score: Optimize _Thread_queue_Enqueue()

Move thread state for _Thread_queue_Enqueue() to the thread queue
context.  This reduces the parameter count of _Thread_queue_Enqueue()
from five to four (ARM for example has only four function parameter
registers).  Since the thread state is used after several function calls
inside _Thread_queue_Enqueue() this parameter was saved on the stack
previously.

20 files changed, 117 insertions, 28 deletions

diff --git a/cpukit/posix/src/condwaitsupp.c b/cpukit/posix/src/condwaitsupp.c
index 9024dbf236..589b5e2810 100644
--- a/cpukit/posix/src/condwaitsupp.c
+++ b/cpukit/posix/src/condwaitsupp.c
@@ -117,6 +117,10 @@ int _POSIX_Condition_variables_Wait_support(
   executing = _Thread_Executing;
 
   if ( !already_timedout ) {
+    _Thread_queue_Context_set_thread_state(
+      &queue_context,
+      STATES_WAITING_FOR_CONDITION_VARIABLE
+    );
     _Thread_queue_Context_set_enqueue_callout(
       &queue_context,
       _POSIX_Condition_variables_Enqueue_callout
@@ -125,7 +129,6 @@ int _POSIX_Condition_variables_Wait_support(
       &the_cond->Wait_queue.Queue,
       POSIX_CONDITION_VARIABLES_TQ_OPERATIONS,
       executing,
-      STATES_WAITING_FOR_CONDITION_VARIABLE,
       &queue_context
     );
     error = _POSIX_Get_error_after_wait( executing );
diff --git a/cpukit/posix/src/nanosleep.c b/cpukit/posix/src/nanosleep.c
index 5deb2b8c02..10158ff44b 100644
--- a/cpukit/posix/src/nanosleep.c
+++ b/cpukit/posix/src/nanosleep.c
@@ -47,6 +47,10 @@ static inline int nanosleep_helper(
   err = 0;
 
   _Thread_queue_Context_initialize( &queue_context );
+  _Thread_queue_Context_set_thread_state(
+    &queue_context,
+    STATES_DELAYING | STATES_INTERRUPTIBLE_BY_SIGNAL
+  );
   _Thread_queue_Context_set_enqueue_callout(
     &queue_context,
     _Thread_queue_Enqueue_do_nothing
@@ -66,7 +70,6 @@ static inline int nanosleep_helper(
     &_Nanosleep_Pseudo_queue.Queue,
     &_Thread_queue_Operations_FIFO,
     _Thread_Executing,
-    STATES_DELAYING | STATES_INTERRUPTIBLE_BY_SIGNAL,
     &queue_context
   );
 
diff --git a/cpukit/posix/src/sigtimedwait.c b/cpukit/posix/src/sigtimedwait.c
index 200d75fdf9..f96c32ef45 100644
--- a/cpukit/posix/src/sigtimedwait.c
+++ b/cpukit/posix/src/sigtimedwait.c
@@ -156,12 +156,15 @@ int sigtimedwait(
 
   executing->Wait.option          = *set;
   executing->Wait.return_argument = the_info;
+  _Thread_queue_Context_set_thread_state(
+    &queue_context,
+    STATES_WAITING_FOR_SIGNAL | STATES_INTERRUPTIBLE_BY_SIGNAL
+  );
   _Thread_queue_Context_set_do_nothing_enqueue_callout( &queue_context );
   _Thread_queue_Enqueue(
     &_POSIX_signals_Wait_queue.Queue,
     POSIX_SIGNALS_TQ_OPERATIONS,
     executing,
-    STATES_WAITING_FOR_SIGNAL | STATES_INTERRUPTIBLE_BY_SIGNAL,
     &queue_context
   );
 
diff --git a/cpukit/rtems/src/regiongetsegment.c b/cpukit/rtems/src/regiongetsegment.c
index 04be73bc4b..dfd476ccc0 100644
--- a/cpukit/rtems/src/regiongetsegment.c
+++ b/cpukit/rtems/src/regiongetsegment.c
@@ -87,6 +87,10 @@ rtems_status_code rtems_region_get_segment(
       executing->Wait.return_argument = segment;
 
