michael@0: /*-
michael@0:  * Copyright (c) 2009-2010 Brad Penoff
michael@0:  * Copyright (c) 2009-2010 Humaira Kamal
michael@0:  * Copyright (c) 2011-2012 Irene Ruengeler
michael@0:  * Copyright (c) 2011-2012 Michael Tuexen
michael@0:  *
michael@0:  * All rights reserved.
michael@0:  *
michael@0:  * Redistribution and use in source and binary forms, with or without
michael@0:  * modification, are permitted provided that the following conditions
michael@0:  * are met:
michael@0:  * 1. Redistributions of source code must retain the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer.
michael@0:  * 2. Redistributions in binary form must reproduce the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer in the
michael@0:  *    documentation and/or other materials provided with the distribution.
michael@0:  *
michael@0:  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
michael@0:  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
michael@0:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
michael@0:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
michael@0:  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
michael@0:  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
michael@0:  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
michael@0:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
michael@0:  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
michael@0:  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
michael@0:  * SUCH DAMAGE.
michael@0:  */
michael@0: 
michael@0: #ifndef _USER_ATOMIC_H_
michael@0: #define _USER_ATOMIC_H_
michael@0: 
michael@0: /* __Userspace__ version of sys/i386/include/atomic.h goes here */
michael@0: 
michael@0: /* TODO In the future, might want to not use i386 specific assembly.
michael@0:  *    The options include:
michael@0:  *       - implement them generically (but maybe not truly atomic?) in userspace
michael@0:  *       - have ifdef's for __Userspace_arch_ perhaps (OS isn't enough...)
michael@0:  */
michael@0: 
michael@0: #include <stdio.h>
michael@0: #include <sys/types.h>
michael@0: 
michael@0: #if defined(__Userspace_os_Darwin) || defined (__Userspace_os_Windows)
michael@0: #if defined (__Userspace_os_Windows)
michael@0: #define atomic_add_int(addr, val) InterlockedExchangeAdd((LPLONG)addr, (LONG)val)
michael@0: #define atomic_fetchadd_int(addr, val) InterlockedExchangeAdd((LPLONG)addr, (LONG)val)
michael@0: #define atomic_subtract_int(addr, val)   InterlockedExchangeAdd((LPLONG)addr,-((LONG)val))
michael@0: #define atomic_cmpset_int(dst, exp, src) InterlockedCompareExchange((LPLONG)dst, src, exp)
michael@0: #define SCTP_DECREMENT_AND_CHECK_REFCOUNT(addr) (InterlockedExchangeAdd((LPLONG)addr, (-1L)) == 1)
michael@0: #else
michael@0: #include <libkern/OSAtomic.h>
michael@0: #define atomic_add_int(addr, val) OSAtomicAdd32Barrier(val, (int32_t *)addr)
michael@0: #define atomic_fetchadd_int(addr, val) OSAtomicAdd32Barrier(val, (int32_t *)addr)
michael@0: #define atomic_subtract_int(addr, val) OSAtomicAdd32Barrier(-val, (int32_t *)addr)
michael@0: #define atomic_cmpset_int(dst, exp, src) OSAtomicCompareAndSwapIntBarrier(exp, src, (int *)dst)
michael@0: #define SCTP_DECREMENT_AND_CHECK_REFCOUNT(addr) (atomic_fetchadd_int(addr, -1) == 0)
michael@0: #endif
michael@0: 
michael@0: #if defined(INVARIANTS)
michael@0: #define SCTP_SAVE_ATOMIC_DECREMENT(addr, val) \
michael@0: { \
michael@0: 	int32_t newval; \
michael@0: 	newval = atomic_fetchadd_int(addr, -val); \
michael@0: 	if (newval < 0) { \
michael@0: 		panic("Counter goes negative"); \
michael@0: 	} \
michael@0: }
michael@0: #else
michael@0: #define SCTP_SAVE_ATOMIC_DECREMENT(addr, val) \
michael@0: { \
michael@0: 	int32_t newval; \
michael@0: 	newval = atomic_fetchadd_int(addr, -val); \
michael@0: 	if (newval < 0) { \
michael@0: 		*addr = 0; \
michael@0: 	} \
michael@0: }
michael@0: #if defined(__Userspace_os_Windows)
michael@0: static void atomic_init() {} /* empty when we are not using atomic_mtx */
michael@0: #else
michael@0: static inline void atomic_init() {} /* empty when we are not using atomic_mtx */
michael@0: #endif
michael@0: #endif
michael@0: 
michael@0: #else
michael@0: /* Using gcc built-in functions for atomic memory operations
michael@0:    Reference: http://gcc.gnu.org/onlinedocs/gcc-4.1.0/gcc/Atomic-Builtins.html
michael@0:    Requires gcc version 4.1.0
michael@0:    compile with -march=i486
michael@0:  */
michael@0: 
michael@0: /*Atomically add V to *P.*/
michael@0: #define atomic_add_int(P, V)	 (void) __sync_fetch_and_add(P, V)
michael@0: 
michael@0: /*Atomically subtrace V from *P.*/
michael@0: #define atomic_subtract_int(P, V) (void) __sync_fetch_and_sub(P, V)
michael@0: 
michael@0: /*
michael@0:  * Atomically add the value of v to the integer pointed to by p and return
michael@0:  * the previous value of *p.
