*
*/
# ifdef THREADS
+ void GC_noop1 GC_PROTO((word));
# ifdef PCR_OBSOLETE /* Faster, but broken with multiple lwp's */
# include "th/PCR_Th.h"
# include "th/PCR_ThCrSec.h"
# define LOCK() RT0u__inCritical++
# define UNLOCK() RT0u__inCritical--
# endif
-# ifdef SOLARIS_THREADS
+# ifdef GC_SOLARIS_THREADS
# include <thread.h>
# include <signal.h>
extern mutex_t GC_allocate_ml;
# define GC_TEST_AND_SET_DEFINED
# endif
# if defined(IA64)
+# include <ia64intrin.h>
inline static int GC_test_and_set(volatile unsigned int *addr) {
- long oldval, n = 1;
- __asm__ __volatile__("xchg4 %0=%1,%2"
- : "=r"(oldval), "=m"(*addr)
- : "r"(n), "1"(*addr) : "memory");
- return oldval;
+ return __sync_lock_test_and_set(addr, 1);
}
# define GC_TEST_AND_SET_DEFINED
- /* Should this handle post-increment addressing?? */
inline static void GC_clear(volatile unsigned int *addr) {
- __asm__ __volatile__("st4.rel %0=r0" : "=m" (*addr) : : "memory");
+ *addr = 0;
}
# define GC_CLEAR_DEFINED
# endif
# if defined(POWERPC)
inline static int GC_test_and_set(volatile unsigned int *addr) {
int oldval;
- int temp = 1; // locked value
+ int temp = 1; /* locked value */
__asm__ __volatile__(
- "1:\tlwarx %0,0,%3\n" // load and reserve
- "\tcmpwi %0, 0\n" // if load is
- "\tbne 2f\n" // non-zero, return already set
- "\tstwcx. %2,0,%1\n" // else store conditional
- "\tbne- 1b\n" // retry if lost reservation
- "2:\t\n" // oldval is zero if we set
+ "1:\tlwarx %0,0,%3\n" /* load and reserve */
+ "\tcmpwi %0, 0\n" /* if load is */
+ "\tbne 2f\n" /* non-zero, return already set */
+ "\tstwcx. %2,0,%1\n" /* else store conditional */
+ "\tbne- 1b\n" /* retry if lost reservation */
+ "\tsync\n" /* import barrier */
+ "2:\t\n" /* oldval is zero if we set */
: "=&r"(oldval), "=p"(addr)
: "r"(temp), "1"(addr)
- : "memory");
- return (int)oldval;
+ : "cr0","memory");
+ return oldval;
}
# define GC_TEST_AND_SET_DEFINED
inline static void GC_clear(volatile unsigned int *addr) {
- __asm__ __volatile__("eieio" ::: "memory");
+ __asm__ __volatile__("eieio" : : : "memory");
*(addr) = 0;
}
# define GC_CLEAR_DEFINED
" bne %2,2f\n"
" xor %0,%3,%0\n"
" stl_c %0,%1\n"
+# ifdef __ELF__
" beq %0,3f\n"
+# else
+ " beq %0,1b\n"
+# endif
" mb\n"
"2:\n"
+# ifdef __ELF__
".section .text2,\"ax\"\n"
"3: br 1b\n"
".previous"
+# endif
:"=&r" (temp), "=m" (*addr), "=&r" (oldvalue)
:"Ir" (1), "m" (*addr)
:"memory");
return oldvalue;
}
# define GC_TEST_AND_SET_DEFINED
- /* Should probably also define GC_clear, since it needs */
- /* a memory barrier ?? */
+ inline static void GC_clear(volatile unsigned int *addr) {
+ __asm__ __volatile__("mb" : : : "memory");
+ *(addr) = 0;
+ }
+# define GC_CLEAR_DEFINED
# endif /* ALPHA */
# ifdef ARM32
inline static int GC_test_and_set(volatile unsigned int *addr) {
}
# define GC_TEST_AND_SET_DEFINED
# endif /* ARM32 */
+# ifdef CRIS
+ inline static int GC_test_and_set(volatile unsigned int *addr) {
+ /* Ripped from linuxthreads/sysdeps/cris/pt-machine.h. */
+ /* Included with Hans-Peter Nilsson's permission. */
+ register unsigned long int ret;
+
+ /* Note the use of a dummy output of *addr to expose the write.
+ * The memory barrier is to stop *other* writes being moved past
+ * this code.
