_ptid, execution_control_state *ecs, int options)
   int num_inferiors = 0;
   int random_selector;
 
-  /* For fairness, we pick the first inferior/target to poll at
-     random, and then continue polling the rest of the inferior list
-     starting from that one in a circular fashion until the whole list
-     is polled once.  */
+  /* For fairness, we pick the first inferior/target to poll at random
+     out of all inferiors that may report events, and then continue
+     polling the rest of the inferior list starting from that one in a
+     circular fashion until the whole list is polled once.  */
 
   auto inferior_matches = [&wait_ptid] (inferior *inf)
     {
       return (inf->process_target () != NULL
-	      && (threads_are_executing (inf->process_target ())
-		  || threads_are_resumed_pending_p (inf))
 	      && ptid_t (inf->pid).matches (wait_ptid));
     };
 
-  /* First see how many resumed inferiors we have.  */
+  /* First see how many matching inferiors we have.  */
   for (inferior *inf : all_inferiors ())
     if (inferior_matches (inf))
       num_inferiors++;
@@ -3649,7 +3633,7 @@ do_target_wait (ptid_t wait_ptid, execution_control_state *ecs, int options)
       return false;
     }
 
-  /* Now randomly pick an inferior out of those that were resumed.  */
+  /* Now randomly pick an inferior out of those that matched.  */
   random_selector = (int)
     ((num_inferiors * (double) rand ()) / (RAND_MAX + 1.0));
 
@@ -3658,7 +3642,7 @@ do_target_wait (ptid_t wait_ptid, execution_control_state *ecs, int options)
 			"infrun: Found %d inferiors, starting at #%d\n",
 			num_inferiors, random_selector);
 
-  /* Select the Nth inferior that was resumed.  */
+  /* Select the Nth inferior that matched.  */
 
   inferior *selected = nullptr;
 
@@ -3670,7 +3654,7 @@ do_target_wait (ptid_t wait_ptid, execution_control_state *ecs, int options)
 	  break;
 	}
 
-  /* Now poll for events out of each of the resumed inferior's
+  /* Now poll for events out of each of the matching inferior's
      targets, starting from the selected one.  */
 
   auto do_wait = [&] (inferior *inf)
@@ -3680,8 +3664,8 @@ do_target_wait (ptid_t wait_ptid, execution_control_state *ecs, int options)
     return (ecs->ws.kind != TARGET_WAITKIND_IGNORE);
   };
 
-  /* Needed in all-stop+target-non-stop mode, because we end up here
-     spuriously after the target is all stopped and we've already
+  /* Needed in 'all-stop + target-non-stop' mode, because we end up
+     here spuriously after the target is all stopped and we've already
      reported the stop to the user, polling for events.  */
   scoped_restore_current_thread restore_thread;
 
@@ -5068,23 +5052,77 @@ handle_no_resumed (struct execution_control_state *ecs)
      have resumed threads _now_.  In the example above, this removes
      thread 3 from the thread list.  If thread 2 was re-resumed, we
      ignore this event.  If we find no thread resumed, then we cancel
-     the synchronous command show "no unwaited-for " to the user.  */
-  update_thread_list ();
+     the synchronous command and show "no unwaited-for " to the
+     user.  */
 