       /* FIXME: This is a home grown condition variable */
+      _Thread_queue_Context_set_thread_state(
+        &queue_context,
+        STATES_WAITING_FOR_SEGMENT
+      );
       _Thread_queue_Context_set_enqueue_callout(
         &queue_context,
         _Region_Enqueue_callout
@@ -96,7 +100,6 @@ rtems_status_code rtems_region_get_segment(
         &the_region->Wait_queue.Queue,
         the_region->wait_operations,
         executing,
-        STATES_WAITING_FOR_SEGMENT,
         &queue_context
       );
       return _Status_Get_after_wait( executing );
diff --git a/cpukit/score/include/rtems/score/coresemimpl.h b/cpukit/score/include/rtems/score/coresemimpl.h
index 7f5b58e76b..20ca30b366 100644
--- a/cpukit/score/include/rtems/score/coresemimpl.h
+++ b/cpukit/score/include/rtems/score/coresemimpl.h
@@ -184,12 +184,15 @@ RTEMS_INLINE_ROUTINE Status_Control _CORE_semaphore_Seize(
     return STATUS_UNSATISFIED;
   }
 
+  _Thread_queue_Context_set_thread_state(
+    queue_context,
+    STATES_WAITING_FOR_SEMAPHORE
+  );
   _Thread_queue_Context_set_do_nothing_enqueue_callout( queue_context );
   _Thread_queue_Enqueue(
     &the_semaphore->Wait_queue.Queue,
     operations,
     executing,
-    STATES_WAITING_FOR_SEMAPHORE,
     queue_context
   );
   return _Thread_Wait_get_status( executing );
diff --git a/cpukit/score/include/rtems/score/threadq.h b/cpukit/score/include/rtems/score/threadq.h
index 61ce5fbee2..dec8e13e9e 100644
--- a/cpukit/score/include/rtems/score/threadq.h
+++ b/cpukit/score/include/rtems/score/threadq.h
@@ -24,6 +24,7 @@
 #include <rtems/score/object.h>
 #include <rtems/score/priority.h>
 #include <rtems/score/rbtree.h>
+#include <rtems/score/states.h>
 #include <rtems/score/watchdog.h>
 
 #ifdef __cplusplus
@@ -193,6 +194,11 @@ struct Thread_queue_Context {
   Thread_queue_Lock_context Lock_context;
 
   /**
+   * @brief The thread state for _Thread_queue_Enqueue().
+   */
+  States_Control thread_state;
+
+  /**
    * @brief The enqueue callout for _Thread_queue_Enqueue().
    *
    * The callout is invoked after the release of the thread queue lock with
diff --git a/cpukit/score/include/rtems/score/threadqimpl.h b/cpukit/score/include/rtems/score/threadqimpl.h
index 6e3c044aed..bb35b1718b 100644
--- a/cpukit/score/include/rtems/score/threadqimpl.h
+++ b/cpukit/score/include/rtems/score/threadqimpl.h
@@ -96,6 +96,24 @@ RTEMS_INLINE_ROUTINE void _Thread_queue_Context_initialize(
 }
 
