diff --git a/tritondse/process_state.py b/tritondse/process_state.py
index 1ad3c31..3901342 100644
--- a/tritondse/process_state.py
+++ b/tritondse/process_state.py
@@ -193,7 +193,12 @@ def spawn_new_thread(self, new_pc: Addr, args: Addr) -> ThreadContext:
         thread.save(self.tt_ctx)
 
         # Concretize pc, bp, sp, and first (argument)
-        regs = [self.program_counter_register, self.stack_pointer_register, self.base_pointer_register, self._get_argument_register(0)]
+        regs = [
+            self.program_counter_register,
+            self.stack_pointer_register,
+            self.base_pointer_register,
+            self._get_argument_register(0)
+        ]
         for reg in regs:
             if reg.getId() in thread.sregs:
                 del thread.sregs[reg.getId()]
@@ -201,8 +206,16 @@ def spawn_new_thread(self, new_pc: Addr, args: Addr) -> ThreadContext:
         thread.cregs[self.program_counter_register.getId()] = new_pc  # set new pc
         thread.cregs[self._get_argument_register(0).getId()] = args   # set args pointer
         stack = self.memory.map_from_name(self.STACK_SEG)
-        thread.cregs[self.base_pointer_register.getId()] = ((stack.start+stack.size) - ((1 << 28) * tid))
-        thread.cregs[self.stack_pointer_register.getId()] = ((stack.start+stack.size) - ((1 << 28) * tid))
+        stack_base_addr = ((stack.start + stack.size - self.ptr_size) - ((1 << 28) * tid))
+        thread.cregs[self.base_pointer_register.getId()] = stack_base_addr
+        thread.cregs[self.stack_pointer_register.getId()] = stack_base_addr
+
+        if self.architecture == Architecture.AARCH64:
+            thread.cregs[getattr(self.registers, 'x30').getId()] = 0xcafecafe
+        elif self.architecture == Architecture.ARM32:
+            thread.cregs[getattr(self.registers, 'r14').getId()] = 0xcafecafe
+        elif self.architecture in [Architecture.X86, Architecture.X86_64]:
+            self.memory.write_ptr(stack_base_addr, 0xcafecafe)
 
         # Add the thread in the pool of threads
         self._threads[tid] = thread
diff --git a/tritondse/routines.py b/tritondse/routines.py
index 07c3f30..9d3b0ea 100644
--- a/tritondse/routines.py
+++ b/tritondse/routines.py
@@ -1219,9 +1219,12 @@ def rtn_pthread_join(se: 'SymbolicExecutor', pstate: 'ProcessState'):
     arg0 = pstate.get_argument_value(0)
     arg1 = pstate.get_argument_value(1)
 
-    if arg0 in pstate.threads:
+    if arg0 in pstate._threads:
+        # This will change the state of the current thread from RUNNING to
+        # JOINING. Therefore, it will not run until the joining thread
+        # finishes.
         pstate.current_thread.join_thread(arg0)
-        logger.info(f"Thread id {pstate.current_thread.tid} joined thread id {arg0}")
+        logger.info(f"Thread id {pstate.current_thread.tid} is waiting thread id {arg0} to join")
     else:
         pstate.current_thread.cancel_join()
         logger.debug(f"Thread id {arg0} already destroyed")
diff --git a/tritondse/symbolic_executor.py b/tritondse/symbolic_executor.py
index 41cc242..7402aca 100644
--- a/tritondse/symbolic_executor.py
+++ b/tritondse/symbolic_executor.py
@@ -232,7 +232,9 @@ def __schedule_thread(self) -> None:
         if len(threads_list) == 1:  # If there is only one thread no need to schedule another thread
             return
 
-        if self.pstate.current_thread.count > self.config.thread_scheduling:
+        if self.pstate.current_thread.count > self.config.thread_scheduling or \
+           self.pstate.current_thread.is_waiting_to_join() or \
+           not self.pstate.current_thread.is_running():
             # Select the next thread to execute
             next_th = self._fetch_next_thread(threads_list)
 
@@ -287,6 +289,17 @@ def step(self) -> bool:
             # Fetch program counter (of the thread selected), at this point the current thread should be running!
             self.current_pc = self.pstate.cpu.program_counter  # should normally be already set but still.
 
+            if self.current_pc == 0xcafecafe:
+                logger.debug(f"Schedule thread ({self.pstate.current_thread.tid}) finished running")
+                self.pstate.current_thread.kill()
+
+                # Check whether the main thread was waiting for the schedule thread to join.
+                if self.pstate._threads[0]._join_th_id == self.pstate.current_thread.tid:
+                    logger.debug(f"Set main thread state to RUNNING")
+                    self.pstate._threads[0]._join_th_id = None
+                    self.pstate._threads[0].state = ThreadState.RUNNING
+                return True
+
             if self.current_pc == self._run_to_target:  # Hit the location we wanted to reach
                 return False
 
diff --git a/tritondse/thread_context.py b/tritondse/thread_context.py
index 6fc480e..61837d4 100644
--- a/tritondse/thread_context.py
+++ b/tritondse/thread_context.py
@@ -5,6 +5,11 @@
 # third-party imports
 from triton import TritonContext
 
+# local imports
+import tritondse.logging
+
+logger = tritondse.logging.get("thread_context")
+
 
 @enum_tools.documentation.document_enum
 class ThreadState(Enum):