glib/glib/gatomic.c
Peter Wu a3d90c0726 gatomic: Fix false positive with Clang+TSAN
__atomic_load_8 and friends do not exist under clang. Use the generic
__atomic_load variant instead that are documented here:
https://gcc.gnu.org/onlinedocs/gcc/_005f_005fatomic-Builtins.html

These have the additional benefit that the exact size of gint (4 bytes)
or gpointer (4 or 8 bytes) no longer have to be checked.

I initially tried `__typeof__(*(atomic)) val;`, but that caused warnings
in Clang (-Wincompatible-pointer-types-discards-qualifiers) when
"atomic" points to a volatile variable. Aside from that, it is
apparently not supported everywhere, see the g_has_typeof macro.
Another reason not to use it are new warnings under Clang, including:

    glib/deprecated/gthread-deprecated.c:683:11: warning: incompatible pointer types initializing 'typeof (*(&mutex->mutex.mutex))' (aka 'union _GMutex *') with an expression of type 'GRecMutex *' (aka 'struct _GRecMutex *') [-Wincompatible-pointer-types]
              g_atomic_pointer_set (&mutex->mutex.mutex, result);

Hence, cast the atomic variable to gint/gpointer pointers, the size was
already statically asserted so the cast should be safe.

The macros use a (hopefully) rare "gaps_temp" name instead of something
like "val" to avoid an issue with GCC builds:

    glib/tests/once.c:123:test_once4: assertion failed (val == "foo"): (NULL == "foo")

Closes #1843
2019-07-26 02:11:35 +01:00

920 lines
22 KiB
C

/*
* Copyright © 2011 Ryan Lortie
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library 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
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, see <http://www.gnu.org/licenses/>.
*
* Author: Ryan Lortie <desrt@desrt.ca>
*/
#include "config.h"
#include "gatomic.h"
/**
* SECTION:atomic_operations
* @title: Atomic Operations
* @short_description: basic atomic integer and pointer operations
* @see_also: #GMutex
*
* The following is a collection of compiler macros to provide atomic
* access to integer and pointer-sized values.
*
* The macros that have 'int' in the name will operate on pointers to
* #gint and #guint. The macros with 'pointer' in the name will operate
* on pointers to any pointer-sized value, including #gsize. There is
* no support for 64bit operations on platforms with 32bit pointers
* because it is not generally possible to perform these operations
* atomically.
*
* The get, set and exchange operations for integers and pointers
* nominally operate on #gint and #gpointer, respectively. Of the
* arithmetic operations, the 'add' operation operates on (and returns)
* signed integer values (#gint and #gssize) and the 'and', 'or', and
* 'xor' operations operate on (and return) unsigned integer values
* (#guint and #gsize).
*
* All of the operations act as a full compiler and (where appropriate)
* hardware memory barrier. Acquire and release or producer and
* consumer barrier semantics are not available through this API.
*
* It is very important that all accesses to a particular integer or
* pointer be performed using only this API and that different sizes of
* operation are not mixed or used on overlapping memory regions. Never
* read or assign directly from or to a value -- always use this API.
*
* For simple reference counting purposes you should use
* g_atomic_int_inc() and g_atomic_int_dec_and_test(). Other uses that
* fall outside of simple reference counting patterns are prone to
* subtle bugs and occasionally undefined behaviour. It is also worth
* noting that since all of these operations require global
* synchronisation of the entire machine, they can be quite slow. In
* the case of performing multiple atomic operations it can often be
* faster to simply acquire a mutex lock around the critical area,
* perform the operations normally and then release the lock.
**/
/**
* G_ATOMIC_LOCK_FREE:
*
* This macro is defined if the atomic operations of GLib are
* implemented using real hardware atomic operations. This means that
* the GLib atomic API can be used between processes and safely mixed
* with other (hardware) atomic APIs.
*
* If this macro is not defined, the atomic operations may be
* emulated using a mutex. In that case, the GLib atomic operations are
* only atomic relative to themselves and within a single process.