 /**
+ * @brief Sets the thread state for the thread to enqueue in the thread queue
+ * context.
+ *
+ * @param queue_context The thread queue context.
+ * @param state The thread state.
+ *
+ * @see _Thread_queue_Enqueue().
+ */
+RTEMS_INLINE_ROUTINE void
+_Thread_queue_Context_set_thread_state(
+  Thread_queue_Context *queue_context,
+  States_Control        thread_state
+)
+{
+  queue_context->thread_state = thread_state;
+}
+
+/**
  * @brief Sets the enqueue callout in the thread queue context.
  *
  * @param queue_context The thread queue context.
@@ -557,6 +575,20 @@ Thread_Control *_Thread_queue_Do_dequeue(
  * to protect the state of objects embedding the thread queue and directly
  * enter _Thread_queue_Enqueue() in case the thread must block.
  *
+ * The thread queue context must be set up with the following functions,
+ * otherwise the behaviour is unpredictable
+ *
+ * - _Thread_queue_Context_set_thread_state(),
+ *
+ * - _Thread_queue_Context_set_enqueue_callout() or
+ *   _Thread_queue_Context_set_do_nothing_enqueue_callout(),
+ *
+ * - _Thread_queue_Context_set_no_timeout() or
+ *   _Thread_queue_Context_set_relative_timeout() or
+ *   _Thread_queue_Context_set_absolute_timeout(), and
+ *
+ * - _Thread_queue_Context_set_deadlock_callout().
+ *
  * @code
  * #include <rtems/score/threadqimpl.h>
  * #include <rtems/score/statesimpl.h>
@@ -564,8 +596,7 @@ Thread_Control *_Thread_queue_Do_dequeue(
  * #define MUTEX_TQ_OPERATIONS &_Thread_queue_Operations_priority
  *
  * typedef struct {
- *   Thread_queue_Control  Queue;
- *   Thread_Control       *owner;
+ *   Thread_queue_Control Queue;
  * } Mutex;
  *
  * void _Mutex_Obtain( Mutex *mutex )
@@ -578,17 +609,24 @@ Thread_Control *_Thread_queue_Do_dequeue(
  *
  *   executing = _Thread_Executing;
  *
- *   if ( mutex->owner == NULL ) {
- *     mutex->owner = executing;
+ *   if ( mutex->Queue.owner == NULL ) {
+ *     mutex->Queue.owner = executing;
  *     _Thread_queue_Release( &mutex->Queue, queue_context );
  *   } else {
+ *     _Thread_queue_Context_set_thread_state(
+ *       &queue_context,
+ *       STATES_WAITING_FOR_MUTEX
+ *     );
  *     _Thread_queue_Context_set_do_nothing_enqueue_callout( &queue_context );
+ *     _Thread_queue_Context_set_no_timeout( &queue_context );
+ *     _Thread_queue_Context_set_deadlock_callout(
+ *       queue_context,
+ *       _Thread_queue_Deadlock_fatal
+ *     );
  *     _Thread_queue_Enqueue(
  *       &mutex->Queue.Queue,
  *       MUTEX_TQ_OPERATIONS,
  *       executing,
- *       STATES_WAITING_FOR_MUTEX,
- *       0,
  *       &queue_context
  *     );
  *   }
@@ -598,14 +636,12 @@ Thread_Control *_Thread_queue_Do_dequeue(
  * @param[in] queue The actual thread queue.
  * @param[in] operations The thread queue operations.
  * @param[in] the_thread The thread to enqueue.
- * @param[in] state The new state of the thread.
  * @param[in] queue_context The thread queue context of the lock acquire.
  */
 void _Thread_queue_Enqueue(
   Thread_queue_Queue            *queue,
   const Thread_queue_Operations *operations,
   Thread_Control                *the_thread,
-  States_Control                 state,
   Thread_queue_Context          *queue_context
 );
 
diff --git a/cpukit/score/src/condition.c b/cpukit/score/src/condition.c
index aaf2f4bd67..ea363169fe 100644
--- a/cpukit/score/src/condition.c
+++ b/cpukit/score/src/condition.c
@@ -109,6 +109,10 @@ static Thread_Control *_Condition_Do_wait(
   context->mutex = _mutex;
   condition = _Condition_Get( _condition );
   executing = _Condition_Queue_acquire_critical( condition, &context->Base );
+  _Thread_queue_Context_set_thread_state(
+    &context->Base,
+    STATES_WAITING_FOR_SYS_LOCK_CONDITION
+  );
   _Thread_queue_Context_set_enqueue_callout(
     &context->Base,
     _Condition_Enqueue_callout
@@ -117,7 +121,6 @@ static Thread_Control *_Condition_Do_wait(
     &condition->Queue.Queue,
     CONDITION_TQ_OPERATIONS,
     executing,
-    STATES_WAITING_FOR_SYS_LOCK_CONDITION,
     &context->Base
   );
 