+ */
+ __asm__ __volatile__("clearf\n"
+ "0:\n\t"
+ "movu.b [%2],%0\n\t"
+ "ax\n\t"
+ "move.b %3,[%2]\n\t"
+ "bwf 0b\n\t"
+ "clearf"
+ : "=&r" (ret), "=m" (*addr)
+ : "r" (addr), "r" ((int) 1), "m" (*addr)
+ : "memory");
+ return ret;
+ }
+# define GC_TEST_AND_SET_DEFINED
+# endif /* CRIS */
+# ifdef S390
+ inline static int GC_test_and_set(volatile unsigned int *addr) {
+ int ret;
+ __asm__ __volatile__ (
+ " l %0,0(%2)\n"
+ "0: cs %0,%1,0(%2)\n"
+ " jl 0b"
+ : "=&d" (ret)
+ : "d" (1), "a" (addr)
+ : "cc", "memory");
+ return ret;
+ }
+# endif
# endif /* __GNUC__ */
# if (defined(ALPHA) && !defined(__GNUC__))
-# define GC_test_and_set(addr) __cxx_test_and_set_atomic(addr, 1)
+# ifndef OSF1
+ --> We currently assume that if gcc is not used, we are
+ --> running under Tru64.
+# endif
+# include <machine/builtins.h>
+# include <c_asm.h>
+# define GC_test_and_set(addr) __ATOMIC_EXCH_LONG(addr, 1)
# define GC_TEST_AND_SET_DEFINED
+# define GC_clear(addr) { asm("mb"); *(volatile unsigned *)addr = 0; }
+# define GC_CLEAR_DEFINED
# endif
# if defined(MSWIN32)
# define GC_test_and_set(addr) InterlockedExchange((LPLONG)addr,1)
# define GC_TEST_AND_SET_DEFINED
# endif
# ifdef MIPS
-# if __mips < 3 || !(defined (_ABIN32) || defined(_ABI64)) \
+# ifdef LINUX
+# include <sys/tas.h>
+# define GC_test_and_set(addr) _test_and_set((int *) addr,1)
+# define GC_TEST_AND_SET_DEFINED
+# elif __mips < 3 || !(defined (_ABIN32) || defined(_ABI64)) \
|| !defined(_COMPILER_VERSION) || _COMPILER_VERSION < 700
-# define GC_test_and_set(addr, v) test_and_set(addr,v)
+# ifdef __GNUC__
+# define GC_test_and_set(addr) _test_and_set((void *)addr,1)
+# else
+# define GC_test_and_set(addr) test_and_set((void *)addr,1)
+# endif
# else
-# define GC_test_and_set(addr, v) __test_and_set(addr,v)
+# define GC_test_and_set(addr) __test_and_set32((void *)addr,1)
# define GC_clear(addr) __lock_release(addr);
# define GC_CLEAR_DEFINED
# endif
# define GC_TEST_AND_SET_DEFINED
# endif /* MIPS */
+# if defined(_AIX)
+# include <sys/atomic_op.h>
+# if (defined(_POWER) || defined(_POWERPC))
+# if defined(__GNUC__)
+ inline static void GC_memsync() {
+ __asm__ __volatile__ ("sync" : : : "memory");
+ }
+# else
+# ifndef inline
+# define inline __inline
+# endif
+# pragma mc_func GC_memsync { \
+ "7c0004ac" /* sync (same opcode used for dcs)*/ \
+ }
+# endif
+# else
+# error dont know how to memsync
+# endif
+ inline static int GC_test_and_set(volatile unsigned int * addr) {
+ int oldvalue = 0;
+ if (compare_and_swap((void *)addr, &oldvalue, 1)) {
+ GC_memsync();
+ return 0;
+ } else return 1;
+ }
+# define GC_TEST_AND_SET_DEFINED
+ inline static void GC_clear(volatile unsigned int *addr) {
+ GC_memsync();
+ *(addr) = 0;
+ }
+# define GC_CLEAR_DEFINED
+
+# endif
# if 0 /* defined(HP_PA) */
/* The official recommendation seems to be to not use ldcw from */
/* user mode. Since multithreaded incremental collection doesn't */
/* "set" means 0 and "clear" means 1 here. */
# define GC_test_and_set(addr) !GC_test_and_clear(addr);
# define GC_TEST_AND_SET_DEFINED
-# define GC_clear(addr) GC_noop1(addr); *(volatile unsigned int *)addr = 1;
+# define GC_clear(addr) GC_noop1((word)(addr)); *(volatile unsigned int *)addr = 1;
