[PATCH 12/27] score: Rework affine ready queue handling

Sebastian Huber sebastian.huber at embedded-brains.de
Mon Nov 15 17:12:44 UTC 2021


Rework the handling of the affine ready queue for the EDF SMP scheduler.
Do the queue handling in the node insert, move, and extract operations.
Remove the queue handling from _Scheduler_EDF_SMP_Allocate_processor().

Update #4531.
---
 cpukit/include/rtems/score/scheduleredfsmp.h |  13 ++-
 cpukit/score/src/scheduleredfsmp.c           | 106 ++++++++++++-------
 2 files changed, 77 insertions(+), 42 deletions(-)

diff --git a/cpukit/include/rtems/score/scheduleredfsmp.h b/cpukit/include/rtems/score/scheduleredfsmp.h
index 85e438e81d..e749b3d419 100644
--- a/cpukit/include/rtems/score/scheduleredfsmp.h
+++ b/cpukit/include/rtems/score/scheduleredfsmp.h
@@ -79,9 +79,18 @@ typedef struct {
   RBTree_Control Queue;
 
   /**
-   * @brief The scheduled thread of the corresponding processor.
+   * @brief If this member is not NULL, then it references the scheduled thread
+   *   affine only to the corresponding processor, otherwise the processor is
+   *   allocated to a thread which may execute on any of the processors owned
+   *   by the scheduler.
    */
-  Scheduler_EDF_SMP_Node *scheduled;
+  Scheduler_EDF_SMP_Node *affine_scheduled;
+
+  /**
+   * @brief This member referneces the thread allocated to the corresponding
+   *   processor.
+   */
+  Scheduler_EDF_SMP_Node *allocated;
 } Scheduler_EDF_SMP_Ready_queue;
 
 typedef struct {
diff --git a/cpukit/score/src/scheduleredfsmp.c b/cpukit/score/src/scheduleredfsmp.c
index a915dbe511..7da777e87a 100644
--- a/cpukit/score/src/scheduleredfsmp.c
+++ b/cpukit/score/src/scheduleredfsmp.c
@@ -196,21 +196,21 @@ static inline Scheduler_Node *_Scheduler_EDF_SMP_Get_highest_ready(
   return &highest_ready->Base.Base;
 }
 
-static inline void _Scheduler_EDF_SMP_Set_scheduled(
+static inline void _Scheduler_EDF_SMP_Set_allocated(
   Scheduler_EDF_SMP_Context *self,
-  Scheduler_EDF_SMP_Node    *scheduled,
+  Scheduler_EDF_SMP_Node    *allocated,
   const Per_CPU_Control     *cpu
 )
 {
-  self->Ready[ _Per_CPU_Get_index( cpu ) + 1 ].scheduled = scheduled;
+  self->Ready[ _Per_CPU_Get_index( cpu ) + 1 ].allocated = allocated;
 }
 
-static inline Scheduler_EDF_SMP_Node *_Scheduler_EDF_SMP_Get_scheduled(
+static inline Scheduler_EDF_SMP_Node *_Scheduler_EDF_SMP_Get_allocated(
   const Scheduler_EDF_SMP_Context *self,
   uint8_t                          rqi
 )
 {
-  return self->Ready[ rqi ].scheduled;
+  return self->Ready[ rqi ].allocated;
 }
 
 static inline Scheduler_Node *_Scheduler_EDF_SMP_Get_lowest_scheduled(
@@ -226,20 +226,62 @@ static inline Scheduler_Node *_Scheduler_EDF_SMP_Get_lowest_scheduled(
 
   if ( rqi != 0 ) {
     Scheduler_EDF_SMP_Context *self;
-    Scheduler_EDF_SMP_Node    *node;
+    Scheduler_EDF_SMP_Node    *affine_scheduled;
 
     self = _Scheduler_EDF_SMP_Get_self( context );
-    node = _Scheduler_EDF_SMP_Get_scheduled( self, rqi );
+    affine_scheduled = self->Ready[ rqi ].affine_scheduled;
 
