score: Simplify FP context allocation

Use the stack area to allocate the FP context.  This considerably
simplifies the application configuration since the task count no longer
influences the configured work space size.  With this change the stack
space size is overestimated since an FP context for each thread is
accounted.  Memory constraint applications can use the stack size for
fine tuning.

Update #3835.

cpukit/include/rtems/confdefs.h

@@ -30,6 +30,7 @@
 #include <rtems/ioimpl.h>
 #include <rtems/sysinit.h>
 #include <rtems/score/apimutex.h>
+#include <rtems/score/context.h>
 #include <rtems/score/percpu.h>
 #include <rtems/score/userextimpl.h>
 #include <rtems/score/wkspace.h>
@@ -1315,10 +1316,10 @@ extern rtems_initialization_tasks_table Initialization_tasks[];
  */
 #ifdef CONFIGURE_TASK_STACK_FROM_ALLOCATOR
   #define _Configure_From_stackspace(_stack_size) \
-    CONFIGURE_TASK_STACK_FROM_ALLOCATOR(_stack_size)
+    CONFIGURE_TASK_STACK_FROM_ALLOCATOR(_stack_size + CONTEXT_FP_SIZE)
 #else
   #define _Configure_From_stackspace(_stack_size) \
-    _Configure_From_workspace(_stack_size)
+    _Configure_From_workspace(_stack_size + CONTEXT_FP_SIZE)
 #endif
 
 /**
@@ -2313,16 +2314,6 @@ struct _reent *__getreent(void)
  */
 #ifndef CONFIGURE_EXECUTIVE_RAM_SIZE
 
-/*
- * Account for allocating the following per object
- *   + array of object control structures
- *   + local pointer table -- pointer per object plus a zero'th
- *     entry in the local pointer table.
- */
-#define _CONFIGURE_MEMORY_FOR_TASKS(_tasks, _number_FP_tasks) \
-  (_Configure_Max_Objects(_number_FP_tasks) \
-    * _Configure_From_workspace(CONTEXT_FP_SIZE))
-
 /**
  * The following macro is used to calculate the memory allocated by RTEMS
  * for the message buffers associated with a particular message queue.
@@ -2358,30 +2349,6 @@ struct _reent *__getreent(void)
   #define CONFIGURE_MEMORY_OVERHEAD 0
 #endif
 
-/**
- * This defines the formula used to compute the amount of memory
- * reserved for internal task control structures.
- */
-#if CPU_IDLE_TASK_IS_FP == TRUE
-  #define _CONFIGURE_MEMORY_FOR_INTERNAL_TASKS \
-    _CONFIGURE_MEMORY_FOR_TASKS( \
-      _CONFIGURE_IDLE_TASKS_COUNT + _CONFIGURE_MPCI_RECEIVE_SERVER_COUNT, \
-      _CONFIGURE_IDLE_TASKS_COUNT + _CONFIGURE_MPCI_RECEIVE_SERVER_COUNT \
-    )
-#else
-  #define _CONFIGURE_MEMORY_FOR_INTERNAL_TASKS \
-    _CONFIGURE_MEMORY_FOR_TASKS( \
-      _CONFIGURE_IDLE_TASKS_COUNT + _CONFIGURE_MPCI_RECEIVE_SERVER_COUNT, \
-      _CONFIGURE_MPCI_RECEIVE_SERVER_COUNT \
-    )
-#endif
-
-/**
- * This macro accounts for general RTEMS system overhead.
- */
-#define _CONFIGURE_MEMORY_FOR_SYSTEM_OVERHEAD \
-  _CONFIGURE_MEMORY_FOR_INTERNAL_TASKS
-
 /**
  * This calculates the memory required for the executive workspace.
  *
@@ -2389,11 +2356,6 @@ struct _reent *__getreent(void)
  */
 #define CONFIGURE_EXECUTIVE_RAM_SIZE \
 ( \
-   _CONFIGURE_MEMORY_FOR_SYSTEM_OVERHEAD + \
-   _CONFIGURE_MEMORY_FOR_TASKS( \
-     _CONFIGURE_TASKS, _CONFIGURE_TASKS) + \
-   _CONFIGURE_MEMORY_FOR_TASKS( \
-     CONFIGURE_MAXIMUM_POSIX_THREADS, CONFIGURE_MAXIMUM_POSIX_THREADS) + \
    _CONFIGURE_MEMORY_FOR_POSIX_MESSAGE_QUEUES( \
      CONFIGURE_MAXIMUM_POSIX_MESSAGE_QUEUES) + \
    _CONFIGURE_MEMORY_FOR_POSIX_SEMAPHORES( \

cpukit/include/rtems/score/context.h

@@ -49,7 +49,9 @@ extern "C" {
  *  to store a full floating point context.
  */
 #if ( CPU_HARDWARE_FP == TRUE ) || ( CPU_SOFTWARE_FP == TRUE )
-  #define CONTEXT_FP_SIZE CPU_CONTEXT_FP_SIZE
+  #define CONTEXT_FP_SIZE \
+    ( ( CPU_CONTEXT_FP_SIZE + CPU_HEAP_ALIGNMENT - 1 ) \
+      & ~( CPU_HEAP_ALIGNMENT - 1 ) )
 #else
   #define CONTEXT_FP_SIZE 0
 #endif

