sparc: Add atomic support for SPARC V8

Use SWAP instruction with one lock for the system in the SMP case.

cpukit/score/cpu/sparc/Makefile.am

@@ -12,6 +12,7 @@ include_rtems_score_HEADERS += rtems/score/cpusmplock.h
 
 noinst_LIBRARIES = libscorecpu.a
 libscorecpu_a_SOURCES = cpu.c cpu_asm.S
+libscorecpu_a_SOURCES += sparcv8-atomic.c
 libscorecpu_a_CPPFLAGS = $(AM_CPPFLAGS)
 
 include $(srcdir)/preinstall.am

cpukit/score/cpu/sparc/sparcv8-atomic.c

@@ -0,0 +1,202 @@
+/*
+ * Copyright (c) 2014 embedded brains GmbH.  All rights reserved.
+ *
+ *  embedded brains GmbH
+ *  Dornierstr. 4
+ *  82178 Puchheim
+ *  Germany
+ *  <rtems@embedded-brains.de>
+ *
+ * The license and distribution terms for this file may be
+ * found in the file LICENSE in this distribution or at
+ * http://www.rtems.com/license/LICENSE.
+ */
+
+#if HAVE_CONFIG_H
+  #include "config.h"
+#endif
+
+#include <rtems/score/cpuopts.h>
+#include <rtems/score/isrlevel.h>
+
+/*
+ * This file is a dirty hack.  A proper solution would be to add RTEMS support
+ * for libatomic in GCC (see also libatomic/configure.tgt).
+ */
+
+#if defined(RTEMS_SMP)
+static volatile uint32_t _SPARCV8_The_one_lock;
+
+static inline uint32_t _SPARCV8_Atomic_swap(
+  volatile uint32_t *address,
+  uint32_t value
+)
+{
+  asm volatile (
+    "swap [%2], %0"
+    : "=r" (value)
+    : "0" (value), "r" (address)
+    : "memory"
+  );
+
+  return value;
+}
+#endif
+
+static ISR_Level _SPARCV8_Acquire_the_one_lock( void )
+{
+  ISR_Level level;
+
+  _ISR_Disable_without_giant( level );
+
+#if defined(RTEMS_SMP)
+  do {
+    while ( _SPARCV8_The_one_lock ) {
+      /* Wait */
+    }
+  } while ( _SPARCV8_Atomic_swap( &_SPARCV8_The_one_lock, 1 ) );
+#endif
+
+  return level;
+}
+
+static void _SPARCV8_Release_the_one_lock( ISR_Level level )
+{
+#if defined(RTEMS_SMP)
+  _SPARCV8_The_one_lock = 0;
+#endif
+  _ISR_Enable_without_giant( level );
+}
+
+uint8_t __atomic_exchange_1( uint8_t *mem, uint8_t val, int model )
+{
+  uint8_t prev;
+  ISR_Level level;
+
+  level = _SPARCV8_Acquire_the_one_lock();
+
+  prev = *mem;
+  *mem = val;
+
+  _SPARCV8_Release_the_one_lock( level );
+
+  return prev;
+}
+
+uint32_t __atomic_exchange_4( uint32_t *mem, uint32_t val, int model )
+{
+  uint32_t prev;
+  ISR_Level level;
+
+  level = _SPARCV8_Acquire_the_one_lock();
+
+  prev = *mem;
+  *mem = val;
+
+  _SPARCV8_Release_the_one_lock( level );
+
+  return prev;
+}
+
+/*
+ * You get probably a warning here which can be ignored: "warning: conflicting
+ * types for built-in function '__atomic_compare_exchange_4' [enabled by
+ * default]"
+ */
+bool __atomic_compare_exchange_4(
+  uint32_t *mem,
+  uint32_t *expected,
+  uint32_t desired,
+  int success,
+  int failure
+)
+{
+  bool equal;
+  ISR_Level level;
+
+  level = _SPARCV8_Acquire_the_one_lock();
+
+  equal = *mem == *expected;
+  if ( equal ) {
+    *mem = desired;
+  }
+
+  _SPARCV8_Release_the_one_lock( level );
+
+  return equal;
+}
+
+uint32_t __atomic_fetch_add_4( uint32_t *mem, uint32_t val, int model )
+{
+  uint32_t prev;
+  ISR_Level level;
+
+  level = _SPARCV8_Acquire_the_one_lock();
+
+  prev = *mem;
+  *mem = prev + val;
+
+  _SPARCV8_Release_the_one_lock( level );
+
+  return prev;
+}
+
+uint32_t __atomic_fetch_sub_4( uint32_t *mem, uint32_t val, int model )
+{
+  uint32_t prev;
+  ISR_Level level;
+
+  level = _SPARCV8_Acquire_the_one_lock();
+
+  prev = *mem;
+  *mem = prev - val;
+
+  _SPARCV8_Release_the_one_lock( level );
+
+  return prev;
+}
+
+uint32_t __atomic_fetch_and_4( uint32_t *mem, uint32_t val, int model )
+{
+  uint32_t prev;
+  ISR_Level level;
+
+  level = _SPARCV8_Acquire_the_one_lock();
+
+  prev = *mem;
+  *mem = prev & val;
+
+  _SPARCV8_Release_the_one_lock( level );
+
+  return prev;
+}
+
+uint32_t __atomic_fetch_or_4( uint32_t *mem, uint32_t val, int model )
+{
+  uint32_t prev;
+  ISR_Level level;
+
+  level = _SPARCV8_Acquire_the_one_lock();
+
+  prev = *mem;
+  *mem = prev | val;
+
+  _SPARCV8_Release_the_one_lock( level );
+
+  return prev;
+}
+
+uint32_t __atomic_fetch_xor_4( uint32_t *mem, uint32_t val, int model )
+{
+  uint32_t prev;
+  ISR_Level level;
+
+  level = _SPARCV8_Acquire_the_one_lock();
+
+  prev = *mem;
+  *mem = prev ^ val;
+
+  _SPARCV8_Release_the_one_lock( level );
+
+  return prev;
+}