-  for (thread_info *thread : all_non_exited_threads (ecs->target))
+  inferior *curr_inf = current_inferior ();
+
+  scoped_restore_current_thread restore_thread;
+
+  for (auto *target : all_non_exited_process_targets ())
     {
-      if (thread->executing
-	  || thread->suspend.waitstatus_pending_p)
+      switch_to_target_no_thread (target);
+      update_thread_list ();
+    }
+
+  /* If:
+
+       - the current target has no thread executing, and
+       - the current inferior is native, and
+       - the current inferior is the one which has the terminal, and
+       - we did nothing,
+
+     then a Ctrl-C from this point on would remain stuck in the
+     kernel, until a thread resumes and dequeues it.  That would
+     result in the GDB CLI not reacting to Ctrl-C, not able to
+     interrupt the program.  To address this, if the current inferior
+     no longer has any thread executing, we give the terminal to some
+     other inferior that has at least one thread executing.  */
+  bool swap_terminal = true;
+
+  /* Whether to ignore this TARGET_WAITKIND_NO_RESUMED event, or
+     whether to report it to the user.  */
+  bool ignore_event = false;
+
+  for (thread_info *thread : all_non_exited_threads ())
+    {
+      if (swap_terminal && thread->executing)
 	{
-	  /* There were no unwaited-for children left in the target at
-	     some point, but there are now.  Just ignore.  */
+	  if (thread->inf != curr_inf)
+	    {
+	      target_terminal::ours ();
+
+	      switch_to_thread (thread);
+	      target_terminal::inferior ();
+	    }
+	  swap_terminal = false;
+	}
+
+      if (!ignore_event
+	  && (thread->executing
+	      || thread->suspend.waitstatus_pending_p))
+	{
+	  /* Either there were no unwaited-for children left in the
+	     target at some point, but there are now, or some target
+	     other than the eventing one has unwaited-for children
+	     left.  Just ignore.  */
 	  if (debug_infrun)
 	    fprintf_unfiltered (gdb_stdlog,
 				"infrun: TARGET_WAITKIND_NO_RESUMED "
 				"(ignoring: found resumed)\n");
-	  prepare_to_wait (ecs);
-	  return 1;
+
+	  ignore_event = true;
 	}
+
+      if (ignore_event && !swap_terminal)
+	break;
+    }
+
+  if (ignore_event)
+    {
+      switch_to_inferior_no_thread (curr_inf);
+      prepare_to_wait (ecs);
+      return 1;
     }
 
   /* Go ahead and report the event.  */
@@ -8116,7 +8154,11 @@ prepare_to_wait (struct execution_control_state *ecs)
 
   ecs->wait_some_more = 1;
 
-  if (!target_is_async_p ())
+  /* If the target can't async, emulate it by marking the infrun event
+     handler such that as soon as we get back to the event-loop, we
+     immediately end up in fetch_inferior_event again calling
+     target_wait.  */
+  if (!target_can_async_p ())
     mark_infrun_async_event_handler ();
 }
 

gdb/remote.c

@@ -5605,7 +5605,7 @@ remote_target::open_1 (const char *name, int from_tty, int extended_p)
 
   /* Register extra event sources in the event loop.  */
   rs->remote_async_inferior_event_token
-    = create_async_event_handler (remote_async_inferior_event_handler, NULL);
+    = create_async_event_handler (remote_async_inferior_event_handler, remote);
   rs->notif_state = remote_notif_state_allocate (remote);
 
   /* Reset the target state; these things will be queried either by
@@ -14164,6 +14164,19 @@ static void
 remote_async_inferior_event_handler (gdb_client_data data)
 {
   inferior_event_handler (INF_REG_EVENT);
+
+  remote_target *remote = (remote_target *) data;
+  remote_state *rs = remote->get_remote_state ();
+
+  /* inferior_event_handler may have consumed an event pending on the
+     infrun side without calling target_wait on the REMOTE target, or
+     may have pulled an event out of a different target.  Keep trying
+     for this remote target as long it still has either pending events
+     or unacknowledged notifications.  */
+
+  if (rs->notif_state->pending_event[notif_client_stop.id] != NULL
+      || !rs->stop_reply_queue.empty ())
+    mark_async_event_handler (rs->remote_async_inferior_event_token);
 }
 
 int

gdb/stack.c

@@ -1836,9 +1836,9 @@ trailing_outermost_frame (int count)
 static void
 select_frame_command_core (struct frame_info *fi, bool ignored)
 {
-  struct frame_info *prev_frame = get_selected_frame_if_set ();
+  frame_info *prev_frame = get_selected_frame (NULL);
   select_frame (fi);
-  if (get_selected_frame_if_set () != prev_frame)
+  if (get_selected_frame (NULL) != prev_frame)
     gdb::observers::user_selected_context_changed.notify (USER_SELECTED_FRAME);
 }
 