cpukit/include/rtems/score/stackimpl.h

@@ -22,6 +22,7 @@
 #define _RTEMS_SCORE_STACKIMPL_H
 
 #include <rtems/score/stack.h>
+#include <rtems/score/context.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -74,15 +75,27 @@ RTEMS_INLINE_ROUTINE uint32_t _Stack_Minimum (void)
  * a valid stack area on this processor, and false otherwise.
  *
  * @param size The stack size to check.
+ * @param is_fp Indicates if the stack is for a floating-point thread.
  *
  * @retval true @a size is large enough.
  * @retval false @a size is not large enough.
  */
-RTEMS_INLINE_ROUTINE bool _Stack_Is_enough (
-  size_t size
+RTEMS_INLINE_ROUTINE bool _Stack_Is_enough(
+  size_t size,
+  bool   is_fp
 )
 {
-  return ( size >= _Stack_Minimum() );
+  size_t minimum;
+
+  minimum = _Stack_Minimum();
+
+#if ( CPU_HARDWARE_FP == TRUE ) || ( CPU_SOFTWARE_FP == TRUE )
+  if ( is_fp ) {
+    minimum += CONTEXT_FP_SIZE;
+  }
+#endif
+
+  return ( size >= minimum );
 }
 
 /**

cpukit/posix/src/pthreadcreate.c

@@ -96,6 +96,12 @@ int pthread_create(
   if ( !the_attr->is_initialized )
     return EINVAL;
 
+  /*
+   *  Currently all POSIX threads are floating point if the hardware
+   *  supports it.
+   */
+  is_fp = true;
+
   /*
    *  Core Thread Initialize ensures we get the minimum amount of
    *  stack space if it is allowed to allocate it itself.
@@ -104,7 +110,7 @@ int pthread_create(
    *        twice the minimum.
    */
   if ( the_attr->stackaddr != NULL ) {
-    if ( !_Stack_Is_enough(the_attr->stacksize) ) {
+    if ( !_Stack_Is_enough( the_attr->stacksize, is_fp ) ) {
       return EINVAL;
     }
 
@@ -191,12 +197,6 @@ int pthread_create(
     return EINVAL;
   }
 
-  /*
-   *  Currently all POSIX threads are floating point if the hardware
-   *  supports it.
-   */
-  is_fp = true;
-
   /*
    *  Allocate the thread control block.
    *

cpukit/score/src/threadinitialize.c

@@ -43,9 +43,6 @@ bool _Thread_Initialize(
 )
 {
   uintptr_t                tls_size = _TLS_Get_size();
-  #if ( CPU_HARDWARE_FP == TRUE ) || ( CPU_SOFTWARE_FP == TRUE )
-    void                  *fp_area = NULL;
-  #endif
   bool                     extension_status;
   size_t                   i;
   Scheduler_Node          *scheduler_node;
@@ -91,6 +88,13 @@ bool _Thread_Initialize(
   if ( stack_area == NULL ) {
 #endif
     stack_size = _Stack_Ensure_minimum( stack_size );
+
+#if ( CPU_HARDWARE_FP == TRUE ) || ( CPU_SOFTWARE_FP == TRUE )
+    if ( is_fp ) {
+      stack_size += CONTEXT_FP_SIZE;
+    }
+#endif
+
     stack_area = _Stack_Allocate( stack_size );
 
     if ( stack_area == NULL ) {
@@ -102,6 +106,16 @@ bool _Thread_Initialize(
   }
 #endif
 
+  /* Allocate floating-point context in stack area */
+#if ( CPU_HARDWARE_FP == TRUE ) || ( CPU_SOFTWARE_FP == TRUE )
+  if ( is_fp ) {
+    the_thread->fp_context = stack_area;
+    the_thread->Start.fp_context = stack_area;
+    stack_size -= CONTEXT_FP_SIZE;
+    stack_area = (char *) stack_area + CONTEXT_FP_SIZE;
+  }
+#endif
+
   _Stack_Initialize(
      &the_thread->Start.Initial_stack,
      stack_area,
@@ -123,19 +137,6 @@ bool _Thread_Initialize(
     }
   }
 
-  /*
-   *  Allocate the floating point area for this thread
-   */
-  #if ( CPU_HARDWARE_FP == TRUE ) || ( CPU_SOFTWARE_FP == TRUE )
-    if ( is_fp ) {
-      fp_area = _Workspace_Allocate( CONTEXT_FP_SIZE );
-      if ( !fp_area )
-        goto failed;
-    }
-    the_thread->fp_context       = fp_area;
-    the_thread->Start.fp_context = fp_area;
-  #endif
-
   /*
    *  Get thread queue heads
    */
@@ -301,16 +302,10 @@ failed:
 #endif
 
   _Workspace_Free( the_thread->Start.tls_area );
-
   _Freechain_Put(
     &information->Thread_queue_heads.Free,
     the_thread->Wait.spare_heads
   );
-
-  #if ( CPU_HARDWARE_FP == TRUE ) || ( CPU_SOFTWARE_FP == TRUE )
-    _Workspace_Free( fp_area );
-  #endif
-
   _Stack_Free( the_thread->Start.allocated_stack );
   return false;
 }

cpukit/score/src/threadrestart.c

@@ -172,8 +172,6 @@ static void _Thread_Free( Thread_Control *the_thread )
   if ( _Thread_Is_allocated_fp( the_thread ) )
     _Thread_Deallocate_fp();
 #endif
-
-  _Workspace_Free( the_thread->Start.fp_context );
 #endif
 
   _Freechain_Put(