When attaching GDB to OpenOCD, the target state is no longer affected.
authoroharboe <oharboe@b42882b7-edfa-0310-969c-e2dbd0fdcd60>
Thu, 24 Sep 2009 06:34:23 +0000 (06:34 +0000)
committeroharboe <oharboe@b42882b7-edfa-0310-969c-e2dbd0fdcd60>
Thu, 24 Sep 2009 06:34:23 +0000 (06:34 +0000)
Added gdb_sync feature that allows GDB to sync up to target state.
Issue "monitor gdb_sync" and the next stepi, will return immediately
with updated register values to GDB.

git-svn-id: svn://svn.berlios.de/openocd/trunk@2754 b42882b7-edfa-0310-969c-e2dbd0fdcd60

src/server/gdb_server.c
src/server/gdb_server.h

index dc14b75c83d17adf8c2c3866855246decfa8491b..44d9de8381222f8454428a8d6107d915838702f5 100644 (file)
@@ -40,6 +40,8 @@
 #define _DEBUG_GDB_IO_
 #endif
 
+static gdb_connection_t *current_gdb_connection;
+
 static int gdb_breakpoint_override;
 static enum breakpoint_type gdb_breakpoint_override_type;
 
@@ -750,6 +752,7 @@ int gdb_new_connection(connection_t *connection)
        gdb_connection->closed = 0;
        gdb_connection->busy = 0;
        gdb_connection->noack_mode = 0;
+       gdb_connection->sync = true;
 
        /* send ACK to GDB for debug request */
        gdb_write(connection, "+", 1);
@@ -767,30 +770,6 @@ int gdb_new_connection(connection_t *connection)
        /* register callback to be informed about target events */
        target_register_event_callback(gdb_target_callback_event_handler, connection);
 
-       /* a gdb session just attached, try to put the target in halt mode.
-        *
-        * DANGER!!!!
-        *
-        * If the halt fails(e.g. target needs a reset, JTAG communication not
-        * working, etc.), then the GDB connect will succeed as
-        * the get_gdb_reg_list() will lie and return a register list with
-        * dummy values.
-        *
-        * This allows GDB monitor commands to be run from a GDB init script to
-        * initialize the target
-        *
-        * Also, since the halt() is asynchronous target connect will be
-        * instantaneous and thus avoiding annoying timeout problems during
-        * connect.
-        */
-       target_halt(gdb_service->target);
-       /* FIX!!!! could extended-remote work better here?
-        *
-        *  wait a tiny bit for halted state or we just continue. The
-        * GDB register packet will then contain garbage
-        */
-       target_wait_state(gdb_service->target, TARGET_HALTED, 500);
-
        /* remove the initial ACK from the incoming buffer */
        if ((retval = gdb_get_char(connection, &initial_ack)) != ERROR_OK)
                return retval;
@@ -1609,7 +1588,11 @@ int gdb_query_packet(connection_t *connection, target_t *target, char *packet, i
                        /* We want to print all debug output to GDB connection */
                        log_add_callback(gdb_log_callback, connection);
                        target_call_timer_callbacks_now();
+                       /* some commands need to know the GDB connection, make note of current
+                        * GDB connection. */
+                       current_gdb_connection = gdb_connection;
                        command_run_line(cmd_ctx, cmd);
+                       current_gdb_connection = NULL;
                        target_call_timer_callbacks_now();
                        log_remove_callback(gdb_log_callback, connection);
                        free(cmd);
@@ -2107,20 +2090,52 @@ int gdb_input_inner(connection_t *connection)
                                case 'c':
                                case 's':
                                        {
-                                               if (target->state != TARGET_HALTED)
+                                               int retval = ERROR_OK;
+
+                                               gdb_connection_t *gdb_con = connection->priv;
+                                               log_add_callback(gdb_log_callback, connection);
+
+                                               bool nostep = false;
+                                               if (target->state == TARGET_RUNNING)
+                                               {
+                                                       LOG_WARNING("The target is already running. Halt target before stepi/continue.");
+                                                       retval = target_halt(target);
+                                                       if (retval == ERROR_OK)
+                                                               retval = target_wait_state(target, TARGET_HALTED, 100);
+                                               } else if (target->state != TARGET_HALTED)
+                                               {
+                                                       LOG_WARNING("The target is not in the halted nor running stated, stepi/continue ignored.");
+                                                       nostep = true;
+                                               } else if ((packet[0] == 's') && gdb_con->sync)
+                                               {
+                                                       /* Hmm..... when you issue a continue in GDB, then a "stepi" is
+                                                        * sent by GDB first to OpenOCD, thus defeating the check to
+                                                        * make only the single stepping have the sync feature...
+                                                        */
+                                                       nostep = true;
+                                                       LOG_WARNING("stepi ignored. GDB will now fetch the register state from the target.");
+                                               }
+                                               gdb_con->sync = false;
+
+                                               if ((retval!=ERROR_OK) || nostep)
                                                {
-                                                       /* If the target isn't in the halted state, then we can't
+                                                       /* Either the target isn't in the halted state, then we can't
                                                         * step/continue. This might be early setup, etc.
+                                                        *
+                                                        * Or we want to allow GDB to pick up a fresh set of
+                                                        * register values without modifying the target state.
+                                                        *
                                                         */
                                                        gdb_sig_halted(connection);
+
+                                                       /* stop forwarding log packets! */
+                                                       log_remove_callback(gdb_log_callback, connection);
                                                } else
                                                {
                                                        /* We're running/stepping, in which case we can
                                                         * forward log output until the target is halted
                                                         */
-                                                       gdb_connection_t *gdb_con = connection->priv;
                                                        gdb_con->frontend_state = TARGET_RUNNING;
-                                                       log_add_callback(gdb_log_callback, connection);
                                                        target_call_event_callbacks(target, TARGET_EVENT_GDB_START);
                                                        int retval = gdb_step_continue_packet(connection, target, packet, packet_size);
                                                        if (retval != ERROR_OK)
@@ -2255,6 +2270,25 @@ int gdb_init(void)
        return ERROR_OK;
 }
 
