atomic.3   [plain text]


.Dd May 26, 2004
.Dt ATOMIC 3
.Os Darwin
.Sh NAME
.Nm OSAtomicAdd32 ,
.Nm OSAtomicAdd32Barrier ,
.Nm OSAtomicIncrement32 ,
.Nm OSAtomicIncrement32Barrier ,
.Nm OSAtomicDecrement32 ,
.Nm OSAtomicDecrement32Barrier ,
.Nm OSAtomicOr32 ,
.Nm OSAtomicOr32Barrier ,
.Nm OSAtomicOr32Orig ,
.Nm OSAtomicOr32OrigBarrier ,
.Nm OSAtomicAnd32 ,
.Nm OSAtomicAnd32Barrier ,
.Nm OSAtomicAnd32Orig ,
.Nm OSAtomicAnd32OrigBarrier ,
.Nm OSAtomicXor32 ,
.Nm OSAtomicXor32Barrier ,
.Nm OSAtomicXor32Orig ,
.Nm OSAtomicXor32OrigBarrier ,
.Nm OSAtomicAdd64 ,
.Nm OSAtomicAdd64Barrier ,
.Nm OSAtomicIncrement64 ,
.Nm OSAtomicIncrement64Barrier ,
.Nm OSAtomicDecrement64 ,
.Nm OSAtomicDecrement64Barrier ,
.Nm OSAtomicCompareAndSwapInt ,
.Nm OSAtomicCompareAndSwapIntBarrier ,
.Nm OSAtomicCompareAndSwapLong ,
.Nm OSAtomicCompareAndSwapLongBarrier ,
.Nm OSAtomicCompareAndSwapPtr ,
.Nm OSAtomicCompareAndSwapPtrBarrier ,
.Nm OSAtomicCompareAndSwap32 ,
.Nm OSAtomicCompareAndSwap32Barrier ,
.Nm OSAtomicCompareAndSwap64 ,
.Nm OSAtomicCompareAndSwap64Barrier ,
.Nm OSAtomicTestAndSet ,
.Nm OSAtomicTestAndSetBarrier ,
.Nm OSAtomicTestAndClear ,
.Nm OSAtomicTestAndClearBarrier ,
.Nm OSSpinLockTry ,
.Nm OSSpinLockLock ,
.Nm OSSpinLockUnlock ,
.Nm OSAtomicEnqueue ,
.Nm OSAtomicDequeue
.Nd atomic add, increment, decrement, or, and, xor, compare and swap, test and set, test and clear, spinlocks, and lockless queues 
.Sh LIBRARY
.Lb libc
.Sh SYNOPSIS
.In libkern/OSAtomic.h
.Ft int32_t
.Fn OSAtomicAdd32 "int32_t theAmount" "volatile int32_t *theValue"
.Ft int32_t
.Fn OSAtomicAdd32Barrier "int32_t theAmount" "volatile int32_t *theValue"
.Ft int32_t
.Fn OSAtomicIncrement32 "volatile int32_t *theValue"
.Ft int32_t
.Fn OSAtomicIncrement32Barrier "volatile int32_t *theValue"
.Ft int32_t
.Fn OSAtomicDecrement32 "volatile int32_t *theValue"
.Ft int32_t
.Fn OSAtomicDecrement32Barrier "volatile int32_t *theValue"
.Ft int32_t
.Fn OSAtomicOr32 "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicOr32Barrier "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicAnd32 "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicAnd32Barrier "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicXor32 "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicXor32Barrier "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicOr32Orig "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicOr32OrigBarrier "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicAnd32Orig "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicAnd32OrigBarrier "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicXor32Orig "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int32_t
.Fn OSAtomicXor32OrigBarrier "uint32_t theMask" "volatile uint32_t *theValue"
.Ft int64_t
.Fn OSAtomicAdd64 "int64_t theAmount" "volatile int64_t *theValue"
.Ft int64_t
.Fn OSAtomicAdd64Barrier "int64_t theAmount" "volatile int64_t *theValue"
.Ft int64_t
.Fn OSAtomicIncrement64 "volatile int64_t *theValue"
.Ft int64_t
.Fn OSAtomicIncrement64Barrier "volatile int64_t *theValue"
.Ft int64_t
.Fn OSAtomicDecrement64 "volatile int64_t *theValue"
.Ft int64_t
.Fn OSAtomicDecrement64Barrier "volatile int64_t *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwapInt "int oldValue" "int newValue" "volatile int *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwapIntBarrier "int oldValue" "int newValue" "volatile int *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwapLong "long oldValue" "long newValue" "volatile long *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwapLongBarrier "long oldValue" "long newValue" "volatile long *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwapPtr "void* oldValue" "void* newValue" "void* volatile *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwapPtrBarrier "void* oldValue" "void* newValue" "void* volatile *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwap32 "int32_t oldValue" "int32_t newValue" "volatile int32_t *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwap32Barrier "int32_t oldValue" "int32_t newValue" "volatile int32_t *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwap64 "int64_t oldValue" "int64_t newValue" "volatile int64_t *theValue"
.Ft bool
.Fn OSAtomicCompareAndSwap64Barrier "int64_t oldValue" "int64_t newValue" "volatile int64_t *theValue"
.Ft bool
.Fn OSAtomicTestAndSet "uint32_t n" "volatile void *theAddress"
.Ft bool
.Fn OSAtomicTestAndSetBarrier "uint32_t n" "volatile void *theAddress"