michael@0:  */
michael@0: #define atomic_fetchadd_int(p, v) __sync_fetch_and_add(p, v)
michael@0: 
michael@0: /* Following explanation from src/sys/i386/include/atomic.h,
michael@0:  * for atomic compare and set
michael@0:  *
michael@0:  * if (*dst == exp) *dst = src (all 32 bit words)
michael@0:  *
michael@0:  * Returns 0 on failure, non-zero on success
michael@0:  */
michael@0: 
michael@0: #define atomic_cmpset_int(dst, exp, src) __sync_bool_compare_and_swap(dst, exp, src)
michael@0: 
michael@0: #define SCTP_DECREMENT_AND_CHECK_REFCOUNT(addr) (atomic_fetchadd_int(addr, -1) == 1)
michael@0: #if defined(INVARIANTS)
michael@0: #define SCTP_SAVE_ATOMIC_DECREMENT(addr, val) \
michael@0: { \
michael@0: 	int32_t oldval; \
michael@0: 	oldval = atomic_fetchadd_int(addr, -val); \
michael@0: 	if (oldval < val) { \
michael@0: 		panic("Counter goes negative"); \
michael@0: 	} \
michael@0: }
michael@0: #else
michael@0: #define SCTP_SAVE_ATOMIC_DECREMENT(addr, val) \
michael@0: { \
michael@0: 	int32_t oldval; \
michael@0: 	oldval = atomic_fetchadd_int(addr, -val); \
michael@0: 	if (oldval < val) { \
michael@0: 		*addr = 0; \
michael@0: 	} \
michael@0: }
michael@0: #endif
michael@0: static inline void atomic_init() {} /* empty when we are not using atomic_mtx */
michael@0: #endif
michael@0: 
michael@0: #if 0 /* using libatomic_ops */
michael@0: #include "user_include/atomic_ops.h"
michael@0: 
michael@0: /*Atomically add incr to *P, and return the original value of *P.*/
michael@0: #define atomic_add_int(P, V)	 AO_fetch_and_add((AO_t*)P, V)
michael@0: 
michael@0: #define atomic_subtract_int(P, V) AO_fetch_and_add((AO_t*)P, -(V))
michael@0: 
michael@0: /*
michael@0:  * Atomically add the value of v to the integer pointed to by p and return
michael@0:  * the previous value of *p.
michael@0:  */
michael@0: #define atomic_fetchadd_int(p, v) AO_fetch_and_add((AO_t*)p, v)
michael@0: 
michael@0: /* Atomically compare *addr to old_val, and replace *addr by new_val
michael@0:    if the first comparison succeeds.  Returns nonzero if the comparison
michael@0:    succeeded and *addr was updated.
michael@0: */
michael@0: /* Following Explanation from src/sys/i386/include/atomic.h, which
michael@0:    matches that of AO_compare_and_swap above.
michael@0:  * Atomic compare and set, used by the mutex functions
michael@0:  *
michael@0:  * if (*dst == exp) *dst = src (all 32 bit words)
michael@0:  *
michael@0:  * Returns 0 on failure, non-zero on success
michael@0:  */
michael@0: 
michael@0: #define atomic_cmpset_int(dst, exp, src) AO_compare_and_swap((AO_t*)dst, exp, src)
michael@0: 
michael@0: static inline void atomic_init() {} /* empty when we are not using atomic_mtx */
michael@0: #endif /* closing #if for libatomic */
michael@0: 
michael@0: #if 0 /* using atomic_mtx */
michael@0: 
michael@0: #include <pthread.h>
michael@0: 
michael@0: extern userland_mutex_t atomic_mtx;
michael@0: 
michael@0: #if defined (__Userspace_os_Windows)
michael@0: static inline void atomic_init() {
michael@0: 	InitializeCriticalSection(&atomic_mtx);
michael@0: }
michael@0: static inline void atomic_destroy() {
michael@0: 	DeleteCriticalSection(&atomic_mtx);
michael@0: }
michael@0: static inline void atomic_lock() {
michael@0: 	EnterCriticalSection(&atomic_mtx);
michael@0: }
michael@0: static inline void atomic_unlock() {
michael@0: 	LeaveCriticalSection(&atomic_mtx);
michael@0: }
michael@0: #else
michael@0: static inline void atomic_init() {
michael@0: 	(void)pthread_mutex_init(&atomic_mtx, NULL);
michael@0: }
michael@0: static inline void atomic_destroy() {
michael@0: 	(void)pthread_mutex_destroy(&atomic_mtx);
michael@0: }
michael@0: static inline void atomic_lock() {
michael@0: 	(void)pthread_mutex_lock(&atomic_mtx);
michael@0: }
michael@0: static inline void atomic_unlock() {
michael@0: 	(void)pthread_mutex_unlock(&atomic_mtx);
michael@0: }
michael@0: #endif
michael@0: /*
michael@0:  * For userland, always use lock prefixes so that the binaries will run
michael@0:  * on both SMP and !SMP systems.