-    if ( node->ready_queue_index > 0 ) {
-      _Assert( node->ready_queue_index == rqi );
-      return &node->Base.Base;
+    if ( affine_scheduled != NULL ) {
+      _Assert( affine_scheduled->ready_queue_index == rqi );
+      return &affine_scheduled->Base.Base;
     }
   }
 
   return _Scheduler_SMP_Get_lowest_scheduled( context, filter_base );
 }
 
+static inline void _Scheduler_EDF_SMP_Insert_scheduled(
+  Scheduler_Context *context,
+  Scheduler_Node    *node_base,
+  Priority_Control   priority_to_insert
+)
+{
+  Scheduler_EDF_SMP_Context     *self;
+  Scheduler_EDF_SMP_Node        *node;
+  uint8_t                        rqi;
+  Scheduler_EDF_SMP_Ready_queue *ready_queue;
+
+  self = _Scheduler_EDF_SMP_Get_self( context );
+  node = _Scheduler_EDF_SMP_Node_downcast( node_base );
+  rqi = node->ready_queue_index;
+  ready_queue = &self->Ready[ rqi ];
+
+  _Scheduler_SMP_Insert_scheduled( context, node_base, priority_to_insert );
+
+  if ( rqi != 0 ) {
+    ready_queue->affine_scheduled = node;
+
+    if ( !_RBTree_Is_empty( &ready_queue->Queue ) ) {
+      _Chain_Extract_unprotected( &ready_queue->Node );
+    }
+  }
+}
+
+static inline void _Scheduler_EDF_SMP_Activate_ready_queue_if_necessary(
+  Scheduler_EDF_SMP_Context     *self,
+  uint8_t                        rqi,
+  Scheduler_EDF_SMP_Ready_queue *ready_queue
+)
+{
+  if (
+    rqi != 0 &&
+    _RBTree_Is_empty( &ready_queue->Queue ) &&
+    ready_queue->affine_scheduled == NULL
+  ) {
+    _Chain_Append_unprotected( &self->Affine_queues, &ready_queue->Node );
+  }
+}
+
 static inline void _Scheduler_EDF_SMP_Insert_ready(
   Scheduler_Context *context,
   Scheduler_Node    *node_base,
@@ -265,6 +307,7 @@ static inline void _Scheduler_EDF_SMP_Insert_ready(
   node->generation = generation;
   self->generations[ generation_index ] = generation + increment;
 
+  _Scheduler_EDF_SMP_Activate_ready_queue_if_necessary( self, rqi, ready_queue );
   _RBTree_Initialize_node( &node->Base.Base.Node.RBTree );
   _RBTree_Insert_inline(
     &ready_queue->Queue,
@@ -272,16 +315,6 @@ static inline void _Scheduler_EDF_SMP_Insert_ready(
     &insert_priority,
     _Scheduler_EDF_SMP_Priority_less_equal
   );
-
-  if ( rqi != 0 && _Chain_Is_node_off_chain( &ready_queue->Node ) ) {
-    Scheduler_EDF_SMP_Node *scheduled;
-
-    scheduled = _Scheduler_EDF_SMP_Get_scheduled( self, rqi );
-
-    if ( scheduled->ready_queue_index == 0 ) {
-      _Chain_Append_unprotected( &self->Affine_queues, &ready_queue->Node );
-    }
-  }
 }
 
 static inline void _Scheduler_EDF_SMP_Extract_from_scheduled(
@@ -305,6 +338,8 @@ static inline void _Scheduler_EDF_SMP_Extract_from_scheduled(
   if ( rqi != 0 && !_RBTree_Is_empty( &ready_queue->Queue ) ) {
     _Chain_Append_unprotected( &self->Affine_queues, &ready_queue->Node );
   }
+
+  ready_queue->affine_scheduled = NULL;
 }
 
 static inline void _Scheduler_EDF_SMP_Extract_from_ready(
@@ -328,10 +363,9 @@ static inline void _Scheduler_EDF_SMP_Extract_from_ready(
   if (
     rqi != 0
       && _RBTree_Is_empty( &ready_queue->Queue )
-      && !_Chain_Is_node_off_chain( &ready_queue->Node )
+      && ready_queue->affine_scheduled == NULL
   ) {
     _Chain_Extract_unprotected( &ready_queue->Node );
-    _Chain_Set_off_chain( &ready_queue->Node );
   }
 }
 
@@ -342,7 +376,7 @@ static inline void _Scheduler_EDF_SMP_Move_from_scheduled_to_ready(
 {
   Priority_Control insert_priority;
 
