src/atomic.cc

/*
 * Copyright (C) 2010 The Android Open Source Project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include "atomic.h"

#include <sched.h>

namespace art {

/*
 * Quasi-atomic 64-bit operations, for platforms that lack the real thing.
 *
 * TODO: unify ARMv6/x86/sh implementations using the to-be-written
 * spin lock implementation.  We don't want to rely on mutex innards,
 * and it would be great if all platforms were running the same code.
 */

#if defined(HAVE_MACOSX_IPC)

#include <libkern/OSAtomic.h>

#if defined(__ppc__)        \
    || defined(__PPC__)     \
    || defined(__powerpc__) \
    || defined(__powerpc)   \
    || defined(__POWERPC__) \
    || defined(_M_PPC)      \
    || defined(__PPC)
#define NEED_QUASIATOMICS 1
#else

int QuasiAtomicCas64(int64_t old_value, int64_t new_value, volatile int64_t* addr) {
  return OSAtomicCompareAndSwap64Barrier(old_value, new_value, (int64_t*)addr) == 0;
}

static inline int64_t QuasiAtomicSwap64Impl(int64_t value, volatile int64_t* addr) {
  int64_t old_value;
  do {
    old_value = *addr;
  } while (QuasiAtomicCas64(old_value, value, addr));
  return old_value;
}

int64_t QuasiAtomicSwap64(int64_t value, volatile int64_t* addr) {
  return QuasiAtomicSwap64Impl(value, addr);
}

int64_t QuasiAtomicSwap64Sync(int64_t value, volatile int64_t* addr) {
  ANDROID_MEMBAR_STORE();
  int64_t old_value = QuasiAtomicSwap64Impl(value, addr);
  /* TUNING: barriers can be avoided on some architectures */
  ANDROID_MEMBAR_FULL();
  return old_value;
}

int64_t QuasiAtomicRead64(volatile const int64_t* addr) {
  return OSAtomicAdd64Barrier(0, addr);
}
#endif

#elif defined(__i386__) || defined(__x86_64__)
#define NEED_QUASIATOMICS 1

#elif __arm__
#include <machine/cpu-features.h>

#ifdef __ARM_HAVE_LDREXD
static inline int64_t QuasiAtomicSwap64Impl(int64_t new_value, volatile int64_t* addr) {
  int64_t prev;
  int status;
  do {
    __asm__ __volatile__("@ QuasiAtomicSwap64\n"
        "ldrexd     %0, %H0, [%3]\n"
        "strexd     %1, %4, %H4, [%3]"
        : "=&r" (prev), "=&r" (status), "+m"(*addr)
        : "r" (addr), "r" (new_value)
        : "cc");
  } while (__builtin_expect(status != 0, 0));
  return prev;
}

int64_t QuasiAtomicSwap64(int64_t new_value, volatile int64_t* addr) {
  return QuasiAtomicSwap64Impl(new_value, addr);
}

int64_t QuasiAtomicSwap64Sync(int64_t new_value, volatile int64_t* addr) {
  ANDROID_MEMBAR_STORE();
  int64_t old_value = QuasiAtomicSwap64Impl(new_value, addr);
  ANDROID_MEMBAR_FULL();
  return old_value;
}

int QuasiAtomicCas64(int64_t old_value, int64_t new_value, volatile int64_t* addr) {
  int64_t prev;
  int status;
  do {
    __asm__ __volatile__("@ QuasiAtomicCas64\n"
        "ldrexd     %0, %H0, [%3]\n"
        "mov        %1, #0\n"
        "teq        %0, %4\n"
        "teqeq      %H0, %H4\n"
        "strexdeq   %1, %5, %H5, [%3]"
        : "=&r" (prev), "=&r" (status), "+m"(*addr)
        : "r" (addr), "Ir" (old_value), "r" (new_value)
        : "cc");
  } while (__builtin_expect(status != 0, 0));
  return prev != old_value;
}

int64_t QuasiAtomicRead64(volatile const int64_t* addr) {
  int64_t value;
  __asm__ __volatile__("@ QuasiAtomicRead64\n"
      "ldrexd     %0, %H0, [%1]"
      : "=&r" (value)
      : "r" (addr));
  return value;
}

#else

// on the device, we implement the 64-bit atomic operations through
// mutex locking. normally, this is bad because we must initialize
// a pthread_mutex_t before being able to use it, and this means
// having to do an initialization check on each function call, and
// that's where really ugly things begin...
//
// BUT, as a special twist, we take advantage of the fact that in our
// pthread library, a mutex is simply a volatile word whose value is always
// initialized to 0. In other words, simply declaring a static mutex
// object initializes it !
//
// another twist is that we use a small array of mutexes to dispatch
// the contention locks from different memory addresses
//