diff --git a/cpukit/score/src/corebarrierwait.c b/cpukit/score/src/corebarrierwait.c
index 4e8709d355..f47c039e7b 100644
--- a/cpukit/score/src/corebarrierwait.c
+++ b/cpukit/score/src/corebarrierwait.c
@@ -44,12 +44,15 @@ Status_Control _CORE_barrier_Seize(
     return STATUS_BARRIER_AUTOMATICALLY_RELEASED;
   } else {
     the_barrier->number_of_waiting_threads = number_of_waiting_threads;
+    _Thread_queue_Context_set_thread_state(
+      queue_context,
+      STATES_WAITING_FOR_BARRIER
+    );
     _Thread_queue_Context_set_do_nothing_enqueue_callout( queue_context );
     _Thread_queue_Enqueue(
       &the_barrier->Wait_queue.Queue,
       CORE_BARRIER_TQ_OPERATIONS,
       executing,
-      STATES_WAITING_FOR_BARRIER,
       queue_context
     );
     return _Thread_Wait_get_status( executing );
diff --git a/cpukit/score/src/coremsgseize.c b/cpukit/score/src/coremsgseize.c
index a38f5adf60..b48a3f93d7 100644
--- a/cpukit/score/src/coremsgseize.c
+++ b/cpukit/score/src/coremsgseize.c
@@ -113,12 +113,15 @@ Status_Control _CORE_message_queue_Seize(
   executing->Wait.return_argument = size_p;
   /* Wait.count will be filled in with the message priority */
 
+  _Thread_queue_Context_set_thread_state(
+    queue_context,
+    STATES_WAITING_FOR_MESSAGE
+  );
   _Thread_queue_Context_set_do_nothing_enqueue_callout( queue_context );
   _Thread_queue_Enqueue(
     &the_message_queue->Wait_queue.Queue,
     the_message_queue->operations,
     executing,
-    STATES_WAITING_FOR_MESSAGE,
     queue_context
   );
   return _Thread_Wait_get_status( executing );
diff --git a/cpukit/score/src/coremsgsubmit.c b/cpukit/score/src/coremsgsubmit.c
index 8ea7fcf6c8..3c961014e3 100644
--- a/cpukit/score/src/coremsgsubmit.c
+++ b/cpukit/score/src/coremsgsubmit.c
@@ -131,12 +131,15 @@ Status_Control _CORE_message_queue_Submit(
     executing->Wait.option = (uint32_t) size;
     executing->Wait.count = submit_type;
 
+    _Thread_queue_Context_set_thread_state(
+      queue_context,
+      STATES_WAITING_FOR_MESSAGE
+    );
     _Thread_queue_Context_set_do_nothing_enqueue_callout( queue_context );
     _Thread_queue_Enqueue(
       &the_message_queue->Wait_queue.Queue,
       the_message_queue->operations,
       executing,
-      STATES_WAITING_FOR_MESSAGE,
       queue_context
     );
     return _Thread_Wait_get_status( executing );
diff --git a/cpukit/score/src/coremutexseize.c b/cpukit/score/src/coremutexseize.c
index 2fef463f04..4309380627 100644
--- a/cpukit/score/src/coremutexseize.c
+++ b/cpukit/score/src/coremutexseize.c
@@ -32,6 +32,10 @@ Status_Control _CORE_mutex_Seize_slow(
 )
 {
   if ( wait ) {
+    _Thread_queue_Context_set_thread_state(
+      queue_context,
+      STATES_WAITING_FOR_MUTEX
+    );
     _Thread_queue_Context_set_do_nothing_enqueue_callout( queue_context );
     _Thread_queue_Context_set_deadlock_callout(
       queue_context,
@@ -41,7 +45,6 @@ Status_Control _CORE_mutex_Seize_slow(
       &the_mutex->Wait_queue.Queue,
       operations,
       executing,
-      STATES_WAITING_FOR_MUTEX,
       queue_context
     );
     return _Thread_Wait_get_status( executing );
diff --git a/cpukit/score/src/corerwlockobtainread.c b/cpukit/score/src/corerwlockobtainread.c
index 001d259265..641945635f 100644
--- a/cpukit/score/src/corerwlockobtainread.c
+++ b/cpukit/score/src/corerwlockobtainread.c
@@ -78,12 +78,15 @@ Status_Control _CORE_RWLock_Seize_for_reading(
 
   executing->Wait.option = CORE_RWLOCK_THREAD_WAITING_FOR_READ;
 