**/
/* NOTE CAREFULLY:
*
* This file is the lowest-level part of GLib.
*
* Other lowlevel parts of GLib (threads, slice allocator, g_malloc,
* messages, etc) call into these functions and macros to get work done.
*
* As such, these functions can not call back into any part of GLib
* without risking recursion.
*/
#ifdef G_ATOMIC_LOCK_FREE
/* if G_ATOMIC_LOCK_FREE was defined by `meson configure` then we MUST
* implement the atomic operations in a lock-free manner.
*/
#if defined (__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4)
/**
* g_atomic_int_get:
* @atomic: a pointer to a #gint or #guint
*
* Gets the current value of @atomic.
*
* This call acts as a full compiler and hardware
* memory barrier (before the get).
*
* Returns: the value of the integer
*
* Since: 2.4
**/
gint
(g_atomic_int_get) (const volatile gint *atomic)
{
return g_atomic_int_get (atomic);
}
/**
* g_atomic_int_set:
* @atomic: a pointer to a #gint or #guint
* @newval: a new value to store
*
* Sets the value of @atomic to @newval.
*
* This call acts as a full compiler and hardware
* memory barrier (after the set).
*
* Since: 2.4
*/
void
(g_atomic_int_set) (volatile gint *atomic,
gint newval)
{
g_atomic_int_set (atomic, newval);
}
/**
* g_atomic_int_inc:
* @atomic: a pointer to a #gint or #guint
*
* Increments the value of @atomic by 1.
*
* Think of this operation as an atomic version of `{ *atomic += 1; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Since: 2.4
**/
void
(g_atomic_int_inc) (volatile gint *atomic)
{
g_atomic_int_inc (atomic);
}
/**
* g_atomic_int_dec_and_test:
* @atomic: a pointer to a #gint or #guint
*
* Decrements the value of @atomic by 1.
*
* Think of this operation as an atomic version of
* `{ *atomic -= 1; return (*atomic == 0); }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Returns: %TRUE if the resultant value is zero
*
* Since: 2.4
**/
gboolean
(g_atomic_int_dec_and_test) (volatile gint *atomic)
{
return g_atomic_int_dec_and_test (atomic);
}
/**
* g_atomic_int_compare_and_exchange:
* @atomic: a pointer to a #gint or #guint
* @oldval: the value to compare with
* @newval: the value to conditionally replace with
*
* Compares @atomic to @oldval and, if equal, sets it to @newval.
* If @atomic was not equal to @oldval then no change occurs.
*
* This compare and exchange is done atomically.
*
* Think of this operation as an atomic version of
* `{ if (*atomic == oldval) { *atomic = newval; return TRUE; } else return FALSE; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Returns: %TRUE if the exchange took place
*
* Since: 2.4
**/
gboolean
(g_atomic_int_compare_and_exchange) (volatile gint *atomic,
gint oldval,
gint newval)
{
return g_atomic_int_compare_and_exchange (atomic, oldval, newval);
}
/**
* g_atomic_int_add:
* @atomic: a pointer to a #gint or #guint
* @val: the value to add
*
* Atomically adds @val to the value of @atomic.
*
* Think of this operation as an atomic version of
* `{ tmp = *atomic; *atomic += val; return tmp; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Before version 2.30, this function did not return a value
* (but g_atomic_int_exchange_and_add() did, and had the same meaning).
*
* Returns: the value of @atomic before the add, signed
*
* Since: 2.4
**/
gint
(g_atomic_int_add) (volatile gint *atomic,
gint val)
{
return g_atomic_int_add (atomic, val);
}
/**
* g_atomic_int_and:
* @atomic: a pointer to a #gint or #guint
* @val: the value to 'and'
*
* Performs an atomic bitwise 'and' of the value of @atomic and @val,
* storing the result back in @atomic.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Think of this operation as an atomic version of
* `{ tmp = *atomic; *atomic &= val; return tmp; }`.