michael@0:  */
michael@0: 
michael@0: #define	MPLOCKED	"lock ; "
michael@0: 
michael@0: /*
michael@0:  * Atomically add the value of v to the integer pointed to by p and return
michael@0:  * the previous value of *p.
michael@0:  */
michael@0: static __inline u_int
michael@0: atomic_fetchadd_int(volatile void *n, u_int v)
michael@0: {
michael@0: 	int *p = (int *) n;
michael@0: 	atomic_lock();
michael@0: 	__asm __volatile(
michael@0: 	"	" MPLOCKED "		"
michael@0: 	"	xaddl	%0, %1 ;	"
michael@0: 	"# atomic_fetchadd_int"
michael@0: 	: "+r" (v),			/* 0 (result) */
michael@0: 	  "=m" (*p)			/* 1 */
michael@0: 	: "m" (*p));			/* 2 */
michael@0: 	atomic_unlock();
michael@0: 
michael@0: 	return (v);
michael@0: }
michael@0: 
michael@0: 
michael@0: #ifdef CPU_DISABLE_CMPXCHG
michael@0: 
michael@0: static __inline int
michael@0: atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
michael@0: {
michael@0: 	u_char res;
michael@0: 
michael@0: 	atomic_lock();
michael@0: 	__asm __volatile(
michael@0: 	"	pushfl ;		"
michael@0: 	"	cli ;			"
michael@0: 	"	cmpl	%3,%4 ;		"
michael@0: 	"	jne	1f ;		"
michael@0: 	"	movl	%2,%1 ;		"
michael@0: 	"1:				"
michael@0: 	"       sete	%0 ;		"
michael@0: 	"	popfl ;			"
michael@0: 	"# atomic_cmpset_int"
michael@0: 	: "=q" (res),			/* 0 */
michael@0: 	  "=m" (*dst)			/* 1 */
michael@0: 	: "r" (src),			/* 2 */
michael@0: 	  "r" (exp),			/* 3 */
michael@0: 	  "m" (*dst)			/* 4 */
michael@0: 	: "memory");
michael@0: 	atomic_unlock();
michael@0: 
michael@0: 	return (res);
michael@0: }
michael@0: 
michael@0: #else /* !CPU_DISABLE_CMPXCHG */
michael@0: 
michael@0: static __inline int
michael@0: atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
michael@0: {
michael@0: 	atomic_lock();
michael@0: 	u_char res;
michael@0: 
michael@0: 	__asm __volatile(
michael@0: 	"	" MPLOCKED "		"
michael@0: 	"	cmpxchgl %2,%1 ;	"
michael@0: 	"	sete	%0 ;		"
michael@0: 	"1:				"
michael@0: 	"# atomic_cmpset_int"
michael@0: 	: "=a" (res),			/* 0 */
michael@0: 	  "=m" (*dst)			/* 1 */
michael@0: 	: "r" (src),			/* 2 */
michael@0: 	  "a" (exp),			/* 3 */
michael@0: 	  "m" (*dst)			/* 4 */
michael@0: 	: "memory");
michael@0: 	atomic_unlock();
michael@0: 
michael@0: 	return (res);
michael@0: }
michael@0: 
michael@0: #endif /* CPU_DISABLE_CMPXCHG */
michael@0: 
michael@0: #define atomic_add_int(P, V)	 do {   \
michael@0: 		atomic_lock();          \
michael@0: 		(*(u_int *)(P) += (V)); \
michael@0: 		atomic_unlock();        \
michael@0: } while(0)
michael@0: #define atomic_subtract_int(P, V)  do {   \
michael@0: 		atomic_lock();            \
michael@0: 		(*(u_int *)(P) -= (V));   \
michael@0: 		atomic_unlock();          \
michael@0: } while(0)
michael@0: 
michael@0: #endif
michael@0: #endif