@@ -1857,10 +1857,9 @@ select_frame_for_mi (struct frame_info *fi)
 static void
 frame_command_core (struct frame_info *fi, bool ignored)
 {
-  struct frame_info *prev_frame = get_selected_frame_if_set ();
-
+  frame_info *prev_frame = get_selected_frame (nullptr);
   select_frame (fi);
-  if (get_selected_frame_if_set () != prev_frame)
+  if (get_selected_frame (nullptr) != prev_frame)
     gdb::observers::user_selected_context_changed.notify (USER_SELECTED_FRAME);
   else
     print_selected_thread_frame (current_uiout, USER_SELECTED_FRAME);

gdb/target.c

@@ -3274,7 +3274,7 @@ target_pass_ctrlc (void)
       if (proc_target == NULL)
 	continue;
 
-      for (thread_info *thr : inf->threads ())
+      for (thread_info *thr : inf->non_exited_threads ())
 	{
 	  /* A thread can be THREAD_STOPPED and executing, while
 	     running an infcall.  */

gdb/testsuite/gdb.multi/multi-target.c

@@ -26,12 +26,14 @@
 
 static pthread_barrier_t barrier;
 
+volatile int exit_thread;
+
 static void *
 thread_start (void *arg)
 {
   pthread_barrier_wait (&barrier);
 
-  while (1)
+  while (!exit_thread)
     sleep (1);
   return NULL;
 }

gdb/testsuite/gdb.multi/multi-target.exp

@@ -439,6 +439,77 @@ proc test_info_inferiors {multi_process} {
     }
 }
 
+# Test that when there's a foreground execution command in progress, a
+# TARGET_WAITKIND_NO_RESUMED for a particular target is ignored when
+# other targets are still resumed.
+
+proc test_no_resumed {} {
+    proc test_no_resumed_infs {inf_A inf_B} {
+	global gdb_prompt
+
+	if {![setup "off"]} {
+	    untested "setup failed"
+	    return
+	}
+
+	gdb_test "thread $inf_A.2" "Switching to thread $inf_A\.2 .*" \
+	    "select thread of target A"
+
+	gdb_test_no_output "set scheduler-locking on"
+
+	gdb_test_multiple "continue &" "" {
+	    -re "Continuing.*$gdb_prompt " {
+		pass $gdb_test_name
+	    }
+	}
+
+	gdb_test "thread $inf_B.2" "Switching to thread $inf_B\.2 .*" \
+	    "select thread of target B"
+	gdb_test "p exit_thread = 1" " = 1" \
+	    "set the thread to exit on resumption"
+
+	# Wait 3 seconds.  If we see any response from GDB, such as
+	# "No unwaited-for children left." it's a bug.
+	gdb_test_multiple "continue" "continue" {
+	    -timeout 3
+	    timeout {
+		pass $gdb_test_name
+	    }
+	}
+
+	# Now stop the program (all targets).
+	send_gdb "\003"
+	gdb_test_multiple "" "send_gdb control C" {
+	    -re "received signal SIGINT.*$gdb_prompt $" {
+		pass $gdb_test_name
+	    }
+	}
+
+	gdb_test_multiple "info threads" "all threads stopped" {
+	    -re "\\\(running\\\).*$gdb_prompt $" {
+		fail $gdb_test_name
+	    }
+	    -re "$gdb_prompt $" {
+		pass $gdb_test_name
+	    }
+	}
+    }
+
+    # inferior 1 -> native
+    # inferior 2 -> extended-remote 1
+    # inferior 5 -> extended-remote 2
+    set inferiors {1 2 5}
+    foreach_with_prefix inf_A $inferiors {
+	foreach_with_prefix inf_B $inferiors {
+	    if {$inf_A == $inf_B} {
+		continue
+	    }
+	    test_no_resumed_infs $inf_A $inf_B
+	}
+    }
+}
+
+
 # Make a core file with two threads upfront.  Several tests load the
 # same core file.
 prepare_core
@@ -467,4 +538,9 @@ with_test_prefix "info-inferiors" {
     }
 }
 
+# Test TARGET_WAITKIND_NO_RESUMED handling with multiple targets.
+with_test_prefix "no-resumed" {
+    test_no_resumed
+}
+
 cleanup_gdbservers

gdb/thread.c

@@ -1325,20 +1325,25 @@ switch_to_thread (process_stratum_target *proc_target, ptid_t ptid)
   switch_to_thread (thr);
 }
 