*
* Returns: the value of @atomic before the operation, unsigned
*
* Since: 2.30
**/
guint
(g_atomic_int_and) (volatile guint *atomic,
guint val)
{
return g_atomic_int_and (atomic, val);
}
/**
* g_atomic_int_or:
* @atomic: a pointer to a #gint or #guint
* @val: the value to 'or'
*
* Performs an atomic bitwise 'or' of the value of @atomic and @val,
* storing the result back in @atomic.
*
* Think of this operation as an atomic version of
* `{ tmp = *atomic; *atomic |= val; return tmp; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Returns: the value of @atomic before the operation, unsigned
*
* Since: 2.30
**/
guint
(g_atomic_int_or) (volatile guint *atomic,
guint val)
{
return g_atomic_int_or (atomic, val);
}
/**
* g_atomic_int_xor:
* @atomic: a pointer to a #gint or #guint
* @val: the value to 'xor'
*
* Performs an atomic bitwise 'xor' of the value of @atomic and @val,
* storing the result back in @atomic.
*
* Think of this operation as an atomic version of
* `{ tmp = *atomic; *atomic ^= val; return tmp; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Returns: the value of @atomic before the operation, unsigned
*
* Since: 2.30
**/
guint
(g_atomic_int_xor) (volatile guint *atomic,
guint val)
{
return g_atomic_int_xor (atomic, val);
}
/**
* g_atomic_pointer_get:
* @atomic: (not nullable): a pointer to a #gpointer-sized value
*
* Gets the current value of @atomic.
*
* This call acts as a full compiler and hardware
* memory barrier (before the get).
*
* Returns: the value of the pointer
*
* Since: 2.4
**/
gpointer
(g_atomic_pointer_get) (const volatile void *atomic)
{
return g_atomic_pointer_get ((const volatile gpointer *) atomic);
}
/**
* g_atomic_pointer_set:
* @atomic: (not nullable): a pointer to a #gpointer-sized value
* @newval: a new value to store
*
* Sets the value of @atomic to @newval.
*
* This call acts as a full compiler and hardware
* memory barrier (after the set).
*
* Since: 2.4
**/
void
(g_atomic_pointer_set) (volatile void *atomic,
gpointer newval)
{
g_atomic_pointer_set ((volatile gpointer *) atomic, newval);
}
/**
* g_atomic_pointer_compare_and_exchange:
* @atomic: (not nullable): a pointer to a #gpointer-sized value
* @oldval: the value to compare with
* @newval: the value to conditionally replace with
*
* Compares @atomic to @oldval and, if equal, sets it to @newval.
* If @atomic was not equal to @oldval then no change occurs.
*
* This compare and exchange is done atomically.
*
* Think of this operation as an atomic version of
* `{ if (*atomic == oldval) { *atomic = newval; return TRUE; } else return FALSE; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Returns: %TRUE if the exchange took place
*
* Since: 2.4
**/
gboolean
(g_atomic_pointer_compare_and_exchange) (volatile void *atomic,
gpointer oldval,
gpointer newval)
{
return g_atomic_pointer_compare_and_exchange ((volatile gpointer *) atomic,
oldval, newval);
}
/**
* g_atomic_pointer_add:
* @atomic: (not nullable): a pointer to a #gpointer-sized value
* @val: the value to add
*
* Atomically adds @val to the value of @atomic.
*
* Think of this operation as an atomic version of
* `{ tmp = *atomic; *atomic += val; return tmp; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Returns: the value of @atomic before the add, signed
*
* Since: 2.30
**/
gssize
(g_atomic_pointer_add) (volatile void *atomic,
gssize val)
{
return g_atomic_pointer_add ((volatile gpointer *) atomic, val);
}
/**
* g_atomic_pointer_and:
* @atomic: (not nullable): a pointer to a #gpointer-sized value
* @val: the value to 'and'
*
* Performs an atomic bitwise 'and' of the value of @atomic and @val,
* storing the result back in @atomic.
