/*
 * Copyright (c) 1999, 2017, Oracle and/or its affiliates. All rights reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
 *
 */

#ifndef SHARE_VM_RUNTIME_ATOMIC_HPP
#define SHARE_VM_RUNTIME_ATOMIC_HPP

#include "memory/allocation.hpp"
#include "metaprogramming/integerTypes.hpp"
#include "metaprogramming/isIntegral.hpp"
#include "metaprogramming/isPointer.hpp"
#include "utilities/align.hpp"
#include "utilities/debug.hpp"
#include "utilities/macros.hpp"

enum cmpxchg_memory_order {
  memory_order_relaxed,
  // Use value which doesn't interfere with C++2011. We need to be more conservative.
  memory_order_conservative = 8
};

class Atomic : AllStatic {
  template<typename T> class Never: public FalseType {};

  template <typename T>
  inline static void specialized_store(T store_value, volatile T* dest) {
    STATIC_ASSERT(sizeof(T) <= size_t(BytesPerWord)); // Does the machine support atomic wide accesses?
    (void)const_cast<T&>(*dest = store_value);
  }

  template <typename T>
  inline static T specialized_load(const volatile T* dest) {
    STATIC_ASSERT(sizeof(T) <= size_t(BytesPerWord)); // Does the machine support atomic wide accesses?
    return *dest;
  }

  template <typename T>
  inline static T specialized_add(T add_value, volatile T* dest) {
    STATIC_ASSERT(Never<T>::value);
    return add_value;
  }

  template <typename T>
  inline static void specialized_inc(volatile T* dest) {
    add(1, dest);
  }

  template <typename T>
  inline static void specialized_dec(volatile T* dest) {
    add(-1, dest);
  }

  template <typename T>
  inline static T specialized_xchg(T exchange_value, volatile T* dest) {
    STATIC_ASSERT(Never<T>::value);
    return exchange_value;
  }

  template <typename T>
  inline static T specialized_cmpxchg(T exchange_value, volatile T* dest, T compare_value, cmpxchg_memory_order order) {
    STATIC_ASSERT(Never<T>::value);
    return exchange_value;
  }

 public:
  // Atomic operations on 64-bit types are not available on all 32-bit
  // platforms. If atomic ops on 64-bit types are defined here they must only
  // be used from code that verifies they are available at runtime and
  // can provide an alternative action if not - see supports_cx8() for
  // a means to test availability.

  // The memory operations that are mentioned with each of the atomic
  // function families come from src/share/vm/runtime/orderAccess.hpp,
  // e.g., <fence> is described in that file and is implemented by the
  // OrderAccess::fence() function. See that file for the gory details
  // on the Memory Access Ordering Model.

  // All of the atomic operations that imply a read-modify-write action
  // guarantee a two-way memory barrier across that operation. Historically
  // these semantics reflect the strength of atomic operations that are
  // provided on SPARC/X86. We assume that strength is necessary unless
  // we can prove that a weaker form is sufficiently safe.

  // Atomically store to a location
  // See comment above about using 64-bit atomics on 32-bit platforms
  template <typename T, typename U>
  inline static void store(T store_value, volatile U* dest);

  // The store_ptr() member functions are deprecated. Use store() instead.
  static void store_ptr(intptr_t store_value, volatile intptr_t* dest) {
    store(store_value, dest);
  }

  static void store_ptr(void*    store_value, volatile void*     dest) {
    store((intptr_t)store_value, (volatile intptr_t*)dest);
  }

  // Atomically load from a location
  // See comment above about using 64-bit atomics on 32-bit platforms
  template <typename T>
  inline static T load(volatile T* src);

  // Atomically add to a location. Returns updated value. add*() provide:
  // <fence> add-value-to-dest <membar StoreLoad|StoreStore>
  // add(I1 v, I* d)
  // add(I1 v, P* d)
  // where I, I1 are integral types, P is a pointer type.
  // Functional behavior is modelled on *dest += add_value.
  template <typename T, typename U>
  inline static U add(T add_value, volatile U* dst);

  template <typename T, typename U>
  inline static U* add(T add_value, U* volatile* dst);

  // The add_ptr() member functions are deprecated. Use add() instead.
  static intptr_t add_ptr(intptr_t add_value, volatile intptr_t* dest) {
    return add(add_value, dest);
  }

  static void*    add_ptr(intptr_t add_value, volatile void*     dest) {
    return (void*)add(add_value, (volatile intptr_t*)dest);
  }

  // Atomically increment location. inc*() provide:
  // <fence> increment-dest <membar StoreLoad|StoreStore>
  // Functional behavior is modelled on *dest++
  template <typename T>
  inline static void inc(volatile T* dest);

  template <typename T>
  inline static void inc(T* volatile* dest);

  // The inc_ptr member functions are deprecated. Use inc() instead.
  static void inc_ptr(volatile intptr_t* dest) {
    inc(dest);
  }

  static void inc_ptr(volatile void*     dest) {
    inc((volatile intptr_t*)dest);
  }