#include <pthread.h>

#define  SWAP_LOCK_COUNT  32U
static pthread_mutex_t  _swap_locks[SWAP_LOCK_COUNT];

#define  SWAP_LOCK(addr) &_swap_locks[((unsigned)(void*)(addr) >> 3U) % SWAP_LOCK_COUNT]

int64_t QuasiAtomicSwap64(int64_t value, volatile int64_t* addr) {
  pthread_mutex_t*  lock = SWAP_LOCK(addr);

  pthread_mutex_lock(lock);

  int64_t old_value = *addr;
  *addr = value;

  pthread_mutex_unlock(lock);
  return old_value;
}

int64_t QuasiAtomicSwap64Sync(int64_t value, volatile int64_t* addr) {
  // Same as QuasiAtomicSwap64 - mutex handles barrier.
  return QuasiAtomicSwap64(value, addr);
}

int QuasiAtomicCas64(int64_t old_value, int64_t new_value, volatile int64_t* addr) {
  int result;
  pthread_mutex_t*  lock = SWAP_LOCK(addr);

  pthread_mutex_lock(lock);

  if (*addr == old_value) {
    *addr  = new_value;
    result = 0;
  } else {
    result = 1;
  }
  pthread_mutex_unlock(lock);
  return result;
}

int64_t QuasiAtomicRead64(volatile const int64_t* addr) {
  int64_t result;
  pthread_mutex_t*  lock = SWAP_LOCK(addr);

  pthread_mutex_lock(lock);
  result = *addr;
  pthread_mutex_unlock(lock);
  return result;
}

#endif /*__ARM_HAVE_LDREXD*/

/*****************************************************************************/
#elif __sh__
#define NEED_QUASIATOMICS 1

#else
#error "Unsupported atomic operations for this platform"
#endif


#if NEED_QUASIATOMICS

/* Note that a spinlock is *not* a good idea in general
 * since they can introduce subtle issues. For example,
 * a real-time thread trying to acquire a spinlock already
 * acquired by another thread will never yeld, making the
 * CPU loop endlessly!
 *
 * However, this code is only used on the Linux simulator
 * so it's probably ok for us.
 *
 * The alternative is to use a pthread mutex, but
 * these must be initialized before being used, and
 * then you have the problem of lazily initializing
 * a mutex without any other synchronization primitive.
 *
 * TODO: these currently use sched_yield(), which is not guaranteed to
 * do anything at all.  We need to use dvmIterativeSleep or a wait /
 * notify mechanism if the initial attempt fails.
 */

/* global spinlock for all 64-bit quasiatomic operations */
static int32_t quasiatomic_spinlock = 0;

int QuasiAtomicCas64(int64_t old_value, int64_t new_value, volatile int64_t* addr) {
  int result;

  while (android_atomic_acquire_cas(0, 1, &quasiatomic_spinlock)) {
#ifdef HAVE_WIN32_THREADS
    Sleep(0);
#else
    sched_yield();
#endif
  }

  if (*addr == old_value) {
    *addr = new_value;
    result = 0;
  } else {
    result = 1;
  }

  android_atomic_release_store(0, &quasiatomic_spinlock);

  return result;
}

int64_t QuasiAtomicRead64(volatile const int64_t* addr) {
  int64_t result;

  while (android_atomic_acquire_cas(0, 1, &quasiatomic_spinlock)) {
#ifdef HAVE_WIN32_THREADS
    Sleep(0);
#else
    sched_yield();
#endif
  }

  result = *addr;
  android_atomic_release_store(0, &quasiatomic_spinlock);

  return result;
}

int64_t QuasiAtomicSwap64(int64_t value, volatile int64_t* addr) {
  int64_t result;

  while (android_atomic_acquire_cas(0, 1, &quasiatomic_spinlock)) {
#ifdef HAVE_WIN32_THREADS
    Sleep(0);
#else
    sched_yield();
#endif
  }

  result = *addr;
  *addr = value;
  android_atomic_release_store(0, &quasiatomic_spinlock);

  return result;
}

int64_t QuasiAtomicSwap64Sync(int64_t value, volatile int64_t* addr) {
  // Same as QuasiAtomicSwap64 - syscall handles barrier.
  return QuasiAtomicSwap64(value, addr);
}

#endif /*NEED_QUASIATOMICS*/

}  // namespace art