-  _Scheduler_SMP_Extract_from_scheduled( context, scheduled_to_ready );
+  _Scheduler_EDF_SMP_Extract_from_scheduled( context, scheduled_to_ready );
   insert_priority = _Scheduler_SMP_Node_priority( scheduled_to_ready );
   _Scheduler_EDF_SMP_Insert_ready(
     context,
@@ -361,7 +395,7 @@ static inline void _Scheduler_EDF_SMP_Move_from_ready_to_scheduled(
   _Scheduler_EDF_SMP_Extract_from_ready( context, ready_to_scheduled );
   insert_priority = _Scheduler_SMP_Node_priority( ready_to_scheduled );
   insert_priority = SCHEDULER_PRIORITY_APPEND( insert_priority );
-  _Scheduler_SMP_Insert_scheduled(
+  _Scheduler_EDF_SMP_Insert_scheduled(
     context,
     ready_to_scheduled,
     insert_priority
@@ -385,24 +419,16 @@ static inline void _Scheduler_EDF_SMP_Allocate_processor(
   rqi = scheduled->ready_queue_index;
 
   if ( rqi != 0 ) {
-    Scheduler_EDF_SMP_Ready_queue *ready_queue;
-    Per_CPU_Control               *desired_cpu;
-
-    ready_queue = &self->Ready[ rqi ];
-
-    if ( !_Chain_Is_node_off_chain( &ready_queue->Node ) ) {
-      _Chain_Extract_unprotected( &ready_queue->Node );
-      _Chain_Set_off_chain( &ready_queue->Node );
-    }
+    Per_CPU_Control *desired_cpu;
 
     desired_cpu = _Per_CPU_Get_by_index( rqi - 1 );
 
     if ( victim_cpu != desired_cpu ) {
       Scheduler_EDF_SMP_Node *node;
 
-      node = _Scheduler_EDF_SMP_Get_scheduled( self, rqi );
+      node = _Scheduler_EDF_SMP_Get_allocated( self, rqi );
       _Assert( node->ready_queue_index == 0 );
-      _Scheduler_EDF_SMP_Set_scheduled( self, node, victim_cpu );
+      _Scheduler_EDF_SMP_Set_allocated( self, node, victim_cpu );
       _Scheduler_SMP_Allocate_processor_exact(
         context,
         &node->Base.Base,
@@ -413,7 +439,7 @@ static inline void _Scheduler_EDF_SMP_Allocate_processor(
     }
   }
 
-  _Scheduler_EDF_SMP_Set_scheduled( self, scheduled, victim_cpu );
+  _Scheduler_EDF_SMP_Set_allocated( self, scheduled, victim_cpu );
   _Scheduler_SMP_Allocate_processor_exact(
     context,
     &scheduled->Base.Base,
@@ -454,7 +480,7 @@ static inline bool _Scheduler_EDF_SMP_Enqueue(
     insert_priority,
     _Scheduler_SMP_Priority_less_equal,
     _Scheduler_EDF_SMP_Insert_ready,
-    _Scheduler_SMP_Insert_scheduled,
+    _Scheduler_EDF_SMP_Insert_scheduled,
     _Scheduler_EDF_SMP_Move_from_scheduled_to_ready,
     _Scheduler_EDF_SMP_Get_lowest_scheduled,
     _Scheduler_EDF_SMP_Allocate_processor
@@ -475,7 +501,7 @@ static inline void _Scheduler_EDF_SMP_Enqueue_scheduled(
     _Scheduler_EDF_SMP_Extract_from_ready,
     _Scheduler_EDF_SMP_Get_highest_ready,
     _Scheduler_EDF_SMP_Insert_ready,
-    _Scheduler_SMP_Insert_scheduled,
+    _Scheduler_EDF_SMP_Insert_scheduled,
     _Scheduler_EDF_SMP_Move_from_ready_to_scheduled,
     _Scheduler_EDF_SMP_Allocate_processor
   );
@@ -510,7 +536,7 @@ static inline bool _Scheduler_EDF_SMP_Do_ask_for_help(
     node,
     _Scheduler_SMP_Priority_less_equal,
     _Scheduler_EDF_SMP_Insert_ready,
-    _Scheduler_SMP_Insert_scheduled,
+    _Scheduler_EDF_SMP_Insert_scheduled,
     _Scheduler_EDF_SMP_Move_from_scheduled_to_ready,
     _Scheduler_EDF_SMP_Get_lowest_scheduled,
     _Scheduler_EDF_SMP_Allocate_processor
@@ -598,7 +624,7 @@ static inline void _Scheduler_EDF_SMP_Register_idle(
 
   self = _Scheduler_EDF_SMP_Get_self( context );
   idle = _Scheduler_EDF_SMP_Node_downcast( idle_base );
-  _Scheduler_EDF_SMP_Set_scheduled( self, idle, cpu );
+  _Scheduler_EDF_SMP_Set_allocated( self, idle, cpu );
 }
 
 void _Scheduler_EDF_SMP_Add_processor(
-- 
2.26.2



More information about the devel mailing list