Index: stable/11/sys/amd64/include/atomic.h =================================================================== --- stable/11/sys/amd64/include/atomic.h (revision 315370) +++ stable/11/sys/amd64/include/atomic.h (revision 315371) @@ -1,625 +1,676 @@ /*- * Copyright (c) 1998 Doug Rabson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_H_ #define _MACHINE_ATOMIC_H_ #ifndef _SYS_CDEFS_H_ #error this file needs sys/cdefs.h as a prerequisite #endif /* * To express interprocessor (as opposed to processor and device) memory * ordering constraints, use the atomic_*() functions with acquire and release * semantics rather than the *mb() functions. An architecture's memory * ordering (or memory consistency) model governs the order in which a * program's accesses to different locations may be performed by an * implementation of that architecture. In general, for memory regions * defined as writeback cacheable, the memory ordering implemented by amd64 * processors preserves the program ordering of a load followed by a load, a * load followed by a store, and a store followed by a store. Only a store * followed by a load to a different memory location may be reordered. * Therefore, except for special cases, like non-temporal memory accesses or * memory regions defined as write combining, the memory ordering effects * provided by the sfence instruction in the wmb() function and the lfence * instruction in the rmb() function are redundant. In contrast, the * atomic_*() functions with acquire and release semantics do not perform * redundant instructions for ordinary cases of interprocessor memory * ordering on any architecture. */ #define mb() __asm __volatile("mfence;" : : : "memory") #define wmb() __asm __volatile("sfence;" : : : "memory") #define rmb() __asm __volatile("lfence;" : : : "memory") /* * Various simple operations on memory, each of which is atomic in the * presence of interrupts and multiple processors. * * atomic_set_char(P, V) (*(u_char *)(P) |= (V)) * atomic_clear_char(P, V) (*(u_char *)(P) &= ~(V)) * atomic_add_char(P, V) (*(u_char *)(P) += (V)) * atomic_subtract_char(P, V) (*(u_char *)(P) -= (V)) * * atomic_set_short(P, V) (*(u_short *)(P) |= (V)) * atomic_clear_short(P, V) (*(u_short *)(P) &= ~(V)) * atomic_add_short(P, V) (*(u_short *)(P) += (V)) * atomic_subtract_short(P, V) (*(u_short *)(P) -= (V)) * * atomic_set_int(P, V) (*(u_int *)(P) |= (V)) * atomic_clear_int(P, V) (*(u_int *)(P) &= ~(V)) * atomic_add_int(P, V) (*(u_int *)(P) += (V)) * atomic_subtract_int(P, V) (*(u_int *)(P) -= (V)) * atomic_swap_int(P, V) (return (*(u_int *)(P)); *(u_int *)(P) = (V);) * atomic_readandclear_int(P) (return (*(u_int *)(P)); *(u_int *)(P) = 0;) * * atomic_set_long(P, V) (*(u_long *)(P) |= (V)) * atomic_clear_long(P, V) (*(u_long *)(P) &= ~(V)) * atomic_add_long(P, V) (*(u_long *)(P) += (V)) * atomic_subtract_long(P, V) (*(u_long *)(P) -= (V)) * atomic_swap_long(P, V) (return (*(u_long *)(P)); *(u_long *)(P) = (V);) * atomic_readandclear_long(P) (return (*(u_long *)(P)); *(u_long *)(P) = 0;) */ /* * The above functions are expanded inline in the statically-linked * kernel. Lock prefixes are generated if an SMP kernel is being * built. * * Kernel modules call real functions which are built into the kernel. * This allows kernel modules to be portable between UP and SMP systems. */ #if defined(KLD_MODULE) || !defined(__GNUCLIKE_ASM) #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \ void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v); \ void atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v) int atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src); int atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src); +int atomic_fcmpset_int(volatile u_int *dst, u_int *expect, u_int src); +int atomic_fcmpset_long(volatile u_long *dst, u_long *expect, u_long src); u_int atomic_fetchadd_int(volatile u_int *p, u_int v); u_long atomic_fetchadd_long(volatile u_long *p, u_long v); int atomic_testandset_int(volatile u_int *p, u_int v); int atomic_testandset_long(volatile u_long *p, u_int v); int atomic_testandclear_int(volatile u_int *p, u_int v); int atomic_testandclear_long(volatile u_long *p, u_int v); void atomic_thread_fence_acq(void); void atomic_thread_fence_acq_rel(void); void atomic_thread_fence_rel(void); void atomic_thread_fence_seq_cst(void); #define ATOMIC_LOAD(TYPE) \ u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p) #define ATOMIC_STORE(TYPE) \ void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v) #else /* !KLD_MODULE && __GNUCLIKE_ASM */ /* * For userland, always use lock prefixes so that the binaries will run * on both SMP and !SMP systems. */ #if defined(SMP) || !defined(_KERNEL) #define MPLOCKED "lock ; " #else #define MPLOCKED #endif /* * The assembly is volatilized to avoid code chunk removal by the compiler. * GCC aggressively reorders operations and memory clobbering is necessary * in order to avoid that for memory barriers. */ #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \ static __inline void \ atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ { \ __asm __volatile(MPLOCKED OP \ : "+m" (*p) \ : CONS (V) \ : "cc"); \ } \ \ static __inline void \ atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ { \ __asm __volatile(MPLOCKED OP \ : "+m" (*p) \ : CONS (V) \ : "memory", "cc"); \ } \ struct __hack /* * Atomic compare and set, used by the mutex functions * * if (*dst == expect) *dst = src (all 32 bit words) * * Returns 0 on failure, non-zero on success */ static __inline int atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src) { u_char res; __asm __volatile( " " MPLOCKED " " " cmpxchgl %3,%1 ; " " sete %0 ; " "# atomic_cmpset_int" : "=q" (res), /* 0 */ "+m" (*dst), /* 1 */ "+a" (expect) /* 2 */ : "r" (src) /* 3 */ : "memory", "cc"); return (res); } static __inline int atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src) { u_char res; __asm __volatile( " " MPLOCKED " " " cmpxchgq %3,%1 ; " " sete %0 ; " "# atomic_cmpset_long" : "=q" (res), /* 0 */ "+m" (*dst), /* 1 */ "+a" (expect) /* 2 */ : "r" (src) /* 3 */ : "memory", "cc"); return (res); } +static __inline int +atomic_fcmpset_int(volatile u_int *dst, u_int *expect, u_int src) +{ + u_char res; + + __asm __volatile( + " " MPLOCKED " " + " cmpxchgl %3,%1 ; " + " sete %0 ; " + "# atomic_fcmpset_int" + : "=r" (res), /* 0 */ + "+m" (*dst), /* 1 */ + "+a" (*expect) /* 2 */ + : "r" (src) /* 3 */ + : "memory", "cc"); + return (res); +} + +static __inline int +atomic_fcmpset_long(volatile u_long *dst, u_long *expect, u_long src) +{ + u_char res; + + __asm __volatile( + " " MPLOCKED " " + " cmpxchgq %3,%1 ; " + " sete %0 ; " + "# atomic_fcmpset_long" + : "=r" (res), /* 0 */ + "+m" (*dst), /* 1 */ + "+a" (*expect) /* 2 */ + : "r" (src) /* 3 */ + : "memory", "cc"); + return (res); +} + /* * Atomically add the value of v to the integer pointed to by p and return * the previous value of *p. */ static __inline u_int atomic_fetchadd_int(volatile u_int *p, u_int v) { __asm __volatile( " " MPLOCKED " " " xaddl %0,%1 ; " "# atomic_fetchadd_int" : "+r" (v), /* 0 */ "+m" (*p) /* 1 */ : : "cc"); return (v); } /* * Atomically add the value of v to the long integer pointed to by p and return * the previous value of *p. */ static __inline u_long atomic_fetchadd_long(volatile u_long *p, u_long v) { __asm __volatile( " " MPLOCKED " " " xaddq %0,%1 ; " "# atomic_fetchadd_long" : "+r" (v), /* 0 */ "+m" (*p) /* 1 */ : : "cc"); return (v); } static __inline int atomic_testandset_int(volatile u_int *p, u_int v) { u_char res; __asm __volatile( " " MPLOCKED " " " btsl %2,%1 ; " " setc %0 ; " "# atomic_testandset_int" : "=q" (res), /* 0 */ "+m" (*p) /* 1 */ : "Ir" (v & 0x1f) /* 2 */ : "cc"); return (res); } static __inline int atomic_testandset_long(volatile u_long *p, u_int v) { u_char res; __asm __volatile( " " MPLOCKED " " " btsq %2,%1 ; " " setc %0 ; " "# atomic_testandset_long" : "=q" (res), /* 0 */ "+m" (*p) /* 1 */ : "Jr" ((u_long)(v & 0x3f)) /* 2 */ : "cc"); return (res); } static __inline int atomic_testandclear_int(volatile u_int *p, u_int v) { u_char res; __asm __volatile( " " MPLOCKED " " " btrl %2,%1 ; " " setc %0 ; " "# atomic_testandclear_int" : "=q" (res), /* 0 */ "+m" (*p) /* 1 */ : "Ir" (v & 0x1f) /* 2 */ : "cc"); return (res); } static __inline int atomic_testandclear_long(volatile u_long *p, u_int v) { u_char res; __asm __volatile( " " MPLOCKED " " " btrq %2,%1 ; " " setc %0 ; " "# atomic_testandclear_long" : "=q" (res), /* 0 */ "+m" (*p) /* 1 */ : "Jr" ((u_long)(v & 0x3f)) /* 2 */ : "cc"); return (res); } /* * We assume that a = b will do atomic loads and stores. Due to the * IA32 memory model, a simple store guarantees release semantics. * * However, a load may pass a store if they are performed on distinct * addresses, so we need a Store/Load barrier for sequentially * consistent fences in SMP kernels. We use "lock addl $0,mem" for a * Store/Load barrier, as recommended by the AMD Software Optimization * Guide, and not mfence. To avoid false data dependencies, we use a * special address for "mem". In the kernel, we use a private per-cpu * cache line. In user space, we use a word in the stack's red zone * (-8(%rsp)). * * For UP kernels, however, the memory of the single processor is * always consistent, so we only need to stop the compiler from * reordering accesses in a way that violates the semantics of acquire * and release. */ #if defined(_KERNEL) /* * OFFSETOF_MONITORBUF == __pcpu_offset(pc_monitorbuf). * * The open-coded number is used instead of the symbolic expression to * avoid a dependency on sys/pcpu.h in machine/atomic.h consumers. * An assertion in amd64/vm_machdep.c ensures that the value is correct. */ #define OFFSETOF_MONITORBUF 0x180 #if defined(SMP) static __inline void __storeload_barrier(void) { __asm __volatile("lock; addl $0,%%gs:%0" : "+m" (*(u_int *)OFFSETOF_MONITORBUF) : : "memory", "cc"); } #else /* _KERNEL && UP */ static __inline void __storeload_barrier(void) { __compiler_membar(); } #endif /* SMP */ #else /* !_KERNEL */ static __inline void __storeload_barrier(void) { __asm __volatile("lock; addl $0,-8(%%rsp)" : : : "memory", "cc"); } #endif /* _KERNEL*/ #define ATOMIC_LOAD(TYPE) \ static __inline u_##TYPE \ atomic_load_acq_##TYPE(volatile u_##TYPE *p) \ { \ u_##TYPE res; \ \ res = *p; \ __compiler_membar(); \ return (res); \ } \ struct __hack #define ATOMIC_STORE(TYPE) \ static __inline void \ atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v) \ { \ \ __compiler_membar(); \ *p = v; \ } \ struct __hack static __inline void atomic_thread_fence_acq(void) { __compiler_membar(); } static __inline void atomic_thread_fence_rel(void) { __compiler_membar(); } static __inline void atomic_thread_fence_acq_rel(void) { __compiler_membar(); } static __inline void atomic_thread_fence_seq_cst(void) { __storeload_barrier(); } #endif /* KLD_MODULE || !__GNUCLIKE_ASM */ ATOMIC_ASM(set, char, "orb %b1,%0", "iq", v); ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v); ATOMIC_ASM(add, char, "addb %b1,%0", "iq", v); ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v); ATOMIC_ASM(set, short, "orw %w1,%0", "ir", v); ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v); ATOMIC_ASM(add, short, "addw %w1,%0", "ir", v); ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v); ATOMIC_ASM(set, int, "orl %1,%0", "ir", v); ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v); ATOMIC_ASM(add, int, "addl %1,%0", "ir", v); ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v); ATOMIC_ASM(set, long, "orq %1,%0", "ir", v); ATOMIC_ASM(clear, long, "andq %1,%0", "ir", ~v); ATOMIC_ASM(add, long, "addq %1,%0", "ir", v); ATOMIC_ASM(subtract, long, "subq %1,%0", "ir", v); #define ATOMIC_LOADSTORE(TYPE) \ ATOMIC_LOAD(TYPE); \ ATOMIC_STORE(TYPE) ATOMIC_LOADSTORE(char); ATOMIC_LOADSTORE(short); ATOMIC_LOADSTORE(int); ATOMIC_LOADSTORE(long); #undef ATOMIC_ASM #undef ATOMIC_LOAD #undef ATOMIC_STORE #undef ATOMIC_LOADSTORE #ifndef WANT_FUNCTIONS /* Read the current value and store a new value in the destination. */ #ifdef __GNUCLIKE_ASM static __inline u_int atomic_swap_int(volatile u_int *p, u_int v) { __asm __volatile( " xchgl %1,%0 ; " "# atomic_swap_int" : "+r" (v), /* 0 */ "+m" (*p)); /* 1 */ return (v); } static __inline u_long atomic_swap_long(volatile u_long *p, u_long v) { __asm __volatile( " xchgq %1,%0 ; " "# atomic_swap_long" : "+r" (v), /* 0 */ "+m" (*p)); /* 1 */ return (v); } #else /* !__GNUCLIKE_ASM */ u_int atomic_swap_int(volatile u_int *p, u_int v); u_long atomic_swap_long(volatile u_long *p, u_long v); #endif /* __GNUCLIKE_ASM */ #define atomic_set_acq_char atomic_set_barr_char #define atomic_set_rel_char atomic_set_barr_char #define atomic_clear_acq_char atomic_clear_barr_char #define atomic_clear_rel_char atomic_clear_barr_char #define atomic_add_acq_char atomic_add_barr_char #define atomic_add_rel_char atomic_add_barr_char #define atomic_subtract_acq_char atomic_subtract_barr_char #define atomic_subtract_rel_char atomic_subtract_barr_char #define atomic_set_acq_short atomic_set_barr_short #define atomic_set_rel_short atomic_set_barr_short #define atomic_clear_acq_short atomic_clear_barr_short #define atomic_clear_rel_short atomic_clear_barr_short #define atomic_add_acq_short atomic_add_barr_short #define atomic_add_rel_short atomic_add_barr_short #define atomic_subtract_acq_short atomic_subtract_barr_short #define atomic_subtract_rel_short atomic_subtract_barr_short #define atomic_set_acq_int atomic_set_barr_int #define atomic_set_rel_int atomic_set_barr_int #define atomic_clear_acq_int atomic_clear_barr_int #define atomic_clear_rel_int atomic_clear_barr_int #define atomic_add_acq_int atomic_add_barr_int #define atomic_add_rel_int atomic_add_barr_int #define atomic_subtract_acq_int atomic_subtract_barr_int #define atomic_subtract_rel_int atomic_subtract_barr_int #define atomic_cmpset_acq_int atomic_cmpset_int #define atomic_cmpset_rel_int atomic_cmpset_int +#define atomic_fcmpset_acq_int atomic_fcmpset_int +#define atomic_fcmpset_rel_int atomic_fcmpset_int #define atomic_set_acq_long atomic_set_barr_long #define atomic_set_rel_long atomic_set_barr_long #define atomic_clear_acq_long atomic_clear_barr_long #define atomic_clear_rel_long atomic_clear_barr_long #define atomic_add_acq_long atomic_add_barr_long #define atomic_add_rel_long atomic_add_barr_long #define atomic_subtract_acq_long atomic_subtract_barr_long #define atomic_subtract_rel_long atomic_subtract_barr_long #define atomic_cmpset_acq_long atomic_cmpset_long #define atomic_cmpset_rel_long atomic_cmpset_long +#define atomic_fcmpset_acq_long atomic_fcmpset_long +#define atomic_fcmpset_rel_long atomic_fcmpset_long #define atomic_readandclear_int(p) atomic_swap_int(p, 0) #define atomic_readandclear_long(p) atomic_swap_long(p, 0) /* Operations on 8-bit bytes. */ #define atomic_set_8 atomic_set_char #define atomic_set_acq_8 atomic_set_acq_char #define atomic_set_rel_8 atomic_set_rel_char #define atomic_clear_8 atomic_clear_char #define atomic_clear_acq_8 atomic_clear_acq_char #define atomic_clear_rel_8 atomic_clear_rel_char #define atomic_add_8 atomic_add_char #define atomic_add_acq_8 atomic_add_acq_char #define atomic_add_rel_8 atomic_add_rel_char #define atomic_subtract_8 atomic_subtract_char #define atomic_subtract_acq_8 atomic_subtract_acq_char #define atomic_subtract_rel_8 atomic_subtract_rel_char #define atomic_load_acq_8 atomic_load_acq_char #define atomic_store_rel_8 atomic_store_rel_char /* Operations on 16-bit words. */ #define atomic_set_16 atomic_set_short #define atomic_set_acq_16 atomic_set_acq_short #define atomic_set_rel_16 atomic_set_rel_short #define atomic_clear_16 atomic_clear_short #define atomic_clear_acq_16 atomic_clear_acq_short #define atomic_clear_rel_16 atomic_clear_rel_short #define atomic_add_16 atomic_add_short #define atomic_add_acq_16 atomic_add_acq_short #define atomic_add_rel_16 atomic_add_rel_short #define atomic_subtract_16 atomic_subtract_short #define atomic_subtract_acq_16 atomic_subtract_acq_short #define atomic_subtract_rel_16 atomic_subtract_rel_short #define atomic_load_acq_16 atomic_load_acq_short #define atomic_store_rel_16 atomic_store_rel_short /* Operations on 32-bit double words. */ #define atomic_set_32 atomic_set_int #define atomic_set_acq_32 atomic_set_acq_int #define atomic_set_rel_32 atomic_set_rel_int #define atomic_clear_32 atomic_clear_int #define atomic_clear_acq_32 atomic_clear_acq_int #define atomic_clear_rel_32 atomic_clear_rel_int #define atomic_add_32 atomic_add_int #define atomic_add_acq_32 atomic_add_acq_int #define atomic_add_rel_32 atomic_add_rel_int #define atomic_subtract_32 atomic_subtract_int #define atomic_subtract_acq_32 atomic_subtract_acq_int #define atomic_subtract_rel_32 atomic_subtract_rel_int #define atomic_load_acq_32 atomic_load_acq_int #define atomic_store_rel_32 atomic_store_rel_int #define atomic_cmpset_32 atomic_cmpset_int #define atomic_cmpset_acq_32 atomic_cmpset_acq_int #define atomic_cmpset_rel_32 atomic_cmpset_rel_int +#define atomic_fcmpset_32 atomic_fcmpset_int +#define atomic_fcmpset_acq_32 atomic_fcmpset_acq_int +#define atomic_fcmpset_rel_32 atomic_fcmpset_rel_int #define atomic_swap_32 atomic_swap_int #define atomic_readandclear_32 atomic_readandclear_int #define atomic_fetchadd_32 atomic_fetchadd_int #define atomic_testandset_32 atomic_testandset_int #define atomic_testandclear_32 atomic_testandclear_int /* Operations on 64-bit quad words. */ #define atomic_set_64 atomic_set_long #define atomic_set_acq_64 atomic_set_acq_long #define atomic_set_rel_64 atomic_set_rel_long #define atomic_clear_64 atomic_clear_long #define atomic_clear_acq_64 atomic_clear_acq_long #define atomic_clear_rel_64 atomic_clear_rel_long #define atomic_add_64 atomic_add_long #define atomic_add_acq_64 atomic_add_acq_long #define atomic_add_rel_64 atomic_add_rel_long #define atomic_subtract_64 atomic_subtract_long #define atomic_subtract_acq_64 atomic_subtract_acq_long #define atomic_subtract_rel_64 atomic_subtract_rel_long #define atomic_load_acq_64 atomic_load_acq_long #define atomic_store_rel_64 atomic_store_rel_long #define atomic_cmpset_64 atomic_cmpset_long #define atomic_cmpset_acq_64 atomic_cmpset_acq_long #define atomic_cmpset_rel_64 atomic_cmpset_rel_long +#define atomic_fcmpset_64 atomic_fcmpset_long +#define atomic_fcmpset_acq_64 atomic_fcmpset_acq_long +#define atomic_fcmpset_rel_64 atomic_fcmpset_rel_long #define atomic_swap_64 atomic_swap_long #define atomic_readandclear_64 atomic_readandclear_long #define atomic_fetchadd_64 atomic_fetchadd_long #define atomic_testandset_64 atomic_testandset_long #define atomic_testandclear_64 atomic_testandclear_long /* Operations on pointers. */ #define atomic_set_ptr atomic_set_long #define atomic_set_acq_ptr atomic_set_acq_long #define atomic_set_rel_ptr atomic_set_rel_long #define atomic_clear_ptr atomic_clear_long #define atomic_clear_acq_ptr atomic_clear_acq_long #define atomic_clear_rel_ptr atomic_clear_rel_long #define atomic_add_ptr atomic_add_long #define atomic_add_acq_ptr atomic_add_acq_long #define atomic_add_rel_ptr atomic_add_rel_long #define atomic_subtract_ptr atomic_subtract_long #define atomic_subtract_acq_ptr atomic_subtract_acq_long #define atomic_subtract_rel_ptr atomic_subtract_rel_long #define atomic_load_acq_ptr atomic_load_acq_long #define atomic_store_rel_ptr atomic_store_rel_long #define atomic_cmpset_ptr atomic_cmpset_long #define atomic_cmpset_acq_ptr atomic_cmpset_acq_long #define atomic_cmpset_rel_ptr atomic_cmpset_rel_long +#define atomic_fcmpset_ptr atomic_fcmpset_long +#define atomic_fcmpset_acq_ptr atomic_fcmpset_acq_long +#define atomic_fcmpset_rel_ptr atomic_fcmpset_rel_long #define atomic_swap_ptr atomic_swap_long #define atomic_readandclear_ptr atomic_readandclear_long #endif /* !WANT_FUNCTIONS */ #endif /* !_MACHINE_ATOMIC_H_ */ Index: stable/11/sys/arm/include/atomic-v4.h =================================================================== --- stable/11/sys/arm/include/atomic-v4.h (revision 315370) +++ stable/11/sys/arm/include/atomic-v4.h (revision 315371) @@ -1,484 +1,535 @@ /* $NetBSD: atomic.h,v 1.1 2002/10/19 12:22:34 bsh Exp $ */ /*- * Copyright (C) 2003-2004 Olivier Houchard * Copyright (C) 1994-1997 Mark Brinicombe * Copyright (C) 1994 Brini * All rights reserved. * * This code is derived from software written for Brini by Mark Brinicombe * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by Brini. * 4. The name of Brini may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY BRINI ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL BRINI BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_V4_H_ #define _MACHINE_ATOMIC_V4_H_ #ifndef _MACHINE_ATOMIC_H_ #error Do not include this file directly, use #endif #if __ARM_ARCH <= 5 #define isb() __asm __volatile("mcr p15, 0, %0, c7, c5, 4" : : "r" (0) : "memory") #define dsb() __asm __volatile("mcr p15, 0, %0, c7, c10, 4" : : "r" (0) : "memory") #define dmb() dsb() #else #error Only use this file with ARMv5 and earlier #endif #define mb() dmb() #define wmb() dmb() #define rmb() dmb() #define __with_interrupts_disabled(expr) \ do { \ u_int cpsr_save, tmp; \ \ __asm __volatile( \ "mrs %0, cpsr;" \ "orr %1, %0, %2;" \ "msr cpsr_fsxc, %1;" \ : "=r" (cpsr_save), "=r" (tmp) \ : "I" (PSR_I | PSR_F) \ : "cc" ); \ (expr); \ __asm __volatile( \ "msr cpsr_fsxc, %0" \ : /* no output */ \ : "r" (cpsr_save) \ : "cc" ); \ } while(0) static __inline uint32_t __swp(uint32_t val, volatile uint32_t *ptr) { __asm __volatile("swp %0, %2, [%3]" : "=&r" (val), "=m" (*ptr) : "r" (val), "r" (ptr), "m" (*ptr) : "memory"); return (val); } #ifdef _KERNEL #define ARM_HAVE_ATOMIC64 static __inline void atomic_add_32(volatile u_int32_t *p, u_int32_t val) { __with_interrupts_disabled(*p += val); } static __inline void atomic_add_64(volatile u_int64_t *p, u_int64_t val) { __with_interrupts_disabled(*p += val); } static __inline void atomic_clear_32(volatile uint32_t *address, uint32_t clearmask) { __with_interrupts_disabled(*address &= ~clearmask); } static __inline void atomic_clear_64(volatile uint64_t *address, uint64_t clearmask) { __with_interrupts_disabled(*address &= ~clearmask); } +static __inline int +atomic_fcmpset_32(volatile u_int32_t *p, volatile u_int32_t *cmpval, volatile u_int32_t newval) +{ + u_int32_t ret; + + __with_interrupts_disabled( + { + ret = *p; + if (*p == *cmpval) { + *p = newval; + ret = 1; + } else { + *cmpval = *p; + ret = 0; + } + }); + return (ret); +} + +static __inline int +atomic_fcmpset_64(volatile u_int64_t *p, volatile u_int64_t *cmpval, volatile u_int64_t newval) +{ + u_int64_t ret; + + __with_interrupts_disabled( + { + if (*p == *cmpval) { + *p = newval; + ret = 1; + } else { + *cmpval = *p; + ret = 0; + } + }); + return (ret); +} + static __inline u_int32_t atomic_cmpset_32(volatile u_int32_t *p, volatile u_int32_t cmpval, volatile u_int32_t newval) { int ret; __with_interrupts_disabled( { if (*p == cmpval) { *p = newval; ret = 1; } else { ret = 0; } }); return (ret); } static __inline u_int64_t atomic_cmpset_64(volatile u_int64_t *p, volatile u_int64_t cmpval, volatile u_int64_t newval) { int ret; __with_interrupts_disabled( { if (*p == cmpval) { *p = newval; ret = 1; } else { ret = 0; } }); return (ret); } static __inline uint32_t atomic_fetchadd_32(volatile uint32_t *p, uint32_t v) { uint32_t value; __with_interrupts_disabled( { value = *p; *p += v; }); return (value); } static __inline uint64_t atomic_fetchadd_64(volatile uint64_t *p, uint64_t v) { uint64_t value; __with_interrupts_disabled( { value = *p; *p += v; }); return (value); } static __inline uint64_t atomic_load_64(volatile uint64_t *p) { uint64_t value; __with_interrupts_disabled(value = *p); return (value); } static __inline void atomic_set_32(volatile uint32_t *address, uint32_t setmask) { __with_interrupts_disabled(*address |= setmask); } static __inline void atomic_set_64(volatile uint64_t *address, uint64_t setmask) { __with_interrupts_disabled(*address |= setmask); } static __inline void atomic_store_64(volatile uint64_t *p, uint64_t value) { __with_interrupts_disabled(*p = value); } static __inline void atomic_subtract_32(volatile u_int32_t *p, u_int32_t val) { __with_interrupts_disabled(*p -= val); } static __inline void atomic_subtract_64(volatile u_int64_t *p, u_int64_t val) { __with_interrupts_disabled(*p -= val); } #else /* !_KERNEL */ static __inline void atomic_add_32(volatile u_int32_t *p, u_int32_t val) { int start, ras_start = ARM_RAS_START; __asm __volatile("1:\n" "adr %1, 1b\n" "str %1, [%0]\n" "adr %1, 2f\n" "str %1, [%0, #4]\n" "ldr %1, [%2]\n" "add %1, %1, %3\n" "str %1, [%2]\n" "2:\n" "mov %1, #0\n" "str %1, [%0]\n" "mov %1, #0xffffffff\n" "str %1, [%0, #4]\n" : "+r" (ras_start), "=r" (start), "+r" (p), "+r" (val) : : "memory"); } static __inline void atomic_clear_32(volatile uint32_t *address, uint32_t clearmask) { int start, ras_start = ARM_RAS_START; __asm __volatile("1:\n" "adr %1, 1b\n" "str %1, [%0]\n" "adr %1, 2f\n" "str %1, [%0, #4]\n" "ldr %1, [%2]\n" "bic %1, %1, %3\n" "str %1, [%2]\n" "2:\n" "mov %1, #0\n" "str %1, [%0]\n" "mov %1, #0xffffffff\n" "str %1, [%0, #4]\n" : "+r" (ras_start), "=r" (start), "+r" (address), "+r" (clearmask) : : "memory"); } static __inline u_int32_t atomic_cmpset_32(volatile u_int32_t *p, volatile u_int32_t cmpval, volatile u_int32_t newval) { register int done, ras_start = ARM_RAS_START; __asm __volatile("1:\n" "adr %1, 1b\n" "str %1, [%0]\n" "adr %1, 2f\n" "str %1, [%0, #4]\n" "ldr %1, [%2]\n" "cmp %1, %3\n" "streq %4, [%2]\n" "2:\n" "mov %1, #0\n" "str %1, [%0]\n" "mov %1, #0xffffffff\n" "str %1, [%0, #4]\n" "moveq %1, #1\n" "movne %1, #0\n" : "+r" (ras_start), "=r" (done) ,"+r" (p), "+r" (cmpval), "+r" (newval) : : "cc", "memory"); return (done); } static __inline uint32_t atomic_fetchadd_32(volatile uint32_t *p, uint32_t v) { uint32_t start, tmp, ras_start = ARM_RAS_START; __asm __volatile("1:\n" "adr %1, 1b\n" "str %1, [%0]\n" "adr %1, 2f\n" "str %1, [%0, #4]\n" "ldr %1, [%3]\n" "mov %2, %1\n" "add %2, %2, %4\n" "str %2, [%3]\n" "2:\n" "mov %2, #0\n" "str %2, [%0]\n" "mov %2, #0xffffffff\n" "str %2, [%0, #4]\n" : "+r" (ras_start), "=r" (start), "=r" (tmp), "+r" (p), "+r" (v) : : "memory"); return (start); } static __inline void atomic_set_32(volatile uint32_t *address, uint32_t setmask) { int start, ras_start = ARM_RAS_START; __asm __volatile("1:\n" "adr %1, 1b\n" "str %1, [%0]\n" "adr %1, 2f\n" "str %1, [%0, #4]\n" "ldr %1, [%2]\n" "orr %1, %1, %3\n" "str %1, [%2]\n" "2:\n" "mov %1, #0\n" "str %1, [%0]\n" "mov %1, #0xffffffff\n" "str %1, [%0, #4]\n" : "+r" (ras_start), "=r" (start), "+r" (address), "+r" (setmask) : : "memory"); } static __inline void atomic_subtract_32(volatile u_int32_t *p, u_int32_t val) { int start, ras_start = ARM_RAS_START; __asm __volatile("1:\n" "adr %1, 1b\n" "str %1, [%0]\n" "adr %1, 2f\n" "str %1, [%0, #4]\n" "ldr %1, [%2]\n" "sub %1, %1, %3\n" "str %1, [%2]\n" "2:\n" "mov %1, #0\n" "str %1, [%0]\n" "mov %1, #0xffffffff\n" "str %1, [%0, #4]\n" : "+r" (ras_start), "=r" (start), "+r" (p), "+r" (val) : : "memory"); } #endif /* _KERNEL */ static __inline uint32_t atomic_readandclear_32(volatile u_int32_t *p) { return (__swp(0, p)); } static __inline uint32_t atomic_swap_32(volatile u_int32_t *p, u_int32_t v) { return (__swp(v, p)); } +#define atomic_fcmpset_rel_32 atomic_fcmpset_32 +#define atomic_fcmpset_acq_32 atomic_fcmpset_32 +#define atomic_fcmpset_rel_64 atomic_fcmpset_64 +#define atomic_fcmpset_acq_64 atomic_fcmpset_64 +#define atomic_fcmpset_acq_long atomic_fcmpset_long +#define atomic_fcmpset_rel_long atomic_fcmpset_long #define atomic_cmpset_rel_32 atomic_cmpset_32 #define atomic_cmpset_acq_32 atomic_cmpset_32 #define atomic_cmpset_rel_64 atomic_cmpset_64 #define atomic_cmpset_acq_64 atomic_cmpset_64 #define atomic_set_rel_32 atomic_set_32 #define atomic_set_acq_32 atomic_set_32 #define atomic_clear_rel_32 atomic_clear_32 #define atomic_clear_acq_32 atomic_clear_32 #define atomic_add_rel_32 atomic_add_32 #define atomic_add_acq_32 atomic_add_32 #define atomic_subtract_rel_32 atomic_subtract_32 #define atomic_subtract_acq_32 atomic_subtract_32 #define atomic_store_rel_32 atomic_store_32 #define atomic_store_rel_long atomic_store_long #define atomic_load_acq_32 atomic_load_32 #define atomic_load_acq_long atomic_load_long #define atomic_add_acq_long atomic_add_long #define atomic_add_rel_long atomic_add_long #define atomic_subtract_acq_long atomic_subtract_long #define atomic_subtract_rel_long atomic_subtract_long #define atomic_clear_acq_long atomic_clear_long #define atomic_clear_rel_long atomic_clear_long #define atomic_set_acq_long atomic_set_long #define atomic_set_rel_long atomic_set_long #define atomic_cmpset_acq_long atomic_cmpset_long #define atomic_cmpset_rel_long atomic_cmpset_long #define atomic_load_acq_long atomic_load_long #undef __with_interrupts_disabled static __inline void atomic_add_long(volatile u_long *p, u_long v) { atomic_add_32((volatile uint32_t *)p, v); } static __inline void atomic_clear_long(volatile u_long *p, u_long v) { atomic_clear_32((volatile uint32_t *)p, v); } static __inline int atomic_cmpset_long(volatile u_long *dst, u_long old, u_long newe) { return (atomic_cmpset_32((volatile uint32_t *)dst, old, newe)); +} + +static __inline u_long +atomic_fcmpset_long(volatile u_long *dst, u_long *old, u_long newe) +{ + + return (atomic_fcmpset_32((volatile uint32_t *)dst, + (uint32_t *)old, newe)); } static __inline u_long atomic_fetchadd_long(volatile u_long *p, u_long v) { return (atomic_fetchadd_32((volatile uint32_t *)p, v)); } static __inline void atomic_readandclear_long(volatile u_long *p) { atomic_readandclear_32((volatile uint32_t *)p); } static __inline void atomic_set_long(volatile u_long *p, u_long v) { atomic_set_32((volatile uint32_t *)p, v); } static __inline void atomic_subtract_long(volatile u_long *p, u_long v) { atomic_subtract_32((volatile uint32_t *)p, v); } /* * ARMv5 does not support SMP. For both kernel and user modes, only a * compiler barrier is needed for fences, since CPU is always * self-consistent. */ static __inline void atomic_thread_fence_acq(void) { __compiler_membar(); } static __inline void atomic_thread_fence_rel(void) { __compiler_membar(); } static __inline void atomic_thread_fence_acq_rel(void) { __compiler_membar(); } static __inline void atomic_thread_fence_seq_cst(void) { __compiler_membar(); } #endif /* _MACHINE_ATOMIC_H_ */ Index: stable/11/sys/arm/include/atomic-v6.h =================================================================== --- stable/11/sys/arm/include/atomic-v6.h (revision 315370) +++ stable/11/sys/arm/include/atomic-v6.h (revision 315371) @@ -1,714 +1,824 @@ /* $NetBSD: atomic.h,v 1.1 2002/10/19 12:22:34 bsh Exp $ */ /*- * Copyright (C) 2003-2004 Olivier Houchard * Copyright (C) 1994-1997 Mark Brinicombe * Copyright (C) 1994 Brini * All rights reserved. * * This code is derived from software written for Brini by Mark Brinicombe * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by Brini. * 4. The name of Brini may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY BRINI ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL BRINI BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_V6_H_ #define _MACHINE_ATOMIC_V6_H_ #ifndef _MACHINE_ATOMIC_H_ #error Do not include this file directly, use #endif #if __ARM_ARCH >= 7 #define isb() __asm __volatile("isb" : : : "memory") #define dsb() __asm __volatile("dsb" : : : "memory") #define dmb() __asm __volatile("dmb" : : : "memory") #elif __ARM_ARCH >= 6 #define isb() __asm __volatile("mcr p15, 0, %0, c7, c5, 4" : : "r" (0) : "memory") #define dsb() __asm __volatile("mcr p15, 0, %0, c7, c10, 4" : : "r" (0) : "memory") #define dmb() __asm __volatile("mcr p15, 0, %0, c7, c10, 5" : : "r" (0) : "memory") #else #error Only use this file with ARMv6 and later #endif #define mb() dmb() #define wmb() dmb() #define rmb() dmb() #define ARM_HAVE_ATOMIC64 #define ATOMIC_ACQ_REL_LONG(NAME) \ static __inline void \ atomic_##NAME##_acq_long(__volatile u_long *p, u_long v) \ { \ atomic_##NAME##_long(p, v); \ dmb(); \ } \ \ static __inline void \ atomic_##NAME##_rel_long(__volatile u_long *p, u_long v) \ { \ dmb(); \ atomic_##NAME##_long(p, v); \ } #define ATOMIC_ACQ_REL(NAME, WIDTH) \ static __inline void \ atomic_##NAME##_acq_##WIDTH(__volatile uint##WIDTH##_t *p, uint##WIDTH##_t v)\ { \ atomic_##NAME##_##WIDTH(p, v); \ dmb(); \ } \ \ static __inline void \ atomic_##NAME##_rel_##WIDTH(__volatile uint##WIDTH##_t *p, uint##WIDTH##_t v)\ { \ dmb(); \ atomic_##NAME##_##WIDTH(p, v); \ } static __inline void atomic_add_32(volatile uint32_t *p, uint32_t val) { uint32_t tmp = 0, tmp2 = 0; __asm __volatile( "1: ldrex %0, [%2] \n" " add %0, %0, %3 \n" " strex %1, %0, [%2] \n" " cmp %1, #0 \n" " it ne \n" " bne 1b \n" : "=&r" (tmp), "+r" (tmp2) ,"+r" (p), "+r" (val) : : "cc", "memory"); } static __inline void atomic_add_64(volatile uint64_t *p, uint64_t val) { uint64_t tmp; uint32_t exflag; __asm __volatile( "1: \n" " ldrexd %Q[tmp], %R[tmp], [%[ptr]] \n" " adds %Q[tmp], %Q[val] \n" " adc %R[tmp], %R[tmp], %R[val] \n" " strexd %[exf], %Q[tmp], %R[tmp], [%[ptr]] \n" " teq %[exf], #0 \n" " it ne \n" " bne 1b \n" : [exf] "=&r" (exflag), [tmp] "=&r" (tmp) : [ptr] "r" (p), [val] "r" (val) : "cc", "memory"); } static __inline void atomic_add_long(volatile u_long *p, u_long val) { atomic_add_32((volatile uint32_t *)p, val); } ATOMIC_ACQ_REL(add, 32) ATOMIC_ACQ_REL(add, 64) ATOMIC_ACQ_REL_LONG(add) static __inline void atomic_clear_32(volatile uint32_t *address, uint32_t setmask) { uint32_t tmp = 0, tmp2 = 0; __asm __volatile( "1: ldrex %0, [%2] \n" " bic %0, %0, %3 \n" " strex %1, %0, [%2] \n" " cmp %1, #0 \n" " it ne \n" " bne 1b \n" : "=&r" (tmp), "+r" (tmp2), "+r" (address), "+r" (setmask) : : "cc", "memory"); } static __inline void atomic_clear_64(volatile uint64_t *p, uint64_t val) { uint64_t tmp; uint32_t exflag; __asm __volatile( "1: \n" " ldrexd %Q[tmp], %R[tmp], [%[ptr]] \n" " bic %Q[tmp], %Q[val] \n" " bic %R[tmp], %R[val] \n" " strexd %[exf], %Q[tmp], %R[tmp], [%[ptr]] \n" " teq %[exf], #0 \n" " it ne \n" " bne 1b \n" : [exf] "=&r" (exflag), [tmp] "=&r" (tmp) : [ptr] "r" (p), [val] "r" (val) : "cc", "memory"); } static __inline void atomic_clear_long(volatile u_long *address, u_long setmask) { atomic_clear_32((volatile uint32_t *)address, setmask); } ATOMIC_ACQ_REL(clear, 32) ATOMIC_ACQ_REL(clear, 64) ATOMIC_ACQ_REL_LONG(clear) +static __inline int +atomic_fcmpset_32(volatile uint32_t *p, uint32_t *cmpval, uint32_t newval) +{ + uint32_t tmp; + uint32_t _cmpval = *cmpval; + int ret; + + __asm __volatile( + "1: mov %0, #1 \n" + " ldrex %1, [%2] \n" + " cmp %1, %3 \n" + " it ne \n" + " bne 2f \n" + " strex %0, %4, [%2] \n" + "2:" + : "=&r" (ret), "=&r" (tmp), "+r" (p), "+r" (_cmpval), "+r" (newval) + : : "cc", "memory"); + *cmpval = tmp; + return (!ret); +} + +static __inline uint64_t +atomic_fcmpset_64(volatile uint64_t *p, uint64_t *cmpval, uint64_t newval) +{ + uint64_t tmp; + uint64_t _cmpval = *cmpval; + int ret; + + __asm __volatile( + "1: mov %[ret], #1 \n" + " ldrexd %Q[tmp], %R[tmp], [%[ptr]] \n" + " teq %Q[tmp], %Q[_cmpval] \n" + " itee eq \n" + " teqeq %R[tmp], %R[_cmpval] \n" + " bne 2f \n" + " strexd %[ret], %Q[newval], %R[newval], [%[ptr]]\n" + "2: \n" + : [ret] "=&r" (ret), + [tmp] "=&r" (tmp) + : [ptr] "r" (p), + [_cmpval] "r" (_cmpval), + [newval] "r" (newval) + : "cc", "memory"); + *cmpval = tmp; + return (!ret); +} + +static __inline u_long +atomic_fcmpset_long(volatile u_long *p, u_long *cmpval, u_long newval) +{ + + return (atomic_fcmpset_32((volatile uint32_t *)p, + (uint32_t *)cmpval, newval)); +} + +static __inline uint64_t +atomic_fcmpset_acq_64(volatile uint64_t *p, uint64_t *cmpval, uint64_t newval) +{ + uint64_t ret; + + ret = atomic_fcmpset_64(p, cmpval, newval); + dmb(); + return (ret); +} + +static __inline u_long +atomic_fcmpset_acq_long(volatile u_long *p, u_long *cmpval, u_long newval) +{ + u_long ret; + + ret = atomic_fcmpset_long(p, cmpval, newval); + dmb(); + return (ret); +} + +static __inline uint32_t +atomic_fcmpset_acq_32(volatile uint32_t *p, uint32_t *cmpval, uint32_t newval) +{ + + uint32_t ret; + + ret = atomic_fcmpset_32(p, cmpval, newval); + dmb(); + return (ret); +} + +static __inline uint32_t +atomic_fcmpset_rel_32(volatile uint32_t *p, uint32_t *cmpval, uint32_t newval) +{ + + dmb(); + return (atomic_fcmpset_32(p, cmpval, newval)); +} + +static __inline uint64_t +atomic_fcmpset_rel_64(volatile uint64_t *p, uint64_t *cmpval, uint64_t newval) +{ + + dmb(); + return (atomic_fcmpset_64(p, cmpval, newval)); +} + +static __inline u_long +atomic_fcmpset_rel_long(volatile u_long *p, u_long *cmpval, u_long newval) +{ + + dmb(); + return (atomic_fcmpset_long(p, cmpval, newval)); +} + static __inline uint32_t atomic_cmpset_32(volatile uint32_t *p, uint32_t cmpval, uint32_t newval) { uint32_t ret; __asm __volatile( "1: ldrex %0, [%1] \n" " cmp %0, %2 \n" " itt ne \n" " movne %0, #0 \n" " bne 2f \n" " strex %0, %3, [%1] \n" " cmp %0, #0 \n" " ite eq \n" " moveq %0, #1 \n" " bne 1b \n" "2:" : "=&r" (ret), "+r" (p), "+r" (cmpval), "+r" (newval) : : "cc", "memory"); return (ret); } static __inline int atomic_cmpset_64(volatile uint64_t *p, uint64_t cmpval, uint64_t newval) { uint64_t tmp; uint32_t ret; __asm __volatile( "1: \n" " ldrexd %Q[tmp], %R[tmp], [%[ptr]] \n" " teq %Q[tmp], %Q[cmpval] \n" " itee eq \n" " teqeq %R[tmp], %R[cmpval] \n" " movne %[ret], #0 \n" " bne 2f \n" " strexd %[ret], %Q[newval], %R[newval], [%[ptr]]\n" " teq %[ret], #0 \n" " it ne \n" " bne 1b \n" " mov %[ret], #1 \n" "2: \n" : [ret] "=&r" (ret), [tmp] "=&r" (tmp) : [ptr] "r" (p), [cmpval] "r" (cmpval), [newval] "r" (newval) : "cc", "memory"); return (ret); } static __inline u_long atomic_cmpset_long(volatile u_long *p, u_long cmpval, u_long newval) { return (atomic_cmpset_32((volatile uint32_t *)p, cmpval, newval)); } static __inline uint32_t atomic_cmpset_acq_32(volatile uint32_t *p, uint32_t cmpval, uint32_t newval) { uint32_t ret; ret = atomic_cmpset_32(p, cmpval, newval); dmb(); return (ret); } static __inline uint64_t atomic_cmpset_acq_64(volatile uint64_t *p, uint64_t cmpval, uint64_t newval) { uint64_t ret; ret = atomic_cmpset_64(p, cmpval, newval); dmb(); return (ret); } static __inline u_long atomic_cmpset_acq_long(volatile u_long *p, u_long cmpval, u_long newval) { u_long ret; ret = atomic_cmpset_long(p, cmpval, newval); dmb(); return (ret); } static __inline uint32_t atomic_cmpset_rel_32(volatile uint32_t *p, uint32_t cmpval, uint32_t newval) { dmb(); return (atomic_cmpset_32(p, cmpval, newval)); } static __inline uint64_t atomic_cmpset_rel_64(volatile uint64_t *p, uint64_t cmpval, uint64_t newval) { dmb(); return (atomic_cmpset_64(p, cmpval, newval)); } static __inline u_long atomic_cmpset_rel_long(volatile u_long *p, u_long cmpval, u_long newval) { dmb(); return (atomic_cmpset_long(p, cmpval, newval)); } static __inline uint32_t atomic_fetchadd_32(volatile uint32_t *p, uint32_t val) { uint32_t tmp = 0, tmp2 = 0, ret = 0; __asm __volatile( "1: ldrex %0, [%3] \n" " add %1, %0, %4 \n" " strex %2, %1, [%3] \n" " cmp %2, #0 \n" " it ne \n" " bne 1b \n" : "+r" (ret), "=&r" (tmp), "+r" (tmp2), "+r" (p), "+r" (val) : : "cc", "memory"); return (ret); } static __inline uint64_t atomic_fetchadd_64(volatile uint64_t *p, uint64_t val) { uint64_t ret, tmp; uint32_t exflag; __asm __volatile( "1: \n" " ldrexd %Q[tmp], %R[tmp], [%[ptr]] \n" " adds %Q[tmp], %Q[ret], %Q[val] \n" " adc %R[tmp], %R[ret], %R[val] \n" " strexd %[exf], %Q[tmp], %R[tmp], [%[ptr]] \n" " teq %[exf], #0 \n" " it ne \n" " bne 1b \n" : [ret] "=&r" (ret), [exf] "=&r" (exflag), [tmp] "=&r" (tmp) : [ptr] "r" (p), [val] "r" (val) : "cc", "memory"); return (ret); } static __inline u_long atomic_fetchadd_long(volatile u_long *p, u_long val) { return (atomic_fetchadd_32((volatile uint32_t *)p, val)); } static __inline uint32_t atomic_load_acq_32(volatile uint32_t *p) { uint32_t v; v = *p; dmb(); return (v); } static __inline uint64_t atomic_load_64(volatile uint64_t *p) { uint64_t ret; /* * The only way to atomically load 64 bits is with LDREXD which puts the * exclusive monitor into the exclusive state, so reset it to open state * with CLREX because we don't actually need to store anything. */ __asm __volatile( "ldrexd %Q[ret], %R[ret], [%[ptr]] \n" "clrex \n" : [ret] "=&r" (ret) : [ptr] "r" (p) : "cc", "memory"); return (ret); } static __inline uint64_t atomic_load_acq_64(volatile uint64_t *p) { uint64_t ret; ret = atomic_load_64(p); dmb(); return (ret); } static __inline u_long atomic_load_acq_long(volatile u_long *p) { u_long v; v = *p; dmb(); return (v); } static __inline uint32_t atomic_readandclear_32(volatile uint32_t *p) { uint32_t ret, tmp = 0, tmp2 = 0; __asm __volatile( "1: ldrex %0, [%3] \n" " mov %1, #0 \n" " strex %2, %1, [%3] \n" " cmp %2, #0 \n" " it ne \n" " bne 1b \n" : "=r" (ret), "=&r" (tmp), "+r" (tmp2), "+r" (p) : : "cc", "memory"); return (ret); } static __inline uint64_t atomic_readandclear_64(volatile uint64_t *p) { uint64_t ret, tmp; uint32_t exflag; __asm __volatile( "1: \n" " ldrexd %Q[ret], %R[ret], [%[ptr]] \n" " mov %Q[tmp], #0 \n" " mov %R[tmp], #0 \n" " strexd %[exf], %Q[tmp], %R[tmp], [%[ptr]] \n" " teq %[exf], #0 \n" " it ne \n" " bne 1b \n" : [ret] "=&r" (ret), [exf] "=&r" (exflag), [tmp] "=&r" (tmp) : [ptr] "r" (p) : "cc", "memory"); return (ret); } static __inline u_long atomic_readandclear_long(volatile u_long *p) { return (atomic_readandclear_32((volatile uint32_t *)p)); } static __inline void atomic_set_32(volatile uint32_t *address, uint32_t setmask) { uint32_t tmp = 0, tmp2 = 0; __asm __volatile( "1: ldrex %0, [%2] \n" " orr %0, %0, %3 \n" " strex %1, %0, [%2] \n" " cmp %1, #0 \n" " it ne \n" " bne 1b \n" : "=&r" (tmp), "+r" (tmp2), "+r" (address), "+r" (setmask) : : "cc", "memory"); } static __inline void atomic_set_64(volatile uint64_t *p, uint64_t val) { uint64_t tmp; uint32_t exflag; __asm __volatile( "1: \n" " ldrexd %Q[tmp], %R[tmp], [%[ptr]] \n" " orr %Q[tmp], %Q[val] \n" " orr %R[tmp], %R[val] \n" " strexd %[exf], %Q[tmp], %R[tmp], [%[ptr]] \n" " teq %[exf], #0 \n" " it ne \n" " bne 1b \n" : [exf] "=&r" (exflag), [tmp] "=&r" (tmp) : [ptr] "r" (p), [val] "r" (val) : "cc", "memory"); } static __inline void atomic_set_long(volatile u_long *address, u_long setmask) { atomic_set_32((volatile uint32_t *)address, setmask); } ATOMIC_ACQ_REL(set, 32) ATOMIC_ACQ_REL(set, 64) ATOMIC_ACQ_REL_LONG(set) static __inline void atomic_subtract_32(volatile uint32_t *p, uint32_t val) { uint32_t tmp = 0, tmp2 = 0; __asm __volatile( "1: ldrex %0, [%2] \n" " sub %0, %0, %3 \n" " strex %1, %0, [%2] \n" " cmp %1, #0 \n" " it ne \n" " bne 1b \n" : "=&r" (tmp), "+r" (tmp2), "+r" (p), "+r" (val) : : "cc", "memory"); } static __inline void atomic_subtract_64(volatile uint64_t *p, uint64_t val) { uint64_t tmp; uint32_t exflag; __asm __volatile( "1: \n" " ldrexd %Q[tmp], %R[tmp], [%[ptr]] \n" " subs %Q[tmp], %Q[val] \n" " sbc %R[tmp], %R[tmp], %R[val] \n" " strexd %[exf], %Q[tmp], %R[tmp], [%[ptr]] \n" " teq %[exf], #0 \n" " it ne \n" " bne 1b \n" : [exf] "=&r" (exflag), [tmp] "=&r" (tmp) : [ptr] "r" (p), [val] "r" (val) : "cc", "memory"); } static __inline void atomic_subtract_long(volatile u_long *p, u_long val) { atomic_subtract_32((volatile uint32_t *)p, val); } ATOMIC_ACQ_REL(subtract, 32) ATOMIC_ACQ_REL(subtract, 64) ATOMIC_ACQ_REL_LONG(subtract) static __inline void atomic_store_64(volatile uint64_t *p, uint64_t val) { uint64_t tmp; uint32_t exflag; /* * The only way to atomically store 64 bits is with STREXD, which will * succeed only if paired up with a preceeding LDREXD using the same * address, so we read and discard the existing value before storing. */ __asm __volatile( "1: \n" " ldrexd %Q[tmp], %R[tmp], [%[ptr]] \n" " strexd %[exf], %Q[val], %R[val], [%[ptr]] \n" " teq %[exf], #0 \n" " it ne \n" " bne 1b \n" : [tmp] "=&r" (tmp), [exf] "=&r" (exflag) : [ptr] "r" (p), [val] "r" (val) : "cc", "memory"); } static __inline void atomic_store_rel_32(volatile uint32_t *p, uint32_t v) { dmb(); *p = v; } static __inline void atomic_store_rel_64(volatile uint64_t *p, uint64_t val) { dmb(); atomic_store_64(p, val); } static __inline void atomic_store_rel_long(volatile u_long *p, u_long v) { dmb(); *p = v; } static __inline int atomic_testandset_32(volatile uint32_t *p, u_int v) { uint32_t tmp, tmp2, res, mask; mask = 1u << (v & 0x1f); tmp = tmp2 = 0; __asm __volatile( "1: ldrex %0, [%4] \n" " orr %1, %0, %3 \n" " strex %2, %1, [%4] \n" " cmp %2, #0 \n" " it ne \n" " bne 1b \n" : "=&r" (res), "=&r" (tmp), "=&r" (tmp2) : "r" (mask), "r" (p) : "cc", "memory"); return ((res & mask) != 0); } static __inline int atomic_testandset_int(volatile u_int *p, u_int v) { return (atomic_testandset_32((volatile uint32_t *)p, v)); } static __inline int atomic_testandset_long(volatile u_long *p, u_int v) { return (atomic_testandset_32((volatile uint32_t *)p, v)); } static __inline int atomic_testandset_64(volatile uint64_t *p, u_int v) { volatile uint32_t *p32; p32 = (volatile uint32_t *)p; /* Assume little-endian */ if (v >= 32) { v &= 0x1f; p32++; } return (atomic_testandset_32(p32, v)); } static __inline uint32_t atomic_swap_32(volatile uint32_t *p, uint32_t v) { uint32_t