  // Atomically decrement a location. dec*() provide:
  // <fence> decrement-dest <membar StoreLoad|StoreStore>
  // Functional behavior is modelled on *dest--
  template <typename T>
  inline static void dec(volatile T* dest);

  template <typename T>
  inline static void dec(T* volatile* dest);

  // The dec_ptr member functions are deprecated. Use dec() instead.
  static void dec_ptr(volatile intptr_t* dest) {
    dec(dest);
  }

  static void dec_ptr(volatile void*     dest) {
    dec((volatile intptr_t*)dest);
  }

  // Performs atomic exchange of *dest with exchange_value. Returns old
  // prior value of *dest. xchg*() provide:
  // <fence> exchange-value-with-dest <membar StoreLoad|StoreStore>
  template <typename T, typename U>
  inline static U xchg(T exchange_value, volatile U* dest);

  // The xchg_ptr() member functions are deprecated. Use xchg() instead.
  static intptr_t xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest) {
    return xchg(exchange_value, dest);
  }

  static void*    xchg_ptr(void*    exchange_value, volatile void*     dest) {
    return (void*)xchg((intptr_t)exchange_value, (volatile intptr_t*)dest);
  }

  // Performs atomic compare of *dest and compare_value, and exchanges
  // *dest with exchange_value if the comparison succeeded. Returns prior
  // value of *dest. cmpxchg*() provide:
  // <fence> compare-and-exchange <membar StoreLoad|StoreStore>
  // See comment above about using 64-bit atomics on 32-bit platforms
  template <typename T, typename U, typename V>
  inline static U cmpxchg(T exchange_value, volatile U* dest, V compare_value, cmpxchg_memory_order order = memory_order_conservative);

  // The cmpxchg_ptr member functions are deprecated. Use cmpxchg() instead.
  inline static intptr_t cmpxchg_ptr(intptr_t exchange_value, volatile intptr_t*  dest,
                                     intptr_t compare_value, cmpxchg_memory_order order = memory_order_conservative) {
    return cmpxchg(exchange_value, dest, compare_value, order);

  }

  inline static void*    cmpxchg_ptr(void*    exchange_value, volatile void*      dest,
                                     void*    compare_value, cmpxchg_memory_order order = memory_order_conservative) {
    return (void*)cmpxchg((intptr_t)exchange_value, (volatile intptr_t*)dest, (intptr_t)compare_value, order);
  }
};

// internal implementation

template <typename T, typename U>
inline void Atomic::store(T store_value, volatile U* dest) {
  typedef typename IntegerTypes::Signed<U>::type Raw;
  U store_value_cast = store_value;
  specialized_store(IntegerTypes::cast_to_signed(store_value_cast), reinterpret_cast<volatile Raw*>(dest));
}

template <typename T>
inline T Atomic::load(volatile T* src) {
  typedef typename IntegerTypes::Signed<T>::type Raw;
  return IntegerTypes::cast<T>(specialized_load(reinterpret_cast<const volatile Raw*>(src)));
}

template <typename T, typename U>
inline U Atomic::add(T add_value, volatile U* dst) {
  STATIC_ASSERT(IsIntegral<T>::value);
  STATIC_ASSERT(IsIntegral<U>::value);
  typedef typename IntegerTypes::Signed<U>::type Raw;
  // Allow -Wconversion or the like to complain about unsafe conversions.
  U value = add_value;
  Raw raw_value = IntegerTypes::cast_to_signed(value);
  Raw result = specialized_add(raw_value, reinterpret_cast<volatile Raw*>(dst));
  return IntegerTypes::cast<U>(result);
}

template <typename T, typename U>
inline U* Atomic::add(T add_value, U* volatile* dst) {
  STATIC_ASSERT(IsIntegral<T>::value);
  typedef typename IntegerTypes::Signed<U*>::type Raw;
  ptrdiff_t value = add_value;
  Raw raw_value = IntegerTypes::cast_to_signed(value * sizeof(U));
  Raw result = specialized_add(raw_value, reinterpret_cast<volatile Raw*>(dst));
  return IntegerTypes::cast<U*>(result);
}

template <typename T>
inline void Atomic::inc(volatile T* src) {
  STATIC_ASSERT(IsIntegral<T>::value);
  typedef typename IntegerTypes::Signed<T>::type Raw;
  specialized_inc(reinterpret_cast<volatile Raw*>(src));
}

template <typename T>
inline void Atomic::inc(T* volatile* src) {
  if (sizeof(T) != 1) {
    add(1, src);
  } else {
    typedef typename IntegerTypes::Signed<T*>::type Raw;
    specialized_inc(reinterpret_cast<volatile Raw*>(src));
  }
}