.Ft bool
.Fn OSAtomicTestAndClear "uint32_t n" "volatile void *theAddress"
.Ft bool
.Fn OSAtomicTestAndClearBarrier "uint32_t n" "volatile void *theAddress"
.Ft bool
.Fn OSSpinLockTry "OSSpinLock *lock"
.Ft void
.Fn OSSpinLockLock "OSSpinLock *lock"
.Ft void
.Fn OSSpinLockUnlock "OSSpinLock *lock"
.Ft void
.Fn OSAtomicEnqueue "OSQueueHead *list" "void *new" "size_t offset"
.Ft void*
.Fn OSAtomicDequeue "OSQueueHead *list" "size_t offset"
.Sh DESCRIPTION
These functions are thread and multiprocessor safe.  For each function, there
is a version that does and another that does not incorporate a memory barrier.
Barriers strictly order memory access on a weakly-ordered
architecture such as PPC.  All loads and stores executed in sequential program
order before the barrier will complete before any load or store executed after
the barrier.  On a uniprocessor, the barrier operation is typically a nop.
On a multiprocessor, the barrier can be quite expensive.
.Pp
Most code will want to use the barrier functions to ensure that memory shared
between threads is properly synchronized.  For example, if you want to initialize
a shared data structure and then atomically increment a variable to indicate
that the initialization is complete, then you must use OSAtomicIncrement32Barrier()
to ensure that the stores to your data structure complete before the atomic add.
Likewise, the consumer of that data structure must use OSAtomicDecrement32Barrier(),
in order to ensure that their loads of the structure are not executed before
the atomic decrement.  On the other hand,
if you are simply incrementing a global counter, then it is safe and potentially much
faster to use OSAtomicIncrement32().  If you are unsure which version to use, prefer
the barrier variants as they are safer.
.Pp
The logical (and, or, xor) and bit test operations are layered on top of the
.Fn OSAtomicCompareAndSwap
primitives.  There are four versions of each logical operation, depending on whether
or not there is a barrier, and whether the return value is the result of the
operation (eg,
.Fn OSAtomicOr32
) or the original value before the operation (eg,
.Fn OSAtomicOr32Orig
).
.Pp
The memory address
.Fa theValue
must be naturally aligned, ie 32-bit aligned for 32-bit operations and 64-bit
aligned for 64-bit operations.
.Pp
The 64-bit operations are not implemented for 32-bit processes on PPC platforms.
.Pp
The 
.Fn OSAtomicCompareAndSwap
operations compare
.Fa oldValue
to
.Fa *theValue ,
and set
.Fa *theValue
to
.Fa newValue
if the comparison is equal.  The comparison and assignment
occur as one atomic operation.
.Pp
.Fn OSAtomicTestAndSet
and
.Fn OSAtomicTestAndClear
operate on bit (0x80 >> (
.Fa n
& 7)) of byte ((char*)
.Fa theAddress
+ (
.Fa n
>> 3)).  They set the named bit to either 1 or 0, respectively.
.Fa theAddress
need not be aligned.
.Pp
The routines
.Fn OSAtomicEnqueue
and
.Fn OSAtomicDequeue
operate on singly linked LIFO queues.  Ie, a dequeue operation will return the
most recently enqueued element, or NULL if the list is empty.  The operations
are lockless, and barriers are used as necessary to permit thread-safe access to
the queue element.
.Fa offset
is the offset in bytes to the link field in the queue element.  For example:
.Bd -literal -offset indent
	typedef struct elem {
		long	data1;
		struct elem *link;
		int	data2;
	} elem_t;
	
	elem_t fred, mary, *p;
	
	OSQueueHead q = OS_ATOMIC_QUEUE_INIT;
	
	OSAtomicEnqueue( &q, &fred, offsetof(elem_t,link) );
	OSAtomicEnqueue( &q, &mary, offsetof(elem_t,link) );
	
	p = OSAtomicDequeue( &q, offsetof(elem_t,link) );
	
.Ed
In this example, the call of
.Fn OSAtomicDequeue
will return a ptr to mary.
.Sh RETURN VALUES
The arithmetic operations return the new value, after the operation has been performed.
The boolean operations come in two styles, one of which returns the new value, and one
of which (the "Orig" versions) returns the old.
The compare-and-swap operations return true if the comparison was equal, ie if the swap occured.
The bit test and set/clear operations return the original value of the bit.
The dequeue operation returns the most recently enqueued element, or NULL if the list in empty.
.Sh SEE ALSO
.Xr spinlock 3 ,
.Xr barrier 3
.Sh HISTORY
Most of these functions first appeared in Mac OS 10.4 (Tiger).  The "Orig" forms of the
boolean operations, the "int", "long" and "ptr" forms of compare-and-swap, and lockless
enqueue/dequeue first appeared in Mac OS 10.5 (Leopard).