+int handle_gdb_sync_command(struct command_context_s *cmd_ctx, char *cmd, char **args, int argc)
+{
+       if (argc != 0)
+       {
+               return ERROR_COMMAND_SYNTAX_ERROR;
+       }
+
+       if (current_gdb_connection == NULL)
+       {
+               command_print(cmd_ctx,
+                               "gdb_sync command can only be run from within gdb using \"monitor gdb_sync\"");
+               return ERROR_FAIL;
+       }
+
+       current_gdb_connection->sync = true;
+
+       return ERROR_OK;
+}
+
 /* daemon configuration command gdb_port */
 int handle_gdb_port_command(struct command_context_s *cmd_ctx, char *cmd, char **args, int argc)
 {
@@ -2399,6 +2433,8 @@ int handle_gdb_breakpoint_override_command(struct command_context_s *cmd_ctx, ch
 
 int gdb_register_commands(command_context_t *command_context)
 {
+       register_command(command_context, NULL, "gdb_sync", handle_gdb_sync_command,
+                       COMMAND_ANY, "next stepi will return immediately allowing GDB fetch register state without affecting target state");
        register_command(command_context, NULL, "gdb_port", handle_gdb_port_command,
                        COMMAND_ANY, "daemon configuration command gdb_port");
        register_command(command_context, NULL, "gdb_detach", handle_gdb_detach_command,
index fb4cdbab6becf713d84aab1218a02d4c8f221f65..ea981f31640f1fd12b9c7d60100f651bced6ec1e 100644 (file)
@@ -43,6 +43,10 @@ typedef struct gdb_connection_s
        int closed;
        int busy;
        int noack_mode;
+       bool sync;      /* set flag to true if you want the next stepi to return immediately.
+                      allowing GDB to pick up a fresh set of register values from the target
+                      without modifying the target state. */
+
 } gdb_connection_t;
 
 typedef struct gdb_service_s