+  _Thread_queue_Context_set_thread_state(
+    queue_context,
+   STATES_WAITING_FOR_RWLOCK
+  );
   _Thread_queue_Context_set_do_nothing_enqueue_callout( queue_context );
   _Thread_queue_Enqueue(
      &the_rwlock->Wait_queue.Queue,
      CORE_RWLOCK_TQ_OPERATIONS,
      executing,
-     STATES_WAITING_FOR_RWLOCK,
      queue_context
   );
   return _Thread_Wait_get_status( executing );
diff --git a/cpukit/score/src/corerwlockobtainwrite.c b/cpukit/score/src/corerwlockobtainwrite.c
index ce47f434b7..7f636daa99 100644
--- a/cpukit/score/src/corerwlockobtainwrite.c
+++ b/cpukit/score/src/corerwlockobtainwrite.c
@@ -66,12 +66,15 @@ Status_Control _CORE_RWLock_Seize_for_writing(
 
   executing->Wait.option = CORE_RWLOCK_THREAD_WAITING_FOR_WRITE;
 
+  _Thread_queue_Context_set_thread_state(
+    queue_context,
+    STATES_WAITING_FOR_RWLOCK
+  );
   _Thread_queue_Context_set_do_nothing_enqueue_callout( queue_context );
   _Thread_queue_Enqueue(
      &the_rwlock->Wait_queue.Queue,
      CORE_RWLOCK_TQ_OPERATIONS,
      executing,
-     STATES_WAITING_FOR_RWLOCK,
      queue_context
   );
   return _Thread_Wait_get_status( executing );
diff --git a/cpukit/score/src/futex.c b/cpukit/score/src/futex.c
index efac831df5..f200895562 100644
--- a/cpukit/score/src/futex.c
+++ b/cpukit/score/src/futex.c
@@ -92,6 +92,10 @@ int _Futex_Wait( struct _Futex_Control *_futex, int *uaddr, int val )
   executing = _Futex_Queue_acquire_critical( futex, &queue_context );
 
   if ( *uaddr == val ) {
+    _Thread_queue_Context_set_thread_state(
+      &queue_context,
+      STATES_WAITING_FOR_SYS_LOCK_FUTEX
+    );
     _Thread_queue_Context_set_do_nothing_enqueue_callout( &queue_context );
     _Thread_queue_Context_set_no_timeout( &queue_context );
     _Thread_queue_Context_set_ISR_level( &queue_context, level );
@@ -99,7 +103,6 @@ int _Futex_Wait( struct _Futex_Control *_futex, int *uaddr, int val )
       &futex->Queue.Queue,
       FUTEX_TQ_OPERATIONS,
       executing,
-      STATES_WAITING_FOR_SYS_LOCK_FUTEX,
       &queue_context
     );
     eno = 0;
diff --git a/cpukit/score/src/mpci.c b/cpukit/score/src/mpci.c
index 9981ab99ad..2f0a2ded2b 100644
--- a/cpukit/score/src/mpci.c
+++ b/cpukit/score/src/mpci.c
@@ -253,6 +253,10 @@ Status_Control _MPCI_Send_request_packet(
       the_packet->timeout = _MPCI_table->default_timeout;
 