*
* Think of this operation as an atomic version of
* `{ tmp = *atomic; *atomic &= val; return tmp; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Returns: the value of @atomic before the operation, unsigned
*
* Since: 2.30
**/
gsize
(g_atomic_pointer_and) (volatile void *atomic,
gsize val)
{
return g_atomic_pointer_and ((volatile gpointer *) atomic, val);
}
/**
* g_atomic_pointer_or:
* @atomic: (not nullable): a pointer to a #gpointer-sized value
* @val: the value to 'or'
*
* Performs an atomic bitwise 'or' of the value of @atomic and @val,
* storing the result back in @atomic.
*
* Think of this operation as an atomic version of
* `{ tmp = *atomic; *atomic |= val; return tmp; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Returns: the value of @atomic before the operation, unsigned
*
* Since: 2.30
**/
gsize
(g_atomic_pointer_or) (volatile void *atomic,
gsize val)
{
return g_atomic_pointer_or ((volatile gpointer *) atomic, val);
}
/**
* g_atomic_pointer_xor:
* @atomic: (not nullable): a pointer to a #gpointer-sized value
* @val: the value to 'xor'
*
* Performs an atomic bitwise 'xor' of the value of @atomic and @val,
* storing the result back in @atomic.
*
* Think of this operation as an atomic version of
* `{ tmp = *atomic; *atomic ^= val; return tmp; }`.
*
* This call acts as a full compiler and hardware memory barrier.
*
* Returns: the value of @atomic before the operation, unsigned
*
* Since: 2.30
**/
gsize
(g_atomic_pointer_xor) (volatile void *atomic,
gsize val)
{
return g_atomic_pointer_xor ((volatile gpointer *) atomic, val);
}
#elif defined (G_PLATFORM_WIN32)
#include <windows.h>
#if !defined(_M_AMD64) && !defined (_M_IA64) && !defined(_M_X64) && !(defined _MSC_VER && _MSC_VER <= 1200)
#define InterlockedAnd _InterlockedAnd
#define InterlockedOr _InterlockedOr
#define InterlockedXor _InterlockedXor
#endif
#if !defined (_MSC_VER) || _MSC_VER <= 1200
#include "gmessages.h"
/* Inlined versions for older compiler */
static LONG
_gInterlockedAnd (volatile guint *atomic,
guint val)
{
LONG i, j;
j = *atomic;
do {
i = j;
j = InterlockedCompareExchange(atomic, i & val, i);
} while (i != j);
return j;
}
#define InterlockedAnd(a,b) _gInterlockedAnd(a,b)
static LONG
_gInterlockedOr (volatile guint *atomic,
guint val)
{
LONG i, j;
j = *atomic;
do {
i = j;
j = InterlockedCompareExchange(atomic, i | val, i);
} while (i != j);
return j;
}
#define InterlockedOr(a,b) _gInterlockedOr(a,b)
static LONG
_gInterlockedXor (volatile guint *atomic,
guint val)
{
LONG i, j;
j = *atomic;
do {
i = j;
j = InterlockedCompareExchange(atomic, i ^ val, i);
} while (i != j);
return j;
}
#define InterlockedXor(a,b) _gInterlockedXor(a,b)
#endif
/*
* http://msdn.microsoft.com/en-us/library/ms684122(v=vs.85).aspx
*/
gint
(g_atomic_int_get) (const volatile gint *atomic)
{
MemoryBarrier ();
return *atomic;
}
void
(g_atomic_int_set) (volatile gint *atomic,
gint newval)
{
*atomic = newval;
MemoryBarrier ();
}
void
(g_atomic_int_inc) (volatile gint *atomic)
{
InterlockedIncrement (atomic);
}
gboolean
(g_atomic_int_dec_and_test) (volatile gint *atomic)
{
return InterlockedDecrement (atomic) == 0;
}
gboolean
(g_atomic_int_compare_and_exchange) (volatile gint *atomic,