-static void
-restore_selected_frame (struct frame_id a_frame_id, int frame_level)
+/* See frame.h.  */
+
+void
+lookup_selected_frame (struct frame_id a_frame_id, int frame_level)
 {
   struct frame_info *frame = NULL;
   int count;
 
-  /* This means there was no selected frame.  */
+  /* This either means there was no selected frame, or the selected
+     frame was the innermost (the current frame).  */
   if (frame_level == -1)
     {
-      select_frame (NULL);
+      select_frame (get_current_frame ());
       return;
     }
 
-  gdb_assert (frame_level >= 0);
+  /* select_frame never saves 0 in SELECTED_FRAME_LEVEL, so we
+     shouldn't see it here.  */
+  gdb_assert (frame_level > 0);
 
   /* Restore by level first, check if the frame id is the same as
      expected.  If that fails, try restoring by frame id.  If that
@@ -1409,22 +1414,14 @@ scoped_restore_current_thread::restore ()
       && target_has_stack
       && target_has_memory)
     restore_selected_frame (m_selected_frame_id, m_selected_frame_level);
+
+  set_language (m_lang);
 }
 
 scoped_restore_current_thread::~scoped_restore_current_thread ()
 {
   if (!m_dont_restore)
-    {
-      try
-	{
-	  restore ();
-	}
-      catch (const gdb_exception &ex)
-	{
-	  /* We're in a dtor, there's really nothing else we can do
-	     but swallow the exception.  */
-	}
-    }
+    restore ();
 
   if (m_thread != NULL)
     m_thread->decref ();
@@ -1433,46 +1430,21 @@ scoped_restore_current_thread::~scoped_restore_current_thread ()
 
 scoped_restore_current_thread::scoped_restore_current_thread ()
 {
-  m_thread = NULL;
   m_inf = current_inferior ();
+  m_inf->incref ();
+
+  m_lang = current_language->la_language;
 
   if (inferior_ptid != null_ptid)
     {
-      thread_info *tp = inferior_thread ();
-      struct frame_info *frame;
+      m_thread = inferior_thread ();
+      m_thread->incref ();
 
-      m_was_stopped = tp->state == THREAD_STOPPED;
-      if (m_was_stopped
-	  && target_has_registers
-	  && target_has_stack
-	  && target_has_memory)
-	{
-	  /* When processing internal events, there might not be a
-	     selected frame.  If we naively call get_selected_frame
-	     here, then we can end up reading debuginfo for the
-	     current frame, but we don't generally need the debuginfo
-	     at this point.  */
-	  frame = get_selected_frame_if_set ();
-	}
-      else
-	frame = NULL;
-
-      try
-	{
-	  m_selected_frame_id = get_frame_id (frame);
-	  m_selected_frame_level = frame_relative_level (frame);
-	}
-      catch (const gdb_exception_error &ex)
-	{
-	  m_selected_frame_id = null_frame_id;
-	  m_selected_frame_level = -1;
-	}
-
-      tp->incref ();
-      m_thread = tp;
+      m_was_stopped = m_thread->state == THREAD_STOPPED;
+      save_selected_frame (&m_selected_frame_id, &m_selected_frame_level);
     }
-
-  m_inf->incref ();
+  else
+    m_thread = NULL;
 }
 
 /* See gdbthread.h.  */

gdb/value.c

@@ -1412,7 +1412,18 @@ value_optimized_out (struct value *value)
 	}
       catch (const gdb_exception_error &ex)
 	{
-	  /* Fall back to checking value->optimized_out.  */
+	  switch (ex.error)
+	    {
+	    case MEMORY_ERROR:
+	    case OPTIMIZED_OUT_ERROR:
+	    case NOT_AVAILABLE_ERROR:
+	      /* These can normally happen when we try to access an
+		 optimized out or unavailable register, either in a
+		 physical register or spilled to memory.  */
+	      break;
+	    default:
+	      throw;
+	    }
 	}
     }