   _Thread_queue_Context_initialize( &queue_context );
+  _Thread_queue_Context_set_thread_state(
+    &queue_context,
+    STATES_WAITING_FOR_RPC_REPLY | extra_state
+  );
   _Thread_queue_Context_set_enqueue_callout(
     &queue_context,
     _MPCI_Enqueue_callout
@@ -276,7 +280,6 @@ Status_Control _MPCI_Send_request_packet(
     &_MPCI_Remote_blocked_threads.Queue,
     &_Thread_queue_Operations_FIFO,
     executing,
-    STATES_WAITING_FOR_RPC_REPLY | extra_state,
     &queue_context
   );
   return _Thread_Wait_get_status( executing );
diff --git a/cpukit/score/src/mutex.c b/cpukit/score/src/mutex.c
index 7c2388df14..edd2e41dfd 100644
--- a/cpukit/score/src/mutex.c
+++ b/cpukit/score/src/mutex.c
@@ -109,6 +109,10 @@ static void _Mutex_Acquire_slow(
   Thread_queue_Context *queue_context
 )
 {
+  _Thread_queue_Context_set_thread_state(
+    queue_context,
+    STATES_WAITING_FOR_SYS_LOCK_MUTEX
+  );
   _Thread_queue_Context_set_do_nothing_enqueue_callout( queue_context );
   _Thread_queue_Context_set_deadlock_callout(
     queue_context,
@@ -119,7 +123,6 @@ static void _Mutex_Acquire_slow(
     &mutex->Queue.Queue,
     MUTEX_TQ_OPERATIONS,
     executing,
-    STATES_WAITING_FOR_SYS_LOCK_MUTEX,
     queue_context
   );
 }
diff --git a/cpukit/score/src/semaphore.c b/cpukit/score/src/semaphore.c
index 27d3dbfc2f..df33532108 100644
--- a/cpukit/score/src/semaphore.c
+++ b/cpukit/score/src/semaphore.c
@@ -103,6 +103,10 @@ void _Semaphore_Wait( struct _Semaphore_Control *_sem )
     sem->count = count - 1;
     _Semaphore_Queue_release( sem, level, &queue_context );
   } else {
+    _Thread_queue_Context_set_thread_state(
+      &queue_context,
+      STATES_WAITING_FOR_SYS_LOCK_SEMAPHORE
+    );
     _Thread_queue_Context_set_do_nothing_enqueue_callout( &queue_context );
     _Thread_queue_Context_set_no_timeout( &queue_context );
     _Thread_queue_Context_set_ISR_level( &queue_context, level );
@@ -110,7 +114,6 @@ void _Semaphore_Wait( struct _Semaphore_Control *_sem )
       &sem->Queue.Queue,
       SEMAPHORE_TQ_OPERATIONS,
       executing,
-      STATES_WAITING_FOR_SYS_LOCK_SEMAPHORE,
       &queue_context
     );
   }
diff --git a/cpukit/score/src/threadqenqueue.c b/cpukit/score/src/threadqenqueue.c
index f04f44c9c0..7ffef336b4 100644
--- a/cpukit/score/src/threadqenqueue.c
+++ b/cpukit/score/src/threadqenqueue.c
@@ -414,7 +414,6 @@ void _Thread_queue_Enqueue(
   Thread_queue_Queue            *queue,
   const Thread_queue_Operations *operations,
   Thread_Control                *the_thread,
-  States_Control                 state,
   Thread_queue_Context          *queue_context
 )
 {
@@ -423,7 +422,7 @@ void _Thread_queue_Enqueue(
 
 #if defined(RTEMS_MULTIPROCESSING)
   if ( _Thread_MP_Is_receive( the_thread ) && the_thread->receive_packet ) {
-    the_thread = _Thread_MP_Allocate_proxy( state );
+    the_thread = _Thread_MP_Allocate_proxy( queue_context->thread_state );
   }
 #endif
 
@@ -456,7 +455,7 @@ void _Thread_queue_Enqueue(
   /*
    *  Set the blocking state for this thread queue in the thread.
    */
-  _Thread_Set_state( the_thread, state );
+  _Thread_Set_state( the_thread, queue_context->thread_state );
 
   /*
    *  If the thread wants to timeout, then schedule its timer.
diff --git a/cpukit/score/src/threadrestart.c b/cpukit/score/src/threadrestart.c
index dd2e0b8b2d..fc934f6529 100644
--- a/cpukit/score/src/threadrestart.c
+++ b/cpukit/score/src/threadrestart.c
@@ -449,11 +449,11 @@ void _Thread_Join(
   executing->Wait.return_argument = NULL;
 #endif
 
+  _Thread_queue_Context_set_thread_state( queue_context, waiting_for_join );
   _Thread_queue_Enqueue(
     &the_thread->Join_queue.Queue,
     THREAD_JOIN_TQ_OPERATIONS,
     executing,
-    waiting_for_join,
     queue_context
   );
 }