gint oldval,
gint newval)
{
return InterlockedCompareExchange (atomic, newval, oldval) == oldval;
}
gint
(g_atomic_int_add) (volatile gint *atomic,
gint val)
{
return InterlockedExchangeAdd (atomic, val);
}
guint
(g_atomic_int_and) (volatile guint *atomic,
guint val)
{
return InterlockedAnd (atomic, val);
}
guint
(g_atomic_int_or) (volatile guint *atomic,
guint val)
{
return InterlockedOr (atomic, val);
}
guint
(g_atomic_int_xor) (volatile guint *atomic,
guint val)
{
return InterlockedXor (atomic, val);
}
gpointer
(g_atomic_pointer_get) (const volatile void *atomic)
{
const volatile gpointer *ptr = atomic;
MemoryBarrier ();
return *ptr;
}
void
(g_atomic_pointer_set) (volatile void *atomic,
gpointer newval)
{
volatile gpointer *ptr = atomic;
*ptr = newval;
MemoryBarrier ();
}
gboolean
(g_atomic_pointer_compare_and_exchange) (volatile void *atomic,
gpointer oldval,
gpointer newval)
{
return InterlockedCompareExchangePointer (atomic, newval, oldval) == oldval;
}
gssize
(g_atomic_pointer_add) (volatile void *atomic,
gssize val)
{
#if GLIB_SIZEOF_VOID_P == 8
return InterlockedExchangeAdd64 (atomic, val);
#else
return InterlockedExchangeAdd (atomic, val);
#endif
}
gsize
(g_atomic_pointer_and) (volatile void *atomic,
gsize val)
{
#if GLIB_SIZEOF_VOID_P == 8
return InterlockedAnd64 (atomic, val);
#else
return InterlockedAnd (atomic, val);
#endif
}
gsize
(g_atomic_pointer_or) (volatile void *atomic,
gsize val)
{
#if GLIB_SIZEOF_VOID_P == 8
return InterlockedOr64 (atomic, val);
#else
return InterlockedOr (atomic, val);
#endif
}
gsize
(g_atomic_pointer_xor) (volatile void *atomic,
gsize val)
{
#if GLIB_SIZEOF_VOID_P == 8
return InterlockedXor64 (atomic, val);
#else
return InterlockedXor (atomic, val);
#endif
}
#else
/* This error occurs when `meson configure` decided that we should be capable
* of lock-free atomics but we find at compile-time that we are not.
*/
#error G_ATOMIC_LOCK_FREE defined, but incapable of lock-free atomics.
#endif /* defined (__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4) */
#else /* G_ATOMIC_LOCK_FREE */
/* We are not permitted to call into any GLib functions from here, so we
* can not use GMutex.
*
* Fortunately, we already take care of the Windows case above, and all
* non-Windows platforms on which glib runs have pthreads. Use those.
*/
#include <pthread.h>
static pthread_mutex_t g_atomic_lock = PTHREAD_MUTEX_INITIALIZER;
gint
(g_atomic_int_get) (const volatile gint *atomic)
{
gint value;
pthread_mutex_lock (&g_atomic_lock);
value = *atomic;
pthread_mutex_unlock (&g_atomic_lock);
return value;
}
void
(g_atomic_int_set) (volatile gint *atomic,
gint value)
{
pthread_mutex_lock (&g_atomic_lock);
*atomic = value;
pthread_mutex_unlock (&g_atomic_lock);
}
void
(g_atomic_int_inc) (volatile gint *atomic)
{
pthread_mutex_lock (&g_atomic_lock);
(*atomic)++;
pthread_mutex_unlock (&g_atomic_lock);
}
gboolean
(g_atomic_int_dec_and_test) (volatile gint *atomic)
{
gboolean is_zero;
pthread_mutex_lock (&g_atomic_lock);
is_zero = --(*atomic) == 0;
pthread_mutex_unlock (&g_atomic_lock);
return is_zero;
}
gboolean
(g_atomic_int_compare_and_exchange) (volatile gint *atomic,
gint oldval,
gint newval)
{
gboolean success;