/* The above needs a memory barrier! */
# define GC_CLEAR_DEFINED
# endif
# define USE_PTHREAD_LOCKS
# endif
-# if defined(LINUX_THREADS) || defined(OSF1_THREADS) \
- || defined(HPUX_THREADS)
+# if defined(GC_PTHREADS) && !defined(GC_SOLARIS_THREADS) \
+ && !defined(GC_IRIX_THREADS) && !defined(GC_WIN32_THREADS)
# define NO_THREAD (pthread_t)(-1)
# include <pthread.h>
# if defined(PARALLEL_MARK)
{
char result;
__asm__ __volatile__("lock; cmpxchgl %2, %0; setz %1"
- : "=m"(*(addr)), "=r"(result)
- : "r" (new_val), "0"(*(addr)), "a"(old) : "memory");
+ : "+m"(*(addr)), "=r"(result)
+ : "r" (new_val), "a"(old) : "memory");
return (GC_bool) result;
}
# endif /* !GENERIC_COMPARE_AND_SWAP */
- inline static void GC_memory_write_barrier()
+ inline static void GC_memory_barrier()
{
/* We believe the processor ensures at least processor */
/* consistent ordering. Thus a compiler barrier */
__asm__ __volatile__("" : : : "memory");
}
# endif /* I386 */
+
+# if defined(POWERPC)
+# if !defined(GENERIC_COMPARE_AND_SWAP)
+ /* Returns TRUE if the comparison succeeded. */
+ inline static GC_bool GC_compare_and_exchange(volatile GC_word *addr,
+ GC_word old, GC_word new_val)
+ {
+ int result, dummy;
+ __asm__ __volatile__(
+ "1:\tlwarx %0,0,%5\n"
+ "\tcmpw %0,%4\n"
+ "\tbne 2f\n"
+ "\tstwcx. %3,0,%2\n"
+ "\tbne- 1b\n"
+ "\tsync\n"
+ "\tli %1, 1\n"
+ "\tb 3f\n"
+ "2:\tli %1, 0\n"
+ "3:\t\n"
+ : "=&r" (dummy), "=r" (result), "=p" (addr)
+ : "r" (new_val), "r" (old), "2"(addr)
+ : "cr0","memory");
+ return (GC_bool) result;
+ }
+# endif /* !GENERIC_COMPARE_AND_SWAP */
+ inline static void GC_memory_barrier()
+ {
+ __asm__ __volatile__("sync" : : : "memory");
+ }
+# endif /* POWERPC */
+
# if defined(IA64)
# if !defined(GENERIC_COMPARE_AND_SWAP)
inline static GC_bool GC_compare_and_exchange(volatile GC_word *addr,
- GC_word old, GC_word new_val)
+ GC_word old,
+ GC_word new_val)
{
- unsigned long oldval;
- __asm__ __volatile__("mov ar.ccv=%4 ;; cmpxchg8.rel %0=%1,%2,ar.ccv"
- : "=r"(oldval), "=m"(*addr)
- : "r"(new_val), "1"(*addr), "r"(old) : "memory");
- return (oldval == old);
+ return __sync_bool_compare_and_swap (addr, old, new_val);
}
# endif /* !GENERIC_COMPARE_AND_SWAP */
# if 0
/* Shouldn't be needed; we use volatile stores instead. */
- inline static void GC_memory_write_barrier()
+ inline static void GC_memory_barrier()
{
- __asm__ __volatile__("mf" : : : "memory");
+ __sync_synchronize ();
}
# endif /* 0 */
# endif /* IA64 */
+# if defined(ALPHA)
+# if !defined(GENERIC_COMPARE_AND_SWAP)
+# if defined(__GNUC__)
+ inline static GC_bool GC_compare_and_exchange(volatile GC_word *addr,
+ GC_word old, GC_word new_val)
+ {
+ unsigned long was_equal;
+ unsigned long temp;
+
+ __asm__ __volatile__(
+ "1: ldq_l %0,%1\n"
+ " cmpeq %0,%4,%2\n"
+ " mov %3,%0\n"
+ " beq %2,2f\n"
+ " stq_c %0,%1\n"
+ " beq %0,1b\n"
+ "2:\n"
+ " mb\n"
+ :"=&r" (temp), "=m" (*addr), "=&r" (was_equal)
+ : "r" (new_val), "Ir" (old)
+ :"memory");
+ return was_equal;
+ }
+# else /* !__GNUC__ */
+ inline static GC_bool GC_compare_and_exchange(volatile GC_word *addr,
+ GC_word old, GC_word new_val)
+ {
+ return __CMP_STORE_QUAD(addr, old, new_val, addr);