ret, exflag; __asm __volatile( "1: ldrex %[ret], [%[ptr]] \n" " strex %[exf], %[val], [%[ptr]] \n" " teq %[exf], #0 \n" " it ne \n" " bne 1b \n" : [ret] "=&r" (ret), [exf] "=&r" (exflag) : [val] "r" (v), [ptr] "r" (p) : "cc", "memory"); return (ret); } static __inline uint64_t atomic_swap_64(volatile uint64_t *p, uint64_t v) { uint64_t ret; uint32_t exflag; __asm __volatile( "1: ldrexd %Q[ret], %R[ret], [%[ptr]] \n" " strexd %[exf], %Q[val], %R[val], [%[ptr]] \n" " teq %[exf], #0 \n" " it ne \n" " bne 1b \n" : [ret] "=&r" (ret), [exf] "=&r" (exflag) : [val] "r" (v), [ptr] "r" (p) : "cc", "memory"); return (ret); } #undef ATOMIC_ACQ_REL #undef ATOMIC_ACQ_REL_LONG static __inline void atomic_thread_fence_acq(void) { dmb(); } static __inline void atomic_thread_fence_rel(void) { dmb(); } static __inline void atomic_thread_fence_acq_rel(void) { dmb(); } static __inline void atomic_thread_fence_seq_cst(void) { dmb(); } #endif /* _MACHINE_ATOMIC_V6_H_ */ Index: stable/11/sys/arm/include/atomic.h =================================================================== --- stable/11/sys/arm/include/atomic.h (revision 315370) +++ stable/11/sys/arm/include/atomic.h (revision 315371) @@ -1,109 +1,115 @@ /* $NetBSD: atomic.h,v 1.1 2002/10/19 12:22:34 bsh Exp $ */ /*- * Copyright (C) 2003-2004 Olivier Houchard * Copyright (C) 1994-1997 Mark Brinicombe * Copyright (C) 1994 Brini * All rights reserved. * * This code is derived from software written for Brini by Mark Brinicombe * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by Brini. * 4. The name of Brini may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY BRINI ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL BRINI BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_H_ #define _MACHINE_ATOMIC_H_ #include #ifndef _KERNEL #include #endif #if __ARM_ARCH >= 6 #include #else /* < armv6 */ #include #endif /* Arch >= v6 */ static __inline int atomic_load_32(volatile uint32_t *v) { return (*v); } static __inline void atomic_store_32(volatile uint32_t *dst, uint32_t src) { *dst = src; } static __inline int atomic_load_long(volatile u_long *v) { return (*v); } static __inline void atomic_store_long(volatile u_long *dst, u_long src) { *dst = src; } #define atomic_clear_ptr atomic_clear_32 #define atomic_set_ptr atomic_set_32 +#define atomic_fcmpset_ptr atomic_fcmpset_32 +#define atomic_fcmpset_rel_ptr atomic_fcmpset_rel_32 +#define atomic_fcmpset_acq_ptr atomic_fcmpset_acq_32 #define atomic_cmpset_ptr atomic_cmpset_32 #define atomic_cmpset_rel_ptr atomic_cmpset_rel_32 #define atomic_cmpset_acq_ptr atomic_cmpset_acq_32 #define atomic_store_ptr atomic_store_32 #define atomic_store_rel_ptr atomic_store_rel_32 #define atomic_add_int atomic_add_32 #define atomic_add_acq_int atomic_add_acq_32 #define atomic_add_rel_int atomic_add_rel_32 #define atomic_subtract_int atomic_subtract_32 #define atomic_subtract_acq_int atomic_subtract_acq_32 #define atomic_subtract_rel_int atomic_subtract_rel_32 #define atomic_clear_int atomic_clear_32 #define atomic_clear_acq_int atomic_clear_acq_32 #define atomic_clear_rel_int atomic_clear_rel_32 #define atomic_set_int atomic_set_32 #define atomic_set_acq_int atomic_set_acq_32 #define atomic_set_rel_int atomic_set_rel_32 +#define atomic_fcmpset_int atomic_fcmpset_32 +#define atomic_fcmpset_acq_int atomic_fcmpset_acq_32 +#define atomic_fcmpset_rel_int atomic_fcmpset_rel_32 #define atomic_cmpset_int atomic_cmpset_32 #define atomic_cmpset_acq_int atomic_cmpset_acq_32 #define atomic_cmpset_rel_int atomic_cmpset_rel_32 #define atomic_fetchadd_int atomic_fetchadd_32 #define atomic_readandclear_int atomic_readandclear_32 #define atomic_load_acq_int atomic_load_acq_32 #define atomic_store_rel_int atomic_store_rel_32 #define atomic_swap_int atomic_swap_32 #endif /* _MACHINE_ATOMIC_H_ */ Index: stable/11/sys/arm64/include/atomic.h =================================================================== --- stable/11/sys/arm64/include/atomic.h (revision 315370) +++ stable/11/sys/arm64/include/atomic.h (revision 315371) @@ -1,411 +1,475 @@ /*- * Copyright (c) 2013 Andrew Turner * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_H_ #define _MACHINE_ATOMIC_H_ #define isb() __asm __volatile("isb" : : : "memory") /* * Options for DMB and DSB: * oshld Outer Shareable, load * oshst Outer Shareable, store * osh Outer Shareable, all * nshld Non-shareable, load * nshst Non-shareable, store * nsh Non-shareable, all * ishld Inner Shareable, load * ishst Inner Shareable, store * ish Inner Shareable, all * ld Full system, load * st Full system, store * sy Full system, all */ #define dsb(opt) __asm __volatile("dsb " __STRING(opt) : : : "memory") #define dmb(opt) __asm __volatile("dmb " __STRING(opt) : : : "memory") #define mb() dmb(sy) /* Full system memory barrier all */ #define wmb() dmb(st) /* Full system memory barrier store */ #define rmb() dmb(ld) /* Full system memory barrier load */ #define ATOMIC_OP(op, asm_op, bar, a, l) \ static __inline void \ atomic_##op##_##bar##32(volatile uint32_t *p, uint32_t val) \ { \ uint32_t tmp; \ int res; \ \ __asm __volatile( \ "1: ld"#a"xr %w0, [%2] \n" \ " "#asm_op" %w0, %w0, %w3 \n" \ " st"#l"xr %w1, %w0, [%2] \n" \ " cbnz %w1, 1b \n" \ : "=&r"(tmp), "=&r"(res) \ : "r" (p), "r" (val) \ : "memory" \ ); \ } \ \ static __inline void \ atomic_##op##_##bar##64(volatile uint64_t *p, uint64_t val) \ { \ uint64_t tmp; \ int res; \ \ __asm __volatile( \ "1: ld"#a"xr %0, [%2] \n" \ " "#asm_op" %0, %0, %3 \n" \ " st"#l"xr %w1, %0, [%2] \n" \ " cbnz %w1, 1b \n" \ : "=&r"(tmp), "=&r"(res) \ : "r" (p), "r" (val) \ : "memory" \ ); \ } #define ATOMIC(op, asm_op) \ ATOMIC_OP(op, asm_op, , , ) \ ATOMIC_OP(op, asm_op, acq_, a, ) \ ATOMIC_OP(op, asm_op, rel_, , l) \ ATOMIC(add, add) ATOMIC(clear, bic) ATOMIC(set, orr) ATOMIC(subtract, sub) +#define ATOMIC_FCMPSET(bar, a, l) \ +static __inline int \ +atomic_fcmpset_##bar##32(volatile uint32_t *p, uint32_t *cmpval, \ + uint32_t newval) \ +{ \ + uint32_t tmp; \ + uint32_t _cmpval = *cmpval; \ + int res; \ + \ + __asm __volatile( \ + "1: mov %w1, #1 \n" \ + " ld"#a"xr %w0, [%2] \n" \ + " cmp %w0, %w3 \n" \ + " b.ne 2f \n" \ + " st"#l"xr %w1, %w4, [%2] \n" \ + "2:" \ + : "=&r"(tmp), "=&r"(res) \ + : "r" (p), "r" (_cmpval), "r" (newval) \ + : "cc", "memory" \ + ); \ + *cmpval = tmp; \ + \ + return (!res); \ +} \ + \ +static __inline int \ +atomic_fcmpset_##bar##64(volatile uint64_t *p, uint64_t *cmpval, \ + uint64_t newval) \ +{ \ + uint64_t tmp; \ + uint64_t _cmpval = *cmpval; \ + int res; \ + \ + __asm __volatile( \ + "1: mov %w1, #1 \n" \ + " ld"#a"xr %0, [%2] \n" \ + " cmp %0, %3 \n" \ + " b.ne 2f \n" \ + " st"#l"xr %w1, %4, [%2] \n" \ + "2:" \ + : "=&r"(tmp), "=&r"(res) \ + : "r" (p), "r" (_cmpval), "r" (newval) \ + : "cc", "memory" \ + ); \ + *cmpval = tmp; \ + \ + return (!res); \ +} + +ATOMIC_FCMPSET( , , ) +ATOMIC_FCMPSET(acq_, a, ) +ATOMIC_FCMPSET(rel_, ,l) + +#undef ATOMIC_FCMPSET + #define ATOMIC_CMPSET(bar, a, l) \ static __inline int \ atomic_cmpset_##bar##32(volatile uint32_t *p, uint32_t cmpval, \ uint32_t newval) \ { \ uint32_t tmp; \ int res; \ \ __asm __volatile( \ "1: mov %w1, #1 \n" \ " ld"#a"xr %w0, [%2] \n" \ " cmp %w0, %w3 \n" \ " b.ne 2f \n" \ " st"#l"xr %w1, %w4, [%2] \n" \ " cbnz %w1, 1b \n" \ "2:" \ : "=&r"(tmp), "=&r"(res) \ : "r" (p), "r" (cmpval), "r" (newval) \ : "cc", "memory" \ ); \ \ return (!res); \ } \ \ static __inline int \ atomic_cmpset_##bar##64(volatile uint64_t *p, uint64_t cmpval, \ uint64_t newval) \ { \ uint64_t tmp; \ int res; \ \ __asm __volatile( \ "1: mov %w1, #1 \n" \ " ld"#a"xr %0, [%2] \n" \ " cmp %0, %3 \n" \ " b.ne 2f \n" \ " st"#l"xr %w1, %4, [%2] \n" \ " cbnz %w1, 1b \n" \ "2:" \ : "=&r"(tmp), "=&r"(res) \ : "r" (p), "r" (cmpval), "r" (newval) \ : "cc", "memory" \ ); \ \ return (!res); \ } ATOMIC_CMPSET( , , ) ATOMIC_CMPSET(acq_, a, ) ATOMIC_CMPSET(rel_, ,l) static __inline uint32_t atomic_fetchadd_32(volatile uint32_t *p, uint32_t val) { uint32_t tmp, ret; int res; __asm __volatile( "1: ldxr %w2, [%3] \n" " add %w0, %w2, %w4 \n" " stxr %w1, %w0, [%3] \n" " cbnz %w1, 1b \n" : "=&r"(tmp), "=&r"(res), "=&r"(ret) : "r" (p), "r" (val) : "memory" ); return (ret); } static __inline uint64_t atomic_fetchadd_64(volatile uint64_t *p, uint64_t val) { uint64_t tmp, ret; int res; __asm __volatile( "1: ldxr %2, [%3] \n" " add %0, %2, %4 \n" " stxr %w1, %0, [%3] \n" " cbnz %w1, 1b \n" : "=&r"(tmp), "=&r"(res), "=&r"(ret) : "r" (p), "r" (val) : "memory" ); return (ret); } static __inline uint32_t atomic_readandclear_32(volatile uint32_t *p) { uint32_t ret; int res; __asm __volatile( "1: ldxr %w1, [%2] \n" " stxr %w0, wzr, [%2] \n" " cbnz %w0, 1b \n" : "=&r"(res), "=&r"(ret) : "r" (p) : "memory" ); return (ret); } static __inline uint64_t atomic_readandclear_64(volatile uint64_t *p) { uint64_t ret; int res; __asm __volatile( "1: ldxr %1, [%2] \n" " stxr %w0, xzr, [%2] \n" " cbnz %w0, 1b \n" : "=&r"(res), "=&r"(ret) : "r" (p) : "memory" ); return (ret); } static __inline uint32_t atomic_swap_32(volatile uint32_t *p, uint32_t val) { uint32_t ret; int res; __asm __volatile( "1: ldxr %w0, [%2] \n" " stxr %w1, %w3, [%2] \n" " cbnz %w1, 1b \n" : "=&r"(ret), "=&r"(res) : "r" (p), "r" (val) : "memory" ); return (ret); } static __inline uint64_t atomic_swap_64(volatile uint64_t *p, uint64_t val) { uint64_t ret; int res; __asm __volatile( "1: ldxr %0, [%2] \n" " stxr %w1, %3, [%2] \n" " cbnz %w1, 1b \n" : "=&r"(ret), "=&r"(res) : "r" (p), "r" (val) : "memory" ); return (ret); } static __inline uint32_t atomic_load_acq_32(volatile uint32_t *p) { uint32_t ret; __asm __volatile( "ldar %w0, [%1] \n" : "=&r" (ret) : "r" (p) : "memory"); return (ret); } static __inline uint64_t atomic_load_acq_64(volatile uint64_t *p) { uint64_t ret; __asm __volatile( "ldar %0, [%1] \n" : "=&r" (ret) : "r" (p) : "memory"); return (ret); } static __inline void atomic_store_rel_32(volatile uint32_t *p, uint32_t val) { __asm __volatile( "stlr %w0, [%1] \n" : : "r" (val), "r" (p) : "memory"); } static __inline void atomic_store_rel_64(volatile uint64_t *p, uint64_t val) { __asm __volatile( "stlr %0, [%1] \n" : : "r" (val), "r" (p) : "memory"); } #define atomic_add_int atomic_add_32 +#define atomic_fcmpset_int atomic_fcmpset_32 #define atomic_clear_int atomic_clear_32 #define atomic_cmpset_int atomic_cmpset_32 #define atomic_fetchadd_int atomic_fetchadd_32 #define atomic_readandclear_int atomic_readandclear_32 #define atomic_set_int atomic_set_32 #define atomic_swap_int atomic_swap_32 #define atomic_subtract_int atomic_subtract_32 #define atomic_add_acq_int atomic_add_acq_32 +#define atomic_fcmpset_acq_int atomic_fcmpset_acq_32 #define atomic_clear_acq_int atomic_clear_acq_32 #define atomic_cmpset_acq_int atomic_cmpset_acq_32 #define atomic_load_acq_int atomic_load_acq_32 #define atomic_set_acq_int atomic_set_acq_32 #define atomic_subtract_acq_int atomic_subtract_acq_32 #define atomic_add_rel_int atomic_add_rel_32 +#define atomic_fcmpset_rel_int atomic_fcmpset_rel_32 #define atomic_clear_rel_int atomic_add_rel_32 #define atomic_cmpset_rel_int atomic_cmpset_rel_32 #define atomic_set_rel_int atomic_set_rel_32 #define atomic_subtract_rel_int atomic_subtract_rel_32 #define atomic_store_rel_int atomic_store_rel_32 #define atomic_add_long atomic_add_64 +#define atomic_fcmpset_long atomic_fcmpset_64 #define atomic_clear_long atomic_clear_64 #define atomic_cmpset_long atomic_cmpset_64 #define atomic_fetchadd_long atomic_fetchadd_64 #define atomic_readandclear_long atomic_readandclear_64 #define atomic_set_long atomic_set_64 #define atomic_swap_long atomic_swap_64 #define atomic_subtract_long atomic_subtract_64 #define atomic_add_ptr atomic_add_64 +#define atomic_fcmpset_ptr atomic_fcmpset_64 #define atomic_clear_ptr atomic_clear_64 #define atomic_cmpset_ptr atomic_cmpset_64 #define atomic_fetchadd_ptr atomic_fetchadd_64 #define atomic_readandclear_ptr atomic_readandclear_64 #define atomic_set_ptr atomic_set_64 #define atomic_swap_ptr atomic_swap_64 #define atomic_subtract_ptr atomic_subtract_64 #define atomic_add_acq_long atomic_add_acq_64 +#define atomic_fcmpset_acq_long atomic_fcmpset_acq_64 #define atomic_clear_acq_long atomic_add_acq_64 #define atomic_cmpset_acq_long atomic_cmpset_acq_64 #define atomic_load_acq_long atomic_load_acq_64 #define atomic_set_acq_long atomic_set_acq_64 #define atomic_subtract_acq_long atomic_subtract_acq_64 #define atomic_add_acq_ptr atomic_add_acq_64 +#define atomic_fcmpset_acq_ptr atomic_fcmpset_acq_64 #define atomic_clear_acq_ptr atomic_add_acq_64 #define atomic_cmpset_acq_ptr atomic_cmpset_acq_64 #define atomic_load_acq_ptr atomic_load_acq_64 #define atomic_set_acq_ptr atomic_set_acq_64 #define atomic_subtract_acq_ptr atomic_subtract_acq_64 #define atomic_add_rel_long atomic_add_rel_64 +#define atomic_fcmpset_rel_long atomic_fcmpset_rel_64 #define atomic_clear_rel_long atomic_clear_rel_64 #define atomic_cmpset_rel_long atomic_cmpset_rel_64 #define atomic_set_rel_long atomic_set_rel_64 #define atomic_subtract_rel_long atomic_subtract_rel_64 #define atomic_store_rel_long atomic_store_rel_64 #define atomic_add_rel_ptr atomic_add_rel_64 +#define atomic_fcmpset_rel_ptr atomic_fcmpset_rel_64 #define atomic_clear_rel_ptr atomic_clear_rel_64 #define atomic_cmpset_rel_ptr atomic_cmpset_rel_64 #define atomic_set_rel_ptr atomic_set_rel_64 #define atomic_subtract_rel_ptr atomic_subtract_rel_64 #define atomic_store_rel_ptr atomic_store_rel_64 static __inline void atomic_thread_fence_acq(void) { dmb(ld); } static __inline void atomic_thread_fence_rel(void) { dmb(sy); } static __inline void atomic_thread_fence_acq_rel(void) { dmb(sy); } static __inline void atomic_thread_fence_seq_cst(void) { dmb(sy); } #endif /* _MACHINE_ATOMIC_H_ */ Index: stable/11/sys/i386/include/atomic.h =================================================================== --- stable/11/sys/i386/include/atomic.h (revision 315370) +++ stable/11/sys/i386/include/atomic.h (revision 315371) @@ -1,783 +1,817 @@ /*- * Copyright (c) 1998 Doug Rabson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_H_ #define _MACHINE_ATOMIC_H_ #ifndef _SYS_CDEFS_H_ #error this file needs sys/cdefs.h as a prerequisite #endif #ifdef _KERNEL #include #include #endif #ifndef __OFFSETOF_MONITORBUF /* * __OFFSETOF_MONITORBUF == __pcpu_offset(pc_monitorbuf). * * The open-coded number is used instead of the symbolic expression to * avoid a dependency on sys/pcpu.h in machine/atomic.h consumers. * An assertion in i386/vm_machdep.c ensures that the value is correct. */ #define __OFFSETOF_MONITORBUF 0x180 static __inline void __mbk(void) { __asm __volatile("lock; addl $0,%%fs:%0" : "+m" (*(u_int *)__OFFSETOF_MONITORBUF) : : "memory", "cc"); } static __inline void __mbu(void) { __asm __volatile("lock; addl $0,(%%esp)" : : : "memory", "cc"); } #endif /* * Various simple operations on memory, each of which is atomic in the * presence of interrupts and multiple processors. * * atomic_set_char(P, V) (*(u_char *)(P) |= (V)) * atomic_clear_char(P, V) (*(u_char *)(P) &= ~(V)) * atomic_add_char(P, V) (*(u_char *)(P) += (V)) * atomic_subtract_char(P, V) (*(u_char *)(P) -= (V)) * * atomic_set_short(P, V) (*(u_short *)(P) |= (V)) * atomic_clear_short(P, V) (*(u_short *)(P) &= ~(V)) * atomic_add_short(P, V) (*(u_short *)(P) += (V)) * atomic_subtract_short(P, V) (*(u_short *)(P) -= (V)) * * atomic_set_int(P, V) (*(u_int *)(P) |= (V)) * atomic_clear_int(P, V) (*(u_int *)(P) &= ~(V)) * atomic_add_int(P, V) (*(u_int *)(P) += (V)) * atomic_subtract_int(P, V) (*(u_int *)(P) -= (V)) * atomic_swap_int(P, V) (return (*(u_int *)(P)); *(u_int *)(P) = (V);) * atomic_readandclear_int(P) (return (*(u_int *)(P)); *(u_int *)(P) = 0;) * * atomic_set_long(P, V) (*(u_long *)(P) |= (V)) * atomic_clear_long(P, V) (*(u_long *)(P) &= ~(V)) * atomic_add_long(P, V) (*(u_long *)(P) += (V)) * atomic_subtract_long(P, V) (*(u_long *)(P) -= (V)) * atomic_swap_long(P, V) (return (*(u_long *)(P)); *(u_long *)(P) = (V);) * atomic_readandclear_long(P) (return (*(u_long *)(P)); *(u_long *)(P) = 0;) */ /* * The above functions are expanded inline in the statically-linked * kernel. Lock prefixes are generated if an SMP kernel is being * built. * * Kernel modules call real functions which are built into the kernel. * This allows kernel modules to be portable between UP and SMP systems. */ #if defined(KLD_MODULE) || !defined(__GNUCLIKE_ASM) #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \ void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v); \ void atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v) int atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src); +int atomic_fcmpset_int(volatile u_int *dst, u_int *expect, u_int src); u_int atomic_fetchadd_int(volatile u_int *p, u_int v); int atomic_testandset_int(volatile u_int *p, u_int v); int atomic_testandclear_int(volatile u_int *p, u_int v); void atomic_thread_fence_acq(void); void atomic_thread_fence_acq_rel(void); void atomic_thread_fence_rel(void); void atomic_thread_fence_seq_cst(void); #define ATOMIC_LOAD(TYPE) \ u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p) #define ATOMIC_STORE(TYPE) \ void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v) int atomic_cmpset_64(volatile uint64_t *, uint64_t, uint64_t); uint64_t atomic_load_acq_64(volatile uint64_t *); void atomic_store_rel_64(volatile uint64_t *, uint64_t); uint64_t atomic_swap_64(volatile uint64_t *, uint64_t); #else /* !KLD_MODULE && __GNUCLIKE_ASM */ /* * For userland, always use lock prefixes so that the binaries will run * on both SMP and !SMP systems. */ #if defined(SMP) || !defined(_KERNEL) #define MPLOCKED "lock ; " #else #define MPLOCKED #endif /* * The assembly is volatilized to avoid code chunk removal by the compiler. * GCC aggressively reorders operations and memory clobbering is necessary * in order to avoid that for memory barriers. */ #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \ static __inline void \ atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ { \ __asm __volatile(MPLOCKED OP \ : "+m" (*p) \ : CONS (V) \ : "cc"); \ } \ \ static __inline void \ atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ { \ __asm __volatile(MPLOCKED OP \ : "+m" (*p) \ : CONS (V) \ : "memory", "cc"); \ } \ struct __hack /* * Atomic compare and set, used by the mutex functions * * if (*dst == expect) *dst = src (all 32 bit words) * * Returns 0 on failure, non-zero on success */ static __inline int atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src) { u_char res; __asm __volatile( " " MPLOCKED " " " cmpxchgl %3,%1 ; " " sete %0 ; " "# atomic_cmpset_int" : "=q" (res), /* 0 */ "+m" (*dst), /* 1 */ "+a" (expect) /* 2 */ : "r" (src) /* 3 */ : "memory", "cc"); return (res); } +static __inline int +atomic_fcmpset_int(volatile u_int *dst, u_int *expect, u_int src) +{ + u_char res; + + __asm __volatile( + " " MPLOCKED " " + " cmpxchgl %3,%1 ; " + " sete %0 ; " + "# atomic_cmpset_int" + : "=q" (res), /* 0 */ + "+m" (*dst), /* 1 */ + "+a" (*expect) /* 2 */ + : "r" (src) /* 3 */ + : "memory", "cc"); + return (res); +} + /* * Atomically add the value of v to the integer pointed to by p and return * the previous value of *p. */ static __inline u_int atomic_fetchadd_int(volatile u_int *p, u_int v) { __asm __volatile( " " MPLOCKED " " " xaddl %0,%1 ; " "# atomic_fetchadd_int" : "+r" (v), /* 0 */ "+m" (*p) /* 1 */ : : "cc"); return (v); } static __inline int atomic_testandset_int(volatile u_int *p, u_int v) { u_char res; __asm __volatile( " " MPLOCKED " " " btsl %2,%1 ; " " setc %0 ; " "# atomic_testandset_int" : "=q" (res), /* 0 */ "+m" (*p) /* 1 */ : "Ir" (v & 0x1f) /* 2 */ : "cc"); return (res); } static __inline int atomic_testandclear_int(volatile u_int *p, u_int v) { u_char res; __asm __volatile( " " MPLOCKED " " " btrl %2,%1 ; " " setc %0 ; " "# atomic_testandclear_int" : "=q" (res), /* 0 */ "+m" (*p) /* 1 */ : "Ir" (v & 0x1f) /* 2 */ : "cc"); return (res); } /* * We assume that a = b will do atomic loads and stores. Due to the * IA32 memory model, a simple store guarantees release semantics. * * However, a load may pass a store if they are performed on distinct * addresses, so we need Store/Load barrier for sequentially * consistent fences in SMP kernels. We use "lock addl $0,mem" for a * Store/Load barrier, as recommended by the AMD Software Optimization * Guide, and not mfence. In the kernel, we use a private per-cpu * cache line for "mem", to avoid introducing false data * dependencies. In user space, we use the word at the top of the * stack. * * For UP kernels, however, the memory of the single processor is * always consistent, so we only need to stop the compiler from * reordering accesses in a way that violates the semantics of acquire * and release. */ #if defined(_KERNEL) #if defined(SMP) #define __storeload_barrier() __mbk() #else /* _KERNEL && UP */ #define __storeload_barrier() __compiler_membar() #endif /* SMP */ #else /* !_KERNEL */ #define __storeload_barrier() __mbu() #endif /* _KERNEL*/ #define ATOMIC_LOAD(TYPE) \ static __inline u_##TYPE \ atomic_load_acq_##TYPE(volatile u_##TYPE *p) \ { \ u_##TYPE res; \ \ res = *p; \ __compiler_membar(); \ return (res); \ } \ struct __hack #define ATOMIC_STORE(TYPE) \ static __inline void \ atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v) \ { \ \ __compiler_membar(); \ *p = v; \ } \ struct __hack static __inline void atomic_thread_fence_acq(void) { __compiler_membar(); } static __inline void atomic_thread_fence_rel(void) { __compiler_membar(); } static __inline void atomic_thread_fence_acq_rel(void) { __compiler_membar(); } static __inline void atomic_thread_fence_seq_cst(void) { __storeload_barrier(); } #ifdef _KERNEL #ifdef WANT_FUNCTIONS int atomic_cmpset_64_i386(volatile uint64_t *, uint64_t, uint64_t); int atomic_cmpset_64_i586(volatile uint64_t *, uint64_t, uint64_t); uint64_t atomic_load_acq_64_i386(volatile uint64_t *); uint64_t atomic_load_acq_64_i586(volatile uint64_t *); void atomic_store_rel_64_i386(volatile uint64_t *, uint64_t); void atomic_store_rel_64_i586(volatile uint64_t *, uint64_t); uint64_t atomic_swap_64_i386(volatile uint64_t *, uint64_t); uint64_t atomic_swap_64_i586(volatile uint64_t *, uint64_t); #endif /* I486 does not support SMP or CMPXCHG8B. */ static __inline int atomic_cmpset_64_i386(volatile uint64_t *dst, uint64_t expect, uint64_t src) { volatile uint32_t *p; u_char res; p = (volatile uint32_t *)dst; __asm __volatile( " pushfl ; " " cli ; " " xorl %1,%%eax ; " " xorl %2,%%edx ; " " orl %%edx,%%eax ; " " jne 1f ; " " movl %4,%1 ; " " movl %5,%2 ; " "1: " " sete %3 ; " " popfl" : "+A" (expect), /* 0 */ "+m" (*p), /* 1 */ "+m" (*(p + 1)), /* 2 */ "=q" (res) /* 3 */ : "r" ((uint32_t)src), /* 4 */ "r" ((uint32_t)(src >> 32)) /* 5 */ : "memory", "cc"); return (res); } static __inline uint64_t atomic_load_acq_64_i386(volatile uint64_t *p) { volatile uint32_t *q; uint64_t res; q = (volatile uint32_t *)p; __asm __volatile( " pushfl ; " " cli ; " " movl %1,%%eax ; " " movl %2,%%edx ; " " popfl" : "=&A" (res) /* 0 */ : "m" (*q), /* 1 */ "m" (*(q + 1)) /* 2 */ : "memory"); return (res); } static __inline void atomic_store_rel_64_i386(volatile uint64_t *p, uint64_t v) { volatile uint32_t *q; q = (volatile uint32_t *)p; __asm __volatile( " pushfl ; " " cli ; " " movl %%eax,%0 ; " " movl %%edx,%1 ; " " popfl" : "=m" (*q), /* 0 */ "=m" (*(q + 1)) /* 1 */ : "A" (v) /* 2 */ : "memory"); } static __inline uint64_t atomic_swap_64_i386(volatile uint64_t *p, uint64_t v) { volatile uint32_t *q; uint64_t res; q = (volatile uint32_t *)p; __asm __volatile( " pushfl ; " " cli ; " " movl %1,%%eax ; " " movl %2,%%edx ; " " movl %4,%2 ; " " movl %3,%1 ; " " popfl" : "=&A" (res), /* 0 */ "+m" (*q), /* 1 */ "+m" (*(q + 1)) /* 2 */ : "r" ((uint32_t)v), /* 3 */ "r" ((uint32_t)(v >> 32))); /* 4 */ return (res); } static __inline int atomic_cmpset_64_i586(volatile uint64_t *dst, uint64_t expect, uint64_t src) { u_char res; __asm __volatile( " " MPLOCKED " " " cmpxchg8b %1 ; " " sete %0" : "=q" (res), /* 0 */ "+m" (*dst), /* 1 */ "+A" (expect) /* 2 */ : "b" ((uint32_t)src), /* 3 */ "c" ((uint32_t)(src >> 32)) /* 4 */ : "memory", "cc"); return (res); } static __inline uint64_t atomic_load_acq_64_i586(volatile uint64_t *p) { uint64_t res; __asm __volatile( " movl %%ebx,%%eax ; " " movl %%ecx,%%edx ; " " " MPLOCKED " " " cmpxchg8b %1" : "=&A" (res), /* 0 */ "+m" (*p) /* 1 */ : : "memory", "cc"); return (res); } static __inline void atomic_store_rel_64_i586(volatile uint64_t *p, uint64_t v) { __asm __volatile( " movl %%eax,%%ebx ; " " movl %%edx,%%ecx ; " "1: " " " MPLOCKED " " " cmpxchg8b %0 ; " " jne 1b" : "+m" (*p), /* 0 */ "+A" (v) /* 1 */ : : "ebx", "ecx", "memory", "cc"); } static __inline uint64_t atomic_swap_64_i586(volatile uint64_t *p, uint64_t v) { __asm __volatile( " movl %%eax,%%ebx ; " " movl %%edx,%%ecx ; " "1: " " " MPLOCKED " " " cmpxchg8b %0 ; " " jne 1b" : "+m" (*p), /* 0 */ "+A" (v) /* 1 */ : : "ebx", "ecx", "memory", "cc"); return (v); } static __inline int atomic_cmpset_64(volatile uint64_t *dst, uint64_t expect, uint64_t src) { if ((cpu_feature & CPUID_CX8) == 0) return (atomic_cmpset_64_i386(dst, expect, src)); else return (atomic_cmpset_64_i586(dst, expect, src)); } static __inline uint64_t atomic_load_acq_64(volatile uint64_t *p) { if ((cpu_feature & CPUID_CX8) == 0) return (atomic_load_acq_64_i386(p)); else return (atomic_load_acq_64_i586(p)); } static __inline void atomic_store_rel_64(volatile uint64_t *p, uint64_t v) { if ((cpu_feature & CPUID_CX8) == 0) atomic_store_rel_64_i386(p, v); else atomic_store_rel_64_i586(p, v); } static __inline uint64_t atomic_swap_64(volatile uint64_t *p, uint64_t v) { if ((cpu_feature & CPUID_CX8) == 0) return (atomic_swap_64_i386(p, v)); else return (atomic_swap_64_i586(p, v)); } #endif /* _KERNEL */ #endif /* KLD_MODULE || !__GNUCLIKE_ASM */ ATOMIC_ASM(set, char, "orb %b1,%0", "iq", v); ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v); ATOMIC_ASM(add, char, "addb %b1,%0", "iq", v); ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v); ATOMIC_ASM(set, short, "orw %w1,%0", "ir", v); ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v); ATOMIC_ASM(add, short, "addw %w1,%0", "ir", v); ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v); ATOMIC_ASM(set, int, "orl %1,%0", "ir", v); ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v); ATOMIC_ASM(add, int, "addl %1,%0", "ir", v); ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v); ATOMIC_ASM(set, long, "orl %1,%0", "ir", v); ATOMIC_ASM(clear, long, "andl %1,%0", "ir", ~v); ATOMIC_ASM(add, long, "addl %1,%0", "ir", v); ATOMIC_ASM(subtract, long, "subl %1,%0", "ir", v); #define ATOMIC_LOADSTORE(TYPE) \ ATOMIC_LOAD(TYPE); \ ATOMIC_STORE(TYPE) ATOMIC_LOADSTORE(char); ATOMIC_LOADSTORE(short); ATOMIC_LOADSTORE(int); ATOMIC_LOADSTORE(long); #undef ATOMIC_ASM #undef ATOMIC_LOAD #undef ATOMIC_STORE #undef ATOMIC_LOADSTORE #ifndef WANT_FUNCTIONS static __inline int atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src) { return (atomic_cmpset_int((volatile u_int *)dst, (u_int)expect, (u_int)src)); } static __inline u_long atomic_fetchadd_long(volatile u_long *p, u_long v) { return (atomic_fetchadd_int((volatile u_int *)p, (u_int)v)); } static __inline int atomic_testandset_long(volatile u_long *p, u_int v) { return (atomic_testandset_int((volatile u_int *)p, v)); } static __inline int atomic_testandclear_long(volatile u_long *p, u_int v) { return (atomic_testandclear_int((volatile u_int *)p, v)); } /* Read the current value and store a new value in the destination. */ #ifdef __GNUCLIKE_ASM static __inline u_int atomic_swap_int(volatile u_int *p, u_int v) { __asm __volatile( " xchgl %1,%0 ; " "# atomic_swap_int" : "+r" (v), /* 0 */ "+m" (*p)); /* 1 */ return (v); } static __inline u_long atomic_swap_long(volatile u_long *p, u_long v) { return (atomic_swap_int((volatile u_int *)p, (u_int)v)); } #else /* !__GNUCLIKE_ASM */ u_int atomic_swap_int(volatile u_int *p, u_int v); u_long atomic_swap_long(volatile u_long *p, u_long v); #endif /* __GNUCLIKE_ASM */ #define atomic_set_acq_char atomic_set_barr_char #define atomic_set_rel_char atomic_set_barr_char #define atomic_clear_acq_char atomic_clear_barr_char #define atomic_clear_rel_char atomic_clear_barr_char #define atomic_add_acq_char atomic_add_barr_char #define atomic_add_rel_char atomic_add_barr_char #define atomic_subtract_acq_char atomic_subtract_barr_char #define atomic_subtract_rel_char atomic_subtract_barr_char #define atomic_set_acq_short atomic_set_barr_short #define atomic_set_rel_short atomic_set_barr_short #define atomic_clear_acq_short atomic_clear_barr_short #define atomic_clear_rel_short atomic_clear_barr_short #define atomic_add_acq_short atomic_add_barr_short #define atomic_add_rel_short atomic_add_barr_short #define atomic_subtract_acq_short atomic_subtract_barr_short #define atomic_subtract_rel_short atomic_subtract_barr_short #define atomic_set_acq_int atomic_set_barr_int #define atomic_set_rel_int atomic_set_barr_int #define atomic_clear_acq_int atomic_clear_barr_int #define atomic_clear_rel_int atomic_clear_barr_int #define atomic_add_acq_int atomic_add_barr_int #define atomic_add_rel_int atomic_add_barr_int #define atomic_subtract_acq_int atomic_subtract_barr_int #define atomic_subtract_rel_int atomic_subtract_barr_int #define atomic_cmpset_acq_int atomic_cmpset_int #define atomic_cmpset_rel_int atomic_cmpset_int +#define atomic_fcmpset_acq_int atomic_fcmpset_int +#define atomic_fcmpset_rel_int atomic_fcmpset_int #define atomic_set_acq_long atomic_set_barr_long #define atomic_set_rel_long atomic_set_barr_long #define atomic_clear_acq_long atomic_clear_barr_long #define atomic_clear_rel_long atomic_clear_barr_long #define atomic_add_acq_long atomic_add_barr_long #define atomic_add_rel_long atomic_add_barr_long #define atomic_subtract_acq_long atomic_subtract_barr_long #define atomic_subtract_rel_long atomic_subtract_barr_long #define atomic_cmpset_acq_long atomic_cmpset_long #define atomic_cmpset_rel_long atomic_cmpset_long +#define atomic_fcmpset_acq_long atomic_fcmpset_long +#define atomic_fcmpset_rel_long atomic_fcmpset_long #define atomic_readandclear_int(p) atomic_swap_int(p, 0) #define