pthread_mutex_lock (&g_atomic_lock);
if ((success = (*atomic == oldval)))
*atomic = newval;
pthread_mutex_unlock (&g_atomic_lock);
return success;
}
gint
(g_atomic_int_add) (volatile gint *atomic,
gint val)
{
gint oldval;
pthread_mutex_lock (&g_atomic_lock);
oldval = *atomic;
*atomic = oldval + val;
pthread_mutex_unlock (&g_atomic_lock);
return oldval;
}
guint
(g_atomic_int_and) (volatile guint *atomic,
guint val)
{
guint oldval;
pthread_mutex_lock (&g_atomic_lock);
oldval = *atomic;
*atomic = oldval & val;
pthread_mutex_unlock (&g_atomic_lock);
return oldval;
}
guint
(g_atomic_int_or) (volatile guint *atomic,
guint val)
{
guint oldval;
pthread_mutex_lock (&g_atomic_lock);
oldval = *atomic;
*atomic = oldval | val;
pthread_mutex_unlock (&g_atomic_lock);
return oldval;
}
guint
(g_atomic_int_xor) (volatile guint *atomic,
guint val)
{
guint oldval;
pthread_mutex_lock (&g_atomic_lock);
oldval = *atomic;
*atomic = oldval ^ val;
pthread_mutex_unlock (&g_atomic_lock);
return oldval;
}
gpointer
(g_atomic_pointer_get) (const volatile void *atomic)
{
const volatile gpointer *ptr = atomic;
gpointer value;
pthread_mutex_lock (&g_atomic_lock);
value = *ptr;
pthread_mutex_unlock (&g_atomic_lock);
return value;
}
void
(g_atomic_pointer_set) (volatile void *atomic,
gpointer newval)
{
volatile gpointer *ptr = atomic;
pthread_mutex_lock (&g_atomic_lock);
*ptr = newval;
pthread_mutex_unlock (&g_atomic_lock);
}
gboolean
(g_atomic_pointer_compare_and_exchange) (volatile void *atomic,
gpointer oldval,
gpointer newval)
{
volatile gpointer *ptr = atomic;
gboolean success;
pthread_mutex_lock (&g_atomic_lock);
if ((success = (*ptr == oldval)))
*ptr = newval;
pthread_mutex_unlock (&g_atomic_lock);
return success;
}
gssize
(g_atomic_pointer_add) (volatile void *atomic,
gssize val)
{
volatile gssize *ptr = atomic;
gssize oldval;
pthread_mutex_lock (&g_atomic_lock);
oldval = *ptr;
*ptr = oldval + val;
pthread_mutex_unlock (&g_atomic_lock);
return oldval;
}
gsize
(g_atomic_pointer_and) (volatile void *atomic,
gsize val)
{
volatile gsize *ptr = atomic;
gsize oldval;
pthread_mutex_lock (&g_atomic_lock);
oldval = *ptr;
*ptr = oldval & val;
pthread_mutex_unlock (&g_atomic_lock);
return oldval;
}
gsize
(g_atomic_pointer_or) (volatile void *atomic,
gsize val)
{
volatile gsize *ptr = atomic;
gsize oldval;
pthread_mutex_lock (&g_atomic_lock);
oldval = *ptr;
*ptr = oldval | val;
pthread_mutex_unlock (&g_atomic_lock);
return oldval;
}
gsize
(g_atomic_pointer_xor) (volatile void *atomic,
gsize val)
{
volatile gsize *ptr = atomic;
gsize oldval;
pthread_mutex_lock (&g_atomic_lock);
oldval = *ptr;
*ptr = oldval ^ val;
pthread_mutex_unlock (&g_atomic_lock);
return oldval;
}
#endif
/**
* g_atomic_int_exchange_and_add:
* @atomic: a pointer to a #gint
* @val: the value to add
*
* This function existed before g_atomic_int_add() returned the prior
* value of the integer (which it now does). It is retained only for
* compatibility reasons. Don't use this function in new code.
*
* Returns: the value of @atomic before the add, signed
* Since: 2.4
* Deprecated: 2.30: Use g_atomic_int_add() instead.
**/
gint
g_atomic_int_exchange_and_add (volatile gint *atomic,
gint val)
{
return (g_atomic_int_add) (atomic, val);
}