+ }
+# endif /* !__GNUC__ */
+# endif /* !GENERIC_COMPARE_AND_SWAP */
+# ifdef __GNUC__
+ inline static void GC_memory_barrier()
+ {
+ __asm__ __volatile__("mb" : : : "memory");
+ }
+# else
+# define GC_memory_barrier() asm("mb")
+# endif /* !__GNUC__ */
+# endif /* ALPHA */
+# if defined(S390)
+# if !defined(GENERIC_COMPARE_AND_SWAP)
+ inline static GC_bool GC_compare_and_exchange(volatile C_word *addr,
+ GC_word old, GC_word new_val)
+ {
+ int retval;
+ __asm__ __volatile__ (
+# ifndef __s390x__
+ " cs %1,%2,0(%3)\n"
+# else
+ " csg %1,%2,0(%3)\n"
+# endif
+ " ipm %0\n"
+ " srl %0,28\n"
+ : "=&d" (retval), "+d" (old)
+ : "d" (new_val), "a" (addr)
+ : "cc", "memory");
+ return retval == 0;
+ }
+# endif
+# endif
# if !defined(GENERIC_COMPARE_AND_SWAP)
/* Returns the original value of *addr. */
inline static GC_word GC_atomic_add(volatile GC_word *addr,
{ GC_ASSERT(I_HOLD_LOCK()); UNSET_LOCK_HOLDER(); \
pthread_mutex_unlock(&GC_allocate_ml); }
# else /* !GC_ASSERTIONS */
+# if defined(NO_PTHREAD_TRYLOCK)
+# define LOCK() GC_lock();
+# else /* !defined(NO_PTHREAD_TRYLOCK) */
# define LOCK() \
{ if (0 != pthread_mutex_trylock(&GC_allocate_ml)) GC_lock(); }
+# endif
# define UNLOCK() pthread_mutex_unlock(&GC_allocate_ml)
# endif /* !GC_ASSERTIONS */
# endif /* USE_PTHREAD_LOCKS */
# ifdef GC_ASSERTIONS
extern pthread_t GC_mark_lock_holder;
# endif
-# endif /* LINUX_THREADS || OSF1_THREADS || HPUX_THREADS */
-# if defined(IRIX_THREADS)
+# endif /* GC_PTHREADS with linux_threads.c implementation */
+# if defined(GC_IRIX_THREADS)
# include <pthread.h>
/* This probably should never be included, but I can't test */
/* on Irix anymore. */
# include <mutex.h>
- extern unsigned long GC_allocate_lock;
+ extern volatile unsigned int GC_allocate_lock;
/* This is not a mutex because mutexes that obey the (optional) */
/* POSIX scheduling rules are subject to convoys in high contention */
/* applications. This is basically a spin lock. */
# define NO_THREAD (pthread_t)(-1)
# define UNSET_LOCK_HOLDER() GC_lock_holder = NO_THREAD
# define I_HOLD_LOCK() (pthread_equal(GC_lock_holder, pthread_self()))
-# define LOCK() { if (GC_test_and_set(&GC_allocate_lock, 1)) GC_lock(); }
+# define LOCK() { if (GC_test_and_set(&GC_allocate_lock)) GC_lock(); }
# define UNLOCK() GC_clear(&GC_allocate_lock);
extern VOLATILE GC_bool GC_collecting;
# define ENTER_GC() \
GC_collecting = 1; \
}
# define EXIT_GC() GC_collecting = 0;
-# endif /* IRIX_THREADS */
-# ifdef WIN32_THREADS
-# include <windows.h>
- GC_API CRITICAL_SECTION GC_allocate_ml;
-# define LOCK() EnterCriticalSection(&GC_allocate_ml);
-# define UNLOCK() LeaveCriticalSection(&GC_allocate_ml);
+# endif /* GC_IRIX_THREADS */
+# if defined(GC_WIN32_THREADS)
+# if defined(GC_PTHREADS)
+# include <pthread.h>
+ extern pthread_mutex_t GC_allocate_ml;
+# define LOCK() pthread_mutex_lock(&GC_allocate_ml)
+# define UNLOCK() pthread_mutex_unlock(&GC_allocate_ml)
+# else
+# include <windows.h>
+ GC_API CRITICAL_SECTION GC_allocate_ml;
+# define LOCK() EnterCriticalSection(&GC_allocate_ml);
+# define UNLOCK() LeaveCriticalSection(&GC_allocate_ml);
+# endif
# endif
# ifndef SET_LOCK_HOLDER
# define SET_LOCK_HOLDER()