template <typename T>
inline void Atomic::dec(volatile T* src) {
  STATIC_ASSERT(IsIntegral<T>::value);
  typedef typename IntegerTypes::Signed<T>::type Raw;
  specialized_dec(reinterpret_cast<volatile Raw*>(src));
}

template <typename T>
inline void Atomic::dec(T* volatile* src) {
  if (sizeof(T) != 1) {
    add(-1, src);
  } else {
    typedef typename IntegerTypes::Signed<T*>::type Raw;
    specialized_dec(reinterpret_cast<volatile Raw*>(src));
  }
}

template <typename T, typename U>
inline U Atomic::xchg(T exchange_value, volatile U* dest) {
  typedef typename IntegerTypes::Signed<U>::type Raw;
  U exchange_value_cast = exchange_value;
  Raw result = specialized_xchg(IntegerTypes::cast_to_signed(exchange_value_cast),
                                reinterpret_cast<volatile Raw*>(dest));
  return IntegerTypes::cast<U>(result);
}

template <typename T, typename U, typename V>
inline U Atomic::cmpxchg(T exchange_value, volatile U* dest, V compare_value, cmpxchg_memory_order order) {
  typedef typename IntegerTypes::Signed<U>::type Raw;
  U exchange_value_cast = exchange_value;
  U compare_value_cast = compare_value;
  Raw result = specialized_cmpxchg(IntegerTypes::cast_to_signed(exchange_value_cast),
                                   reinterpret_cast<volatile Raw*>(dest),
                                   IntegerTypes::cast_to_signed(compare_value_cast), order);
  return IntegerTypes::cast<U>(result);
}

// platform specific in-line definitions - must come before shared definitions

#include OS_CPU_HEADER(atomic)

// shared in-line definitions

#ifndef VM_HAS_SPECIALIZED_CMPXCHG_BYTE
/*
 * This is the default implementation of byte-sized cmpxchg. It emulates 8-bit-sized cmpxchg
 * in terms of 32-bit-sized cmpxchg. Platforms may override this by defining their own inline definition
 * as well as defining VM_HAS_SPECIALIZED_CMPXCHG_BYTE. This will cause the platform specific
 * implementation to be used instead.
 */
template <>
inline int8_t Atomic::specialized_cmpxchg<int8_t>(int8_t exchange_value, volatile int8_t* dest,
                                                  int8_t compare_value, cmpxchg_memory_order order) {
  volatile int32_t* dest_int =
      reinterpret_cast<volatile int32_t*>(align_down(dest, sizeof(int32_t)));
  size_t offset = pointer_delta(dest, dest_int, 1);
  int32_t cur = *dest_int;
  int8_t* cur_as_bytes = reinterpret_cast<int8_t*>(&cur);

  // current value may not be what we are looking for, so force it
  // to that value so the initial cmpxchg will fail if it is different
  cur_as_bytes[offset] = compare_value;

  // always execute a real cmpxchg so that we get the required memory
  // barriers even on initial failure
  do {
    // value to swap in matches current value ...
    int32_t new_value = cur;
    // ... except for the one byte we want to update
    reinterpret_cast<int8_t*>(&new_value)[offset] = exchange_value;

    int32_t res = cmpxchg(new_value, dest_int, cur, order);
    if (res == cur) break; // success

    // at least one byte in the int changed value, so update
    // our view of the current int
    cur = res;
    // if our byte is still as cur we loop and try again
  } while (cur_as_bytes[offset] == compare_value);

  return cur_as_bytes[offset];
}

#endif // VM_HAS_SPECIALIZED_CMPXCHG_BYTE

template <>
inline int16_t Atomic::specialized_add<int16_t>(int16_t add_value, volatile int16_t* dest) {
  // Most platforms do not support atomic add on a 2-byte value. However,
  // if the value occupies the most significant 16 bits of an aligned 32-bit
  // word, then we can do this with an atomic add of (add_value << 16)
  // to the 32-bit word.
  //
  // The least significant parts of this 32-bit word will never be affected, even
  // in case of overflow/underflow.
  //
  // Use the ATOMIC_SHORT_PAIR macro (see macros.hpp) to get the desired alignment.
#ifdef VM_LITTLE_ENDIAN
  assert((intx(dest) & 0x03) == 0x02, "wrong alignment");
  int32_t new_value = Atomic::add(int32_t(add_value) << 16, (volatile int32_t*)(dest-1));
#else
  assert((intx(dest) & 0x03) == 0x00, "wrong alignment");
  int32_t new_value = Atomic::add(int32_t(add_value) << 16, (volatile int32_t*)(dest));
#endif
  return (int16_t)(new_value >> 16); // preserves sign
}

template <>
inline void Atomic::specialized_inc<int16_t>(volatile int16_t* dest) {
  (void)add(int16_t(1), dest);
}

template <>
inline void Atomic::specialized_dec<int16_t>(volatile int16_t* dest) {
  (void)add(int16_t(-1), dest);
}

#endif // SHARE_VM_RUNTIME_ATOMIC_HPP