atomic_readandclear_long(p) atomic_swap_long(p, 0) /* Operations on 8-bit bytes. */ #define atomic_set_8 atomic_set_char #define atomic_set_acq_8 atomic_set_acq_char #define atomic_set_rel_8 atomic_set_rel_char #define atomic_clear_8 atomic_clear_char #define atomic_clear_acq_8 atomic_clear_acq_char #define atomic_clear_rel_8 atomic_clear_rel_char #define atomic_add_8 atomic_add_char #define atomic_add_acq_8 atomic_add_acq_char #define atomic_add_rel_8 atomic_add_rel_char #define atomic_subtract_8 atomic_subtract_char #define atomic_subtract_acq_8 atomic_subtract_acq_char #define atomic_subtract_rel_8 atomic_subtract_rel_char #define atomic_load_acq_8 atomic_load_acq_char #define atomic_store_rel_8 atomic_store_rel_char /* Operations on 16-bit words. */ #define atomic_set_16 atomic_set_short #define atomic_set_acq_16 atomic_set_acq_short #define atomic_set_rel_16 atomic_set_rel_short #define atomic_clear_16 atomic_clear_short #define atomic_clear_acq_16 atomic_clear_acq_short #define atomic_clear_rel_16 atomic_clear_rel_short #define atomic_add_16 atomic_add_short #define atomic_add_acq_16 atomic_add_acq_short #define atomic_add_rel_16 atomic_add_rel_short #define atomic_subtract_16 atomic_subtract_short #define atomic_subtract_acq_16 atomic_subtract_acq_short #define atomic_subtract_rel_16 atomic_subtract_rel_short #define atomic_load_acq_16 atomic_load_acq_short #define atomic_store_rel_16 atomic_store_rel_short /* Operations on 32-bit double words. */ #define atomic_set_32 atomic_set_int #define atomic_set_acq_32 atomic_set_acq_int #define atomic_set_rel_32 atomic_set_rel_int #define atomic_clear_32 atomic_clear_int #define atomic_clear_acq_32 atomic_clear_acq_int #define atomic_clear_rel_32 atomic_clear_rel_int #define atomic_add_32 atomic_add_int #define atomic_add_acq_32 atomic_add_acq_int #define atomic_add_rel_32 atomic_add_rel_int #define atomic_subtract_32 atomic_subtract_int #define atomic_subtract_acq_32 atomic_subtract_acq_int #define atomic_subtract_rel_32 atomic_subtract_rel_int #define atomic_load_acq_32 atomic_load_acq_int #define atomic_store_rel_32 atomic_store_rel_int #define atomic_cmpset_32 atomic_cmpset_int #define atomic_cmpset_acq_32 atomic_cmpset_acq_int #define atomic_cmpset_rel_32 atomic_cmpset_rel_int +#define atomic_fcmpset_32 atomic_fcmpset_int +#define atomic_fcmpset_acq_32 atomic_fcmpset_acq_int +#define atomic_fcmpset_rel_32 atomic_fcmpset_rel_int #define atomic_swap_32 atomic_swap_int #define atomic_readandclear_32 atomic_readandclear_int #define atomic_fetchadd_32 atomic_fetchadd_int #define atomic_testandset_32 atomic_testandset_int #define atomic_testandclear_32 atomic_testandclear_int /* Operations on pointers. */ #define atomic_set_ptr(p, v) \ atomic_set_int((volatile u_int *)(p), (u_int)(v)) #define atomic_set_acq_ptr(p, v) \ atomic_set_acq_int((volatile u_int *)(p), (u_int)(v)) #define atomic_set_rel_ptr(p, v) \ atomic_set_rel_int((volatile u_int *)(p), (u_int)(v)) #define atomic_clear_ptr(p, v) \ atomic_clear_int((volatile u_int *)(p), (u_int)(v)) #define atomic_clear_acq_ptr(p, v) \ atomic_clear_acq_int((volatile u_int *)(p), (u_int)(v)) #define atomic_clear_rel_ptr(p, v) \ atomic_clear_rel_int((volatile u_int *)(p), (u_int)(v)) #define atomic_add_ptr(p, v) \ atomic_add_int((volatile u_int *)(p), (u_int)(v)) #define atomic_add_acq_ptr(p, v) \ atomic_add_acq_int((volatile u_int *)(p), (u_int)(v)) #define atomic_add_rel_ptr(p, v) \ atomic_add_rel_int((volatile u_int *)(p), (u_int)(v)) #define atomic_subtract_ptr(p, v) \ atomic_subtract_int((volatile u_int *)(p), (u_int)(v)) #define atomic_subtract_acq_ptr(p, v) \ atomic_subtract_acq_int((volatile u_int *)(p), (u_int)(v)) #define atomic_subtract_rel_ptr(p, v) \ atomic_subtract_rel_int((volatile u_int *)(p), (u_int)(v)) #define atomic_load_acq_ptr(p) \ atomic_load_acq_int((volatile u_int *)(p)) #define atomic_store_rel_ptr(p, v) \ atomic_store_rel_int((volatile u_int *)(p), (v)) #define atomic_cmpset_ptr(dst, old, new) \ atomic_cmpset_int((volatile u_int *)(dst), (u_int)(old), (u_int)(new)) #define atomic_cmpset_acq_ptr(dst, old, new) \ atomic_cmpset_acq_int((volatile u_int *)(dst), (u_int)(old), \ (u_int)(new)) #define atomic_cmpset_rel_ptr(dst, old, new) \ atomic_cmpset_rel_int((volatile u_int *)(dst), (u_int)(old), \ + (u_int)(new)) +#define atomic_fcmpset_ptr(dst, old, new) \ + atomic_fcmpset_int((volatile u_int *)(dst), (u_int *)(old), (u_int)(new)) +#define atomic_fcmpset_acq_ptr(dst, old, new) \ + atomic_fcmpset_acq_int((volatile u_int *)(dst), (u_int *)(old), \ + (u_int)(new)) +#define atomic_fcmpset_rel_ptr(dst, old, new) \ + atomic_fcmpset_rel_int((volatile u_int *)(dst), (u_int *)(old), \ (u_int)(new)) #define atomic_swap_ptr(p, v) \ atomic_swap_int((volatile u_int *)(p), (u_int)(v)) #define atomic_readandclear_ptr(p) \ atomic_readandclear_int((volatile u_int *)(p)) #endif /* !WANT_FUNCTIONS */ #if defined(_KERNEL) #define mb() __mbk() #define wmb() __mbk() #define rmb() __mbk() #else #define mb() __mbu() #define wmb() __mbu() #define rmb() __mbu() #endif #endif /* !_MACHINE_ATOMIC_H_ */ Index: stable/11/sys/mips/include/atomic.h =================================================================== --- stable/11/sys/mips/include/atomic.h (revision 315370) +++ stable/11/sys/mips/include/atomic.h (revision 315371) @@ -1,672 +1,771 @@ /*- * Copyright (c) 1998 Doug Rabson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: src/sys/alpha/include/atomic.h,v 1.21.2.3 2005/10/06 18:12:05 jhb * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_H_ #define _MACHINE_ATOMIC_H_ #ifndef _SYS_CDEFS_H_ #error this file needs sys/cdefs.h as a prerequisite #endif /* * Note: All the 64-bit atomic operations are only atomic when running * in 64-bit mode. It is assumed that code compiled for n32 and n64 * fits into this definition and no further safeties are needed. * * It is also assumed that the add, subtract and other arithmetic is * done on numbers not pointers. The special rules for n32 pointers * do not have atomic operations defined for them, but generally shouldn't * need atomic operations. */ #ifndef __MIPS_PLATFORM_SYNC_NOPS #define __MIPS_PLATFORM_SYNC_NOPS "" #endif static __inline void mips_sync(void) { __asm __volatile (".set noreorder\n" "\tsync\n" __MIPS_PLATFORM_SYNC_NOPS ".set reorder\n" : : : "memory"); } #define mb() mips_sync() #define wmb() mips_sync() #define rmb() mips_sync() /* * Various simple arithmetic on memory which is atomic in the presence * of interrupts and SMP safe. */ void atomic_set_8(__volatile uint8_t *, uint8_t); void atomic_clear_8(__volatile uint8_t *, uint8_t); void atomic_add_8(__volatile uint8_t *, uint8_t); void atomic_subtract_8(__volatile uint8_t *, uint8_t); void atomic_set_16(__volatile uint16_t *, uint16_t); void atomic_clear_16(__volatile uint16_t *, uint16_t); void atomic_add_16(__volatile uint16_t *, uint16_t); void atomic_subtract_16(__volatile uint16_t *, uint16_t); static __inline void atomic_set_32(__volatile uint32_t *p, uint32_t v) { uint32_t temp; __asm __volatile ( "1:\tll %0, %3\n\t" /* load old value */ "or %0, %2, %0\n\t" /* calculate new value */ "sc %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* spin if failed */ : "=&r" (temp), "=m" (*p) : "r" (v), "m" (*p) : "memory"); } static __inline void atomic_clear_32(__volatile uint32_t *p, uint32_t v) { uint32_t temp; v = ~v; __asm __volatile ( "1:\tll %0, %3\n\t" /* load old value */ "and %0, %2, %0\n\t" /* calculate new value */ "sc %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* spin if failed */ : "=&r" (temp), "=m" (*p) : "r" (v), "m" (*p) : "memory"); } static __inline void atomic_add_32(__volatile uint32_t *p, uint32_t v) { uint32_t temp; __asm __volatile ( "1:\tll %0, %3\n\t" /* load old value */ "addu %0, %2, %0\n\t" /* calculate new value */ "sc %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* spin if failed */ : "=&r" (temp), "=m" (*p) : "r" (v), "m" (*p) : "memory"); } static __inline void atomic_subtract_32(__volatile uint32_t *p, uint32_t v) { uint32_t temp; __asm __volatile ( "1:\tll %0, %3\n\t" /* load old value */ "subu %0, %2\n\t" /* calculate new value */ "sc %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* spin if failed */ : "=&r" (temp), "=m" (*p) : "r" (v), "m" (*p) : "memory"); } static __inline uint32_t atomic_readandclear_32(__volatile uint32_t *addr) { uint32_t result,temp; __asm __volatile ( "1:\tll %0,%3\n\t" /* load current value, asserting lock */ "li %1,0\n\t" /* value to store */ "sc %1,%2\n\t" /* attempt to store */ "beqz %1, 1b\n\t" /* if the store failed, spin */ : "=&r"(result), "=&r"(temp), "=m" (*addr) : "m" (*addr) : "memory"); return result; } static __inline uint32_t atomic_readandset_32(__volatile uint32_t *addr, uint32_t value) { uint32_t result,temp; __asm __volatile ( "1:\tll %0,%3\n\t" /* load current value, asserting lock */ "or %1,$0,%4\n\t" "sc %1,%2\n\t" /* attempt to store */ "beqz %1, 1b\n\t" /* if the store failed, spin */ : "=&r"(result), "=&r"(temp), "=m" (*addr) : "m" (*addr), "r" (value) : "memory"); return result; } #if defined(__mips_n64) || defined(__mips_n32) static __inline void atomic_set_64(__volatile uint64_t *p, uint64_t v) { uint64_t temp; __asm __volatile ( "1:\n\t" "lld %0, %3\n\t" /* load old value */ "or %0, %2, %0\n\t" /* calculate new value */ "scd %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* spin if failed */ : "=&r" (temp), "=m" (*p) : "r" (v), "m" (*p) : "memory"); } static __inline void atomic_clear_64(__volatile uint64_t *p, uint64_t v) { uint64_t temp; v = ~v; __asm __volatile ( "1:\n\t" "lld %0, %3\n\t" /* load old value */ "and %0, %2, %0\n\t" /* calculate new value */ "scd %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* spin if failed */ : "=&r" (temp), "=m" (*p) : "r" (v), "m" (*p) : "memory"); } static __inline void atomic_add_64(__volatile uint64_t *p, uint64_t v) { uint64_t temp; __asm __volatile ( "1:\n\t" "lld %0, %3\n\t" /* load old value */ "daddu %0, %2, %0\n\t" /* calculate new value */ "scd %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* spin if failed */ : "=&r" (temp), "=m" (*p) : "r" (v), "m" (*p) : "memory"); } static __inline void atomic_subtract_64(__volatile uint64_t *p, uint64_t v) { uint64_t temp; __asm __volatile ( "1:\n\t" "lld %0, %3\n\t" /* load old value */ "dsubu %0, %2\n\t" /* calculate new value */ "scd %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* spin if failed */ : "=&r" (temp), "=m" (*p) : "r" (v), "m" (*p) : "memory"); } static __inline uint64_t atomic_readandclear_64(__volatile uint64_t *addr) { uint64_t result,temp; __asm __volatile ( "1:\n\t" "lld %0, %3\n\t" /* load old value */ "li %1, 0\n\t" /* value to store */ "scd %1, %2\n\t" /* attempt to store */ "beqz %1, 1b\n\t" /* if the store failed, spin */ : "=&r"(result), "=&r"(temp), "=m" (*addr) : "m" (*addr) : "memory"); return result; } static __inline uint64_t atomic_readandset_64(__volatile uint64_t *addr, uint64_t value) { uint64_t result,temp; __asm __volatile ( "1:\n\t" "lld %0,%3\n\t" /* Load old value*/ "or %1,$0,%4\n\t" "scd %1,%2\n\t" /* attempt to store */ "beqz %1, 1b\n\t" /* if the store failed, spin */ : "=&r"(result), "=&r"(temp), "=m" (*addr) : "m" (*addr), "r" (value) : "memory"); return result; } #endif #define ATOMIC_ACQ_REL(NAME, WIDTH) \ static __inline void \ atomic_##NAME##_acq_##WIDTH(__volatile uint##WIDTH##_t *p, uint##WIDTH##_t v)\ { \ atomic_##NAME##_##WIDTH(p, v); \ mips_sync(); \ } \ \ static __inline void \ atomic_##NAME##_rel_##WIDTH(__volatile uint##WIDTH##_t *p, uint##WIDTH##_t v)\ { \ mips_sync(); \ atomic_##NAME##_##WIDTH(p, v); \ } /* Variants of simple arithmetic with memory barriers. */ ATOMIC_ACQ_REL(set, 8) ATOMIC_ACQ_REL(clear, 8) ATOMIC_ACQ_REL(add, 8) ATOMIC_ACQ_REL(subtract, 8) ATOMIC_ACQ_REL(set, 16) ATOMIC_ACQ_REL(clear, 16) ATOMIC_ACQ_REL(add, 16) ATOMIC_ACQ_REL(subtract, 16) ATOMIC_ACQ_REL(set, 32) ATOMIC_ACQ_REL(clear, 32) ATOMIC_ACQ_REL(add, 32) ATOMIC_ACQ_REL(subtract, 32) #if defined(__mips_n64) || defined(__mips_n32) ATOMIC_ACQ_REL(set, 64) ATOMIC_ACQ_REL(clear, 64) ATOMIC_ACQ_REL(add, 64) ATOMIC_ACQ_REL(subtract, 64) #endif #undef ATOMIC_ACQ_REL /* * We assume that a = b will do atomic loads and stores. */ #define ATOMIC_STORE_LOAD(WIDTH) \ static __inline uint##WIDTH##_t \ atomic_load_acq_##WIDTH(__volatile uint##WIDTH##_t *p) \ { \ uint##WIDTH##_t v; \ \ v = *p; \ mips_sync(); \ return (v); \ } \ \ static __inline void \ atomic_store_rel_##WIDTH(__volatile uint##WIDTH##_t *p, uint##WIDTH##_t v)\ { \ mips_sync(); \ *p = v; \ } ATOMIC_STORE_LOAD(32) ATOMIC_STORE_LOAD(64) #if !defined(__mips_n64) && !defined(__mips_n32) void atomic_store_64(__volatile uint64_t *, uint64_t *); void atomic_load_64(__volatile uint64_t *, uint64_t *); #else static __inline void atomic_store_64(__volatile uint64_t *p, uint64_t *v) { *p = *v; } static __inline void atomic_load_64(__volatile uint64_t *p, uint64_t *v) { *v = *p; } #endif #undef ATOMIC_STORE_LOAD /* * Atomically compare the value stored at *p with cmpval and if the * two values are equal, update the value of *p with newval. Returns * zero if the compare failed, nonzero otherwise. */ static __inline uint32_t -atomic_cmpset_32(__volatile uint32_t* p, uint32_t cmpval, uint32_t newval) +atomic_cmpset_32(__volatile uint32_t *p, uint32_t cmpval, uint32_t newval) { uint32_t ret; __asm __volatile ( "1:\tll %0, %4\n\t" /* load old value */ "bne %0, %2, 2f\n\t" /* compare */ "move %0, %3\n\t" /* value to store */ "sc %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* if it failed, spin */ "j 3f\n\t" "2:\n\t" "li %0, 0\n\t" "3:\n" : "=&r" (ret), "=m" (*p) : "r" (cmpval), "r" (newval), "m" (*p) : "memory"); return ret; } /* * Atomically compare the value stored at *p with cmpval and if the * two values are equal, update the value of *p with newval. Returns * zero if the compare failed, nonzero otherwise. */ static __inline uint32_t atomic_cmpset_acq_32(__volatile uint32_t *p, uint32_t cmpval, uint32_t newval) { int retval; retval = atomic_cmpset_32(p, cmpval, newval); mips_sync(); return (retval); } static __inline uint32_t atomic_cmpset_rel_32(__volatile uint32_t *p, uint32_t cmpval, uint32_t newval) { mips_sync(); return (atomic_cmpset_32(p, cmpval, newval)); } +static __inline uint32_t +atomic_fcmpset_32(__volatile uint32_t *p, uint32_t *cmpval, uint32_t newval) +{ + uint32_t ret; + + __asm __volatile ( + "1:\n\t" + "ll %0, %1\n\t" /* load old value */ + "bne %0, %4, 2f\n\t" /* compare */ + "move %0, %3\n\t" /* value to store */ + "sc %0, %1\n\t" /* attempt to store */ + "beqz %0, 1b\n\t" /* if it failed, spin */ + "j 3f\n\t" + "2:\n\t" + "sw %0, %2\n\t" /* save old value */ + "li %0, 0\n\t" + "3:\n" + : "=&r" (ret), "+m" (*p), "=m" (*cmpval) + : "r" (newval), "r" (*cmpval) + : "memory"); + return ret; +} + +static __inline uint32_t +atomic_fcmpset_acq_32(__volatile uint32_t *p, uint32_t *cmpval, uint32_t newval) +{ + int retval; + + retval = atomic_fcmpset_32(p, cmpval, newval); + mips_sync(); + return (retval); +} + +static __inline uint32_t +atomic_fcmpset_rel_32(__volatile uint32_t *p, uint32_t *cmpval, uint32_t newval) +{ + mips_sync(); + return (atomic_fcmpset_32(p, cmpval, newval)); +} + /* * Atomically add the value of v to the integer pointed to by p and return * the previous value of *p. */ static __inline uint32_t atomic_fetchadd_32(__volatile uint32_t *p, uint32_t v) { uint32_t value, temp; __asm __volatile ( "1:\tll %0, %1\n\t" /* load old value */ "addu %2, %3, %0\n\t" /* calculate new value */ "sc %2, %1\n\t" /* attempt to store */ "beqz %2, 1b\n\t" /* spin if failed */ : "=&r" (value), "=m" (*p), "=&r" (temp) : "r" (v), "m" (*p)); return (value); } #if defined(__mips_n64) || defined(__mips_n32) /* * Atomically compare the value stored at *p with cmpval and if the * two values are equal, update the value of *p with newval. Returns * zero if the compare failed, nonzero otherwise. */ static __inline uint64_t -atomic_cmpset_64(__volatile uint64_t* p, uint64_t cmpval, uint64_t newval) +atomic_cmpset_64(__volatile uint64_t *p, uint64_t cmpval, uint64_t newval) { uint64_t ret; __asm __volatile ( "1:\n\t" "lld %0, %4\n\t" /* load old value */ "bne %0, %2, 2f\n\t" /* compare */ "move %0, %3\n\t" /* value to store */ "scd %0, %1\n\t" /* attempt to store */ "beqz %0, 1b\n\t" /* if it failed, spin */ "j 3f\n\t" "2:\n\t" "li %0, 0\n\t" "3:\n" : "=&r" (ret), "=m" (*p) : "r" (cmpval), "r" (newval), "m" (*p) : "memory"); return ret; } /* * Atomically compare the value stored at *p with cmpval and if the * two values are equal, update the value of *p with newval. Returns * zero if the compare failed, nonzero otherwise. */ static __inline uint64_t atomic_cmpset_acq_64(__volatile uint64_t *p, uint64_t cmpval, uint64_t newval) { int retval; retval = atomic_cmpset_64(p, cmpval, newval); mips_sync(); return (retval); } static __inline uint64_t atomic_cmpset_rel_64(__volatile uint64_t *p, uint64_t cmpval, uint64_t newval) { mips_sync(); return (atomic_cmpset_64(p, cmpval, newval)); } +static __inline uint32_t +atomic_fcmpset_64(__volatile uint64_t *p, uint64_t *cmpval, uint64_t newval) +{ + uint32_t ret; + + __asm __volatile ( + "1:\n\t" + "lld %0, %1\n\t" /* load old value */ + "bne %0, %4, 2f\n\t" /* compare */ + "move %0, %3\n\t" /* value to store */ + "scd %0, %1\n\t" /* attempt to store */ + "beqz %0, 1b\n\t" /* if it failed, spin */ + "j 3f\n\t" + "2:\n\t" + "sd %0, %2\n\t" /* save old value */ + "li %0, 0\n\t" + "3:\n" + : "=&r" (ret), "+m" (*p), "=m" (*cmpval) + : "r" (newval), "r" (*cmpval) + : "memory"); + + return ret; +} + +static __inline uint64_t +atomic_fcmpset_acq_64(__volatile uint64_t *p, uint64_t *cmpval, uint64_t newval) +{ + int retval; + + retval = atomic_fcmpset_64(p, cmpval, newval); + mips_sync(); + return (retval); +} + +static __inline uint64_t +atomic_fcmpset_rel_64(__volatile uint64_t *p, uint64_t *cmpval, uint64_t newval) +{ + mips_sync(); + return (atomic_fcmpset_64(p, cmpval, newval)); +} + /* * Atomically add the value of v to the integer pointed to by p and return * the previous value of *p. */ static __inline uint64_t atomic_fetchadd_64(__volatile uint64_t *p, uint64_t v) { uint64_t value, temp; __asm __volatile ( "1:\n\t" "lld %0, %1\n\t" /* load old value */ "daddu %2, %3, %0\n\t" /* calculate new value */ "scd %2, %1\n\t" /* attempt to store */ "beqz %2, 1b\n\t" /* spin if failed */ : "=&r" (value), "=m" (*p), "=&r" (temp) : "r" (v), "m" (*p)); return (value); } #endif static __inline void atomic_thread_fence_acq(void) { mips_sync(); } static __inline void atomic_thread_fence_rel(void) { mips_sync(); } static __inline void atomic_thread_fence_acq_rel(void) { mips_sync(); } static __inline void atomic_thread_fence_seq_cst(void) { mips_sync(); } /* Operations on chars. */ #define atomic_set_char atomic_set_8 #define atomic_set_acq_char atomic_set_acq_8 #define atomic_set_rel_char atomic_set_rel_8 #define atomic_clear_char atomic_clear_8 #define atomic_clear_acq_char atomic_clear_acq_8 #define atomic_clear_rel_char atomic_clear_rel_8 #define atomic_add_char atomic_add_8 #define atomic_add_acq_char atomic_add_acq_8 #define atomic_add_rel_char atomic_add_rel_8 #define atomic_subtract_char atomic_subtract_8 #define atomic_subtract_acq_char atomic_subtract_acq_8 #define atomic_subtract_rel_char atomic_subtract_rel_8 /* Operations on shorts. */ #define atomic_set_short atomic_set_16 #define atomic_set_acq_short atomic_set_acq_16 #define atomic_set_rel_short atomic_set_rel_16 #define atomic_clear_short atomic_clear_16 #define atomic_clear_acq_short atomic_clear_acq_16 #define atomic_clear_rel_short atomic_clear_rel_16 #define atomic_add_short atomic_add_16 #define atomic_add_acq_short atomic_add_acq_16 #define atomic_add_rel_short atomic_add_rel_16 #define atomic_subtract_short atomic_subtract_16 #define atomic_subtract_acq_short atomic_subtract_acq_16 #define atomic_subtract_rel_short atomic_subtract_rel_16 /* Operations on ints. */ #define atomic_set_int atomic_set_32 #define atomic_set_acq_int atomic_set_acq_32 #define atomic_set_rel_int atomic_set_rel_32 #define atomic_clear_int atomic_clear_32 #define atomic_clear_acq_int atomic_clear_acq_32 #define atomic_clear_rel_int atomic_clear_rel_32 #define atomic_add_int atomic_add_32 #define atomic_add_acq_int atomic_add_acq_32 #define atomic_add_rel_int atomic_add_rel_32 #define atomic_subtract_int atomic_subtract_32 #define atomic_subtract_acq_int atomic_subtract_acq_32 #define atomic_subtract_rel_int atomic_subtract_rel_32 #define atomic_cmpset_int atomic_cmpset_32 #define atomic_cmpset_acq_int atomic_cmpset_acq_32 #define atomic_cmpset_rel_int atomic_cmpset_rel_32 +#define atomic_fcmpset_int atomic_fcmpset_32 +#define atomic_fcmpset_acq_int atomic_fcmpset_acq_32 +#define atomic_fcmpset_rel_int atomic_fcmpset_rel_32 #define atomic_load_acq_int atomic_load_acq_32 #define atomic_store_rel_int atomic_store_rel_32 #define atomic_readandclear_int atomic_readandclear_32 #define atomic_readandset_int atomic_readandset_32 #define atomic_fetchadd_int atomic_fetchadd_32 /* * I think the following is right, even for n32. For n32 the pointers * are still 32-bits, so we need to operate on them as 32-bit quantities, * even though they are sign extended in operation. For longs, there's * no question because they are always 32-bits. */ #ifdef __mips_n64 /* Operations on longs. */ #define atomic_set_long atomic_set_64 #define atomic_set_acq_long atomic_set_acq_64 #define atomic_set_rel_long atomic_set_rel_64 #define atomic_clear_long atomic_clear_64 #define atomic_clear_acq_long atomic_clear_acq_64 #define atomic_clear_rel_long atomic_clear_rel_64 #define atomic_add_long atomic_add_64 #define atomic_add_acq_long atomic_add_acq_64 #define atomic_add_rel_long atomic_add_rel_64 #define atomic_subtract_long atomic_subtract_64 #define atomic_subtract_acq_long atomic_subtract_acq_64 #define atomic_subtract_rel_long atomic_subtract_rel_64 #define atomic_cmpset_long atomic_cmpset_64 #define atomic_cmpset_acq_long atomic_cmpset_acq_64 #define atomic_cmpset_rel_long atomic_cmpset_rel_64 +#define atomic_fcmpset_long atomic_fcmpset_64 +#define atomic_fcmpset_acq_long atomic_fcmpset_acq_64 +#define atomic_fcmpset_rel_long atomic_fcmpset_rel_64 #define atomic_load_acq_long atomic_load_acq_64 #define atomic_store_rel_long atomic_store_rel_64 #define atomic_fetchadd_long atomic_fetchadd_64 #define atomic_readandclear_long atomic_readandclear_64 #else /* !__mips_n64 */ /* Operations on longs. */ #define atomic_set_long(p, v) \ atomic_set_32((volatile u_int *)(p), (u_int)(v)) #define atomic_set_acq_long(p, v) \ atomic_set_acq_32((volatile u_int *)(p), (u_int)(v)) #define atomic_set_rel_long(p, v) \ atomic_set_rel_32((volatile u_int *)(p), (u_int)(v)) #define atomic_clear_long(p, v) \ atomic_clear_32((volatile u_int *)(p), (u_int)(v)) #define atomic_clear_acq_long(p, v) \ atomic_clear_acq_32((volatile u_int *)(p), (u_int)(v)) #define atomic_clear_rel_long(p, v) \ atomic_clear_rel_32((volatile u_int *)(p), (u_int)(v)) #define atomic_add_long(p, v) \ atomic_add_32((volatile u_int *)(p), (u_int)(v)) #define atomic_add_acq_long(p, v) \ atomic_add_32((volatile u_int *)(p), (u_int)(v)) #define atomic_add_rel_long(p, v) \ atomic_add_32((volatile u_int *)(p), (u_int)(v)) #define atomic_subtract_long(p, v) \ atomic_subtract_32((volatile u_int *)(p), (u_int)(v)) #define atomic_subtract_acq_long(p, v) \ atomic_subtract_acq_32((volatile u_int *)(p), (u_int)(v)) #define atomic_subtract_rel_long(p, v) \ atomic_subtract_rel_32((volatile u_int *)(p), (u_int)(v)) #define atomic_cmpset_long(p, cmpval, newval) \ atomic_cmpset_32((volatile u_int *)(p), (u_int)(cmpval), \ (u_int)(newval)) #define atomic_cmpset_acq_long(p, cmpval, newval) \ atomic_cmpset_acq_32((volatile u_int *)(p), (u_int)(cmpval), \ (u_int)(newval)) #define atomic_cmpset_rel_long(p, cmpval, newval) \ atomic_cmpset_rel_32((volatile u_int *)(p), (u_int)(cmpval), \ (u_int)(newval)) +#define atomic_fcmpset_long(p, cmpval, newval) \ + atomic_fcmpset_32((volatile u_int *)(p), (u_int *)(cmpval), \ + (u_int)(newval)) +#define atomic_fcmpset_acq_long(p, cmpval, newval) \ + atomic_fcmpset_acq_32((volatile u_int *)(p), (u_int *)(cmpval), \ + (u_int)(newval)) +#define atomic_fcmpset_rel_long(p, cmpval, newval) \ + atomic_fcmpset_rel_32((volatile u_int *)(p), (u_int *)(cmpval), \ + (u_int)(newval)) #define atomic_load_acq_long(p) \ (u_long)atomic_load_acq_32((volatile u_int *)(p)) #define atomic_store_rel_long(p, v) \ atomic_store_rel_32((volatile u_int *)(p), (u_int)(v)) #define atomic_fetchadd_long(p, v) \ atomic_fetchadd_32((volatile u_int *)(p), (u_int)(v)) #define atomic_readandclear_long(p) \ atomic_readandclear_32((volatile u_int *)(p)) #endif /* __mips_n64 */ /* Operations on pointers. */ #define atomic_set_ptr atomic_set_long #define atomic_set_acq_ptr atomic_set_acq_long #define atomic_set_rel_ptr atomic_set_rel_long #define atomic_clear_ptr atomic_clear_long #define atomic_clear_acq_ptr atomic_clear_acq_long #define atomic_clear_rel_ptr atomic_clear_rel_long #define atomic_add_ptr atomic_add_long #define atomic_add_acq_ptr atomic_add_acq_long #define atomic_add_rel_ptr atomic_add_rel_long #define atomic_subtract_ptr atomic_subtract_long #define atomic_subtract_acq_ptr atomic_subtract_acq_long #define atomic_subtract_rel_ptr atomic_subtract_rel_long #define atomic_cmpset_ptr atomic_cmpset_long #define atomic_cmpset_acq_ptr atomic_cmpset_acq_long #define atomic_cmpset_rel_ptr atomic_cmpset_rel_long +#define atomic_fcmpset_ptr atomic_fcmpset_long +#define atomic_fcmpset_acq_ptr atomic_fcmpset_acq_long +#define atomic_fcmpset_rel_ptr atomic_fcmpset_rel_long #define atomic_load_acq_ptr atomic_load_acq_long #define atomic_store_rel_ptr atomic_store_rel_long #define atomic_readandclear_ptr atomic_readandclear_long #endif /* ! _MACHINE_ATOMIC_H_ */ Index: stable/11/sys/powerpc/include/atomic.h =================================================================== --- stable/11/sys/powerpc/include/atomic.h (revision 315370) +++ stable/11/sys/powerpc/include/atomic.h (revision 315371) @@ -1,773 +1,896 @@ /*- * Copyright (c) 2008 Marcel Moolenaar * Copyright (c) 2001 Benno Rice * Copyright (c) 2001 David E. O'Brien * Copyright (c) 1998 Doug Rabson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_H_ #define _MACHINE_ATOMIC_H_ #ifndef _SYS_CDEFS_H_ #error this file needs sys/cdefs.h as a prerequisite #endif /* * The __ATOMIC_REL/ACQ() macros provide memory barriers only in conjunction * with the atomic lXarx/stXcx. sequences below. They are not exposed outside * of this file. See also Appendix B.2 of Book II of the architecture manual. * * Note that not all Book-E processors accept the light-weight sync variant. * In particular, early models of E500 cores are known to wedge. Bank on all * 64-bit capable CPUs to accept lwsync properly and pressimize 32-bit CPUs * to use the heavier-weight sync. */ #ifdef __powerpc64__ #define mb() __asm __volatile("sync" : : : "memory") #define rmb() __asm __volatile("lwsync" : : : "memory") #define wmb() __asm __volatile("lwsync" : : : "memory") #define __ATOMIC_REL() __asm __volatile("lwsync" : : : "memory") #define __ATOMIC_ACQ() __asm __volatile("isync" : : : "memory") #else #define mb() __asm __volatile("sync" : : : "memory") #define rmb() __asm __volatile("sync" : : : "memory") #define wmb() __asm __volatile("sync" : : : "memory") #define __ATOMIC_REL() __asm __volatile("sync" : : : "memory") #define __ATOMIC_ACQ() __asm __volatile("isync" : : : "memory") #endif static __inline void powerpc_lwsync(void) { #ifdef __powerpc64__ __asm __volatile("lwsync" : : : "memory"); #else __asm __volatile("sync" : : : "memory"); #endif } /* * atomic_add(p, v) * { *p += v; } */ #define __atomic_add_int(p, v, t) \ __asm __volatile( \ "1: lwarx %0, 0, %2\n" \ " add %0, %3, %0\n" \ " stwcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_add_int */ #ifdef __powerpc64__ #define __atomic_add_long(p, v, t) \ __asm __volatile( \ "1: ldarx %0, 0, %2\n" \ " add %0, %3, %0\n" \ " stdcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_add_long */ #else #define __atomic_add_long(p, v, t) \ __asm __volatile( \ "1: lwarx %0, 0, %2\n" \ " add %0, %3, %0\n" \ " stwcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_add_long */ #endif #define _ATOMIC_ADD(type) \ static __inline void \ atomic_add_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __atomic_add_##type(p, v, t); \ } \ \ static __inline void \ atomic_add_acq_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __atomic_add_##type(p, v, t); \ __ATOMIC_ACQ(); \ } \ \ static __inline void \ atomic_add_rel_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __ATOMIC_REL(); \ __atomic_add_##type(p, v, t); \ } \ /* _ATOMIC_ADD */ _ATOMIC_ADD(int) _ATOMIC_ADD(long) #define atomic_add_32 atomic_add_int #define atomic_add_acq_32 atomic_add_acq_int #define atomic_add_rel_32 atomic_add_rel_int #ifdef __powerpc64__ #define atomic_add_64 atomic_add_long #define atomic_add_acq_64 atomic_add_acq_long #define atomic_add_rel_64 atomic_add_rel_long #define atomic_add_ptr atomic_add_long #define atomic_add_acq_ptr atomic_add_acq_long #define atomic_add_rel_ptr atomic_add_rel_long #else #define atomic_add_ptr atomic_add_int #define atomic_add_acq_ptr atomic_add_acq_int #define atomic_add_rel_ptr atomic_add_rel_int #endif #undef _ATOMIC_ADD #undef __atomic_add_long #undef __atomic_add_int /* * atomic_clear(p, v) * { *p &= ~v; } */ #define __atomic_clear_int(p, v, t) \ __asm __volatile( \ "1: lwarx %0, 0, %2\n" \ " andc %0, %0, %3\n" \ " stwcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_clear_int */ #ifdef __powerpc64__ #define __atomic_clear_long(p, v, t) \ __asm __volatile( \ "1: ldarx %0, 0, %2\n" \ " andc %0, %0, %3\n" \ " stdcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_clear_long */ #else #define __atomic_clear_long(p, v, t) \ __asm __volatile( \ "1: lwarx %0, 0, %2\n" \ " andc %0, %0, %3\n" \ " stwcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_clear_long */ #endif #define _ATOMIC_CLEAR(type) \ static __inline void \ atomic_clear_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __atomic_clear_##type(p, v, t); \ } \ \ static __inline void \ atomic_clear_acq_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __atomic_clear_##type(p, v, t); \ __ATOMIC_ACQ(); \ } \ \ static __inline void \ atomic_clear_rel_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __ATOMIC_REL(); \ __atomic_clear_##type(p, v, t); \ } \ /* _ATOMIC_CLEAR */ _ATOMIC_CLEAR(int) _ATOMIC_CLEAR(long) #define atomic_clear_32 atomic_clear_int #define atomic_clear_acq_32 atomic_clear_acq_int #define atomic_clear_rel_32 atomic_clear_rel_int #ifdef __powerpc64__ #define atomic_clear_64 atomic_clear_long #define atomic_clear_acq_64 atomic_clear_acq_long #define atomic_clear_rel_64 atomic_clear_rel_long #define atomic_clear_ptr atomic_clear_long #define atomic_clear_acq_ptr atomic_clear_acq_long #define atomic_clear_rel_ptr atomic_clear_rel_long #else #define atomic_clear_ptr atomic_clear_int #define atomic_clear_acq_ptr atomic_clear_acq_int #define atomic_clear_rel_ptr atomic_clear_rel_int #endif #undef _ATOMIC_CLEAR #undef __atomic_clear_long #undef __atomic_clear_int /* * atomic_cmpset(p, o, n) */ /* TODO -- see below */ /* * atomic_load_acq(p) */ /* TODO -- see below */ /* * atomic_readandclear(p) */ /* TODO -- see below */ /* * atomic_set(p, v) * { *p |= v; } */ #define __atomic_set_int(p, v, t) \ __asm __volatile( \ "1: lwarx %0, 0, %2\n" \ " or %0, %3, %0\n" \ " stwcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_set_int */ #ifdef __powerpc64__ #define __atomic_set_long(p, v, t) \ __asm __volatile( \ "1: ldarx %0, 0, %2\n" \ " or %0, %3, %0\n" \ " stdcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_set_long */ #else #define __atomic_set_long(p, v, t) \ __asm __volatile( \ "1: lwarx %0, 0, %2\n" \ " or %0, %3, %0\n" \ " stwcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_set_long */ #endif #define _ATOMIC_SET(type) \ static __inline void \ atomic_set_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __atomic_set_##type(p, v, t); \ } \ \ static __inline void \ atomic_set_acq_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __atomic_set_##type(p, v, t); \ __ATOMIC_ACQ(); \ } \ \ static __inline void \ atomic_set_rel_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __ATOMIC_REL(); \ __atomic_set_##type(p, v, t); \ } \ /* _ATOMIC_SET */ _ATOMIC_SET(int) _ATOMIC_SET(long) #define atomic_set_32 atomic_set_int #define atomic_set_acq_32 atomic_set_acq_int #define atomic_set_rel_32 atomic_set_rel_int #ifdef __powerpc64__ #define atomic_set_64 atomic_set_long #define atomic_set_acq_64 atomic_set_acq_long #define atomic_set_rel_64 atomic_set_rel_long #define atomic_set_ptr atomic_set_long #define atomic_set_acq_ptr atomic_set_acq_long #define atomic_set_rel_ptr atomic_set_rel_long #else #define atomic_set_ptr atomic_set_int #define atomic_set_acq_ptr atomic_set_acq_int #define atomic_set_rel_ptr atomic_set_rel_int #endif #undef _ATOMIC_SET #undef __atomic_set_long #undef __atomic_set_int /* * atomic_subtract(p, v) * { *p -= v; } */ #define __atomic_subtract_int(p, v, t) \ __asm __volatile( \ "1: lwarx %0, 0, %2\n" \ " subf %0, %3, %0\n" \ " stwcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_subtract_int */ #ifdef __powerpc64__ #define __atomic_subtract_long(p, v, t) \ __asm __volatile( \ "1: ldarx %0, 0, %2\n" \ " subf %0, %3, %0\n" \ " stdcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_subtract_long */ #else #define __atomic_subtract_long(p, v, t) \ __asm __volatile( \ "1: lwarx %0, 0, %2\n" \ " subf %0, %3, %0\n" \ " stwcx. %0, 0, %2\n" \ " bne- 1b\n" \ : "=&r" (t), "=m" (*p) \ : "r" (p), "r" (v), "m" (*p) \ : "cr0", "memory") \ /* __atomic_subtract_long */ #endif #define _ATOMIC_SUBTRACT(type) \ static __inline void \ atomic_subtract_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __atomic_subtract_##type(p, v, t); \ } \ \ static __inline void \ atomic_subtract_acq_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __atomic_subtract_##type(p, v, t); \ __ATOMIC_ACQ(); \ } \ \ static __inline void \ atomic_subtract_rel_##type(volatile u_##type *p, u_##type v) { \ u_##type t; \ __ATOMIC_REL(); \ __atomic_subtract_##type(p, v, t); \ } \ /* _ATOMIC_SUBTRACT */ _ATOMIC_SUBTRACT(int) _ATOMIC_SUBTRACT(long) #define atomic_subtract_32 atomic_subtract_int #define atomic_subtract_acq_32 atomic_subtract_acq_int #define atomic_subtract_rel_32 atomic_subtract_rel_int #ifdef __powerpc64__ #define atomic_subtract_64 atomic_subtract_long #define atomic_subtract_acq_64 atomic_subract_acq_long #define atomic_subtract_rel_64 atomic_subtract_rel_long #define atomic_subtract_ptr atomic_subtract_long #define atomic_subtract_acq_ptr atomic_subtract_acq_long #define atomic_subtract_rel_ptr atomic_subtract_rel_long #else #define atomic_subtract_ptr atomic_subtract_int #define atomic_subtract_acq_ptr atomic_subtract_acq_int #define atomic_subtract_rel_ptr atomic_subtract_rel_int #endif #undef _ATOMIC_SUBTRACT #undef __atomic_subtract_long #undef __atomic_subtract_int /* * atomic_store_rel(p, v) */ /* TODO -- see below */ /* * Old/original implementations that still need revisiting. */ static __inline u_int atomic_readandclear_int(volatile u_int *addr) { u_int result,temp; #ifdef __GNUCLIKE_ASM __asm __volatile ( "\tsync\n" /* drain writes */ "1:\tlwarx %0, 0, %3\n\t" /* load old value */ "li %1, 0\n\t" /* load new value */ "stwcx. %1, 0, %3\n\t" /* attempt to store */ "bne- 1b\n\t" /* spin if failed */ : "=&r"(result), "=&r"(temp), "=m" (*addr) : "r" (addr), "m" (*addr) : "cr0", "memory"); #endif return (result); } #ifdef __powerpc64__ static __inline u_long atomic_readandclear_long(volatile u_long *addr) { u_long result,temp; #ifdef __GNUCLIKE_ASM __asm __volatile ( "\tsync\n" /* drain writes */ "1:\tldarx %0, 0, %3\n\t" /* load old value */ "li %1, 0\n\t" /* load new value */ "stdcx. %1, 0, %3\n\t" /* attempt to store */ "bne- 1b\n\t" /* spin if failed */ : "=&r"(result), "=&r"(temp), "=m" (*addr) : "r" (addr), "m" (*addr) : "cr0", "memory"); #endif return (result); } #endif #define atomic_readandclear_32 atomic_readandclear_int #ifdef __powerpc64__ #define atomic_readandclear_64 atomic_readandclear_long #define atomic_readandclear_ptr atomic_readandclear_long #else static __inline u_long atomic_readandclear_long(volatile u_long *addr) { return ((u_long)atomic_readandclear_int((volatile u_int *)addr)); } #define atomic_readandclear_ptr atomic_readandclear_int #endif /* * We assume that a = b will do atomic loads and stores. */ #define ATOMIC_STORE_LOAD(TYPE) \ static __inline u_##TYPE \ atomic_load_acq_##TYPE(volatile u_##TYPE *p) \ { \ u_##TYPE v; \ \ v = *p; \ mb(); \ return (v); \ } \ \ static __inline void \ atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v) \ { \ \ powerpc_lwsync(); \ *p = v; \ } ATOMIC_STORE_LOAD(int) #define atomic_load_acq_32 atomic_load_acq_int #define atomic_store_rel_32 atomic_store_rel_int #ifdef __powerpc64__ ATOMIC_STORE_LOAD(long) #define atomic_load_acq_64 atomic_load_acq_long #define atomic_store_rel_64 atomic_store_rel_long #define atomic_load_acq_ptr atomic_load_acq_long #define atomic_store_rel_ptr atomic_store_rel_long #else static __inline u_long atomic_load_acq_long(volatile u_long *addr) { return ((u_long)atomic_load_acq_int((volatile u_int *)addr)); } static __inline void atomic_store_rel_long(volatile u_long *addr, u_long val) { atomic_store_rel_int((volatile u_int *)addr, (u_int)val); } #define atomic_load_acq_ptr atomic_load_acq_int #define atomic_store_rel_ptr atomic_store_rel_int #endif #undef ATOMIC_STORE_LOAD /* * Atomically compare the value stored at *p with cmpval and if the * two values are equal, update the value of *p with newval. Returns * zero if the compare failed, nonzero otherwise. */ static __inline int atomic_cmpset_int(volatile u_int* p, u_int cmpval, u_int newval) { int ret; #ifdef __GNUCLIKE_ASM __asm __volatile ( "1:\tlwarx %0, 0, %2\n\t" /* load old value */ "cmplw %3, %0\n\t" /* compare */ "bne 2f\n\t" /* exit if not equal */ "stwcx. %4, 0, %2\n\t" /* attempt to store */ "bne- 1b\n\t" /* spin if failed */ "li %0, 1\n\t" /* success - retval = 1 */ "b 3f\n\t" /* we've succeeded */ "2:\n\t" "stwcx. %0, 0, %2\n\t" /* clear reservation (74xx) */ "li %0, 0\n\t" /* failure - retval = 0 */ "3:\n\t" : "=&r" (ret), "=m" (*p) : "r" (p), "r" (cmpval), "r" (newval), "m" (*p) : "cr0", "memory"); #endif return (ret); } static __inline int atomic_cmpset_long(volatile u_long* p, u_long cmpval, u_long newval) { int ret; #ifdef __GNUCLIKE_ASM __asm __volatile ( #ifdef __powerpc64__ "1:\tldarx %0, 0, %2\n\t" /* load old value */ "cmpld %3, %0\n\t" /* compare */ "bne 2f\n\t" /* exit if not equal */ "stdcx. %4, 0, %2\n\t" /* attempt to store */ #else "1:\tlwarx %0, 0, %2\n\t" /* load old value */ "cmplw %3, %0\n\t" /* compare */ "bne 2f\n\t" /* exit if not equal */ "stwcx. %4, 0, %2\n\t" /* attempt to store */ #endif "bne- 1b\n\t" /* spin if failed */ "li %0, 1\n\t" /* success - retval = 1 */ "b 3f\n\t" /* we've succeeded */ "2:\n\t" #ifdef __powerpc64__ "stdcx. %0, 0, %2\n\t" /* clear reservation (74xx) */ #else "stwcx. %0, 0, %2\n\t" /* clear reservation (74xx) */ #endif "li %0, 0\n\t" /* failure - retval = 0 */ "3:\n\t" : "=&r" (ret), "=m" (*p) : "r" (p), "r" (cmpval), "r" (newval), "m" (*p) : "cr0", "memory"); #endif return (ret); } static __inline int atomic_cmpset_acq_int(volatile u_int *p, u_int cmpval, u_int newval) { int retval; retval = atomic_cmpset_int(p, cmpval, newval); __ATOMIC_ACQ(); return (retval); } static __inline int atomic_cmpset_rel_int(volatile u_int *p, u_int cmpval, u_int newval) { __ATOMIC_REL(); return (atomic_cmpset_int(p, cmpval, newval)); } static __inline int atomic_cmpset_acq_long(volatile u_long *p, u_long cmpval, u_long newval) { u_long retval; retval = atomic_cmpset_long(p, cmpval, newval); __ATOMIC_ACQ(); return (retval); } static __inline int atomic_cmpset_rel_long(volatile u_long *p, u_long cmpval, u_long newval) { __ATOMIC_REL(); return (atomic_cmpset_long(p, cmpval, newval)); } #define atomic_cmpset_32 atomic_cmpset_int #define atomic_cmpset_acq_32 atomic_cmpset_acq_int #define atomic_cmpset_rel_32 atomic_cmpset_rel_int #ifdef __powerpc64__ #define atomic_cmpset_64 atomic_cmpset_long #define atomic_cmpset_acq_64 atomic_cmpset_acq_long #define atomic_cmpset_rel_64 atomic_cmpset_rel_long #define atomic_cmpset_ptr atomic_cmpset_long #define atomic_cmpset_acq_ptr atomic_cmpset_acq_long #define atomic_cmpset_rel_ptr atomic_cmpset_rel_long #else #define atomic_cmpset_ptr atomic_cmpset_int #define atomic_cmpset_acq_ptr atomic_cmpset_acq_int #define atomic_cmpset_rel_ptr atomic_cmpset_rel_int #endif +/* + * Atomically compare the value stored at *p with *cmpval and if the + * two values are equal, update the value of *p with newval. Returns + * zero if the compare failed and sets *cmpval to the read value from *p, + * nonzero otherwise. + */ +static __inline int +atomic_fcmpset_int(volatile u_int *p, u_int *cmpval, u_int newval) +{ + int ret; + +#ifdef __GNUCLIKE_ASM + __asm __volatile ( + "lwarx %0, 0, %3\n\t" /* load old value */ + "cmplw %4, %0\n\t" /* compare */ + "bne 1f\n\t" /* exit if not equal */ + "stwcx. %5, 0, %3\n\t" /* attempt to store */ + "bne- 1f\n\t" /* exit if failed */ + "li %0, 1\n\t" /* success - retval = 1 */ + "b 2f\n\t" /* we've succeeded */ + "1:\n\t" + "stwcx. %0, 0, %3\n\t" /* clear reservation (74xx) */ + "stwx %0, 0, %7\n\t" + "li %0, 0\n\t" /* failure - retval = 0 */ + "2:\n\t" + : "=&r" (ret), "=m" (*p), "=m" (*cmpval) + : "r" (p), "r" (*cmpval), "r" (newval), "m" (*p), "r"(cmpval) + : "cr0", "memory"); +#endif + + return (ret); +} +static __inline int +atomic_fcmpset_long(volatile u_long *p, u_long *cmpval, u_long newval) +{ + int ret; + +#ifdef __GNUCLIKE_ASM + __asm __volatile ( + #ifdef __powerpc64__ + "ldarx %0, 0, %3\n\t" /* load old value */ + "cmpld %4, %0\n\t" /* compare */ + "bne 1f\n\t" /* exit if not equal */ + "stdcx. %5, 0, %3\n\t" /* attempt to store */ + #else + "lwarx %0, 0, %3\n\t" /* load old value */ + "cmplw %4, %0\n\t" /* compare */ + "bne 1f\n\t" /* exit if not equal */ + "stwcx. %5, 0, %3\n\t" /* attempt to store */ + #endif + "bne- 1f\n\t" /* exit if failed */ + "li %0, 1\n\t" /* success - retval = 1 */ + "b 2f\n\t" /* we've succeeded */ + "1:\n\t" + #ifdef __powerpc64__ + "stdcx. %0, 0, %3\n\t" /* clear reservation (74xx) */ + "stdx %0, 0, %7\n\t" + #else + "stwcx. %0, 0, %3\n\t" /* clear reservation (74xx) */ + "stwx %0, 0, %7\n\t" + #endif + "li %0, 0\n\t" /* failure - retval = 0 */ + "2:\n\t" + : "=&r" (ret), "=m" (*p), "=m" (*cmpval) + : "r" (p), "r" (*cmpval), "r" (newval), "m" (*p), "r"(cmpval) + : "cr0", "memory"); +#endif + + return (ret); +} + +static __inline int +atomic_fcmpset_acq_int(volatile u_int *p, u_int *cmpval, u_int newval) +{ + int retval; + + retval = atomic_fcmpset_int(p, cmpval, newval); + __ATOMIC_ACQ(); + return (retval); +} + +static __inline int +atomic_fcmpset_rel_int(volatile u_int *p, u_int *cmpval, u_int newval) +{ + __ATOMIC_REL(); + return (atomic_fcmpset_int(p, cmpval, newval)); +} + +static __inline int +atomic_fcmpset_acq_long(volatile u_long *p, u_long *cmpval, u_long newval) +{ + u_long retval; + + retval = atomic_fcmpset_long(p, cmpval, newval); + __ATOMIC_ACQ(); + return (retval); +} + +static __inline int +atomic_fcmpset_rel_long(volatile u_long *p, u_long *cmpval, u_long newval) +{ + __ATOMIC_REL(); + return (atomic_fcmpset_long(p, cmpval, newval)); +} + +#define atomic_fcmpset_32 atomic_fcmpset_int +#define atomic_fcmpset_acq_32 atomic_fcmpset_acq_int +#define atomic_fcmpset_rel_32 atomic_fcmpset_rel_int + +#ifdef __powerpc64__ +#define atomic_fcmpset_64 atomic_fcmpset_long +#define atomic_fcmpset_acq_64 atomic_fcmpset_acq_long +#define atomic_fcmpset_rel_64 atomic_fcmpset_rel_long + +#define atomic_fcmpset_ptr atomic_fcmpset_long +#define atomic_fcmpset_acq_ptr atomic_fcmpset_acq_long +#define atomic_fcmpset_rel_ptr atomic_fcmpset_rel_long +#else +#define atomic_fcmpset_ptr atomic_fcmpset_int +#define atomic_fcmpset_acq_ptr atomic_fcmpset_acq_int +#define atomic_fcmpset_rel_ptr atomic_fcmpset_rel_int +#endif + static __inline u_int atomic_fetchadd_int(volatile u_int *p, u_int v) { u_int value; do { value = *p; } while (!atomic_cmpset_int(p, value, value + v)); return (value); } static __inline u_long atomic_fetchadd_long(volatile u_long *p, u_long v) { u_long value; do { value = *p; } while (!atomic_cmpset_long(p, value, value + v)); return (value); } static __inline u_int atomic_swap_32(volatile u_int *p, u_int v) { u_int prev; __asm __volatile( "1: lwarx %0,0,%2\n" " stwcx. %3,0,%2\n" " bne- 1b\n" : "=&r" (prev), "+m" (*(volatile u_int *)p) : "r" (p), "r" (v) : "cr0", "memory"); return (prev); } #ifdef __powerpc64__ static __inline u_long atomic_swap_64(volatile u_long *p, u_long v) { u_long prev; __asm __volatile( "1: ldarx %0,0,%2\n" " stdcx. %3,0,%2\n" " bne- 1b\n" : "=&r" (prev), "+m" (*(volatile u_long *)p) : "r" (p), "r" (v) : "cr0", "memory"); return (prev); } #endif #define atomic_fetchadd_32 atomic_fetchadd_int #define atomic_swap_int atomic_swap_32 #ifdef __powerpc64__ #define atomic_fetchadd_64 atomic_fetchadd_long #define atomic_swap_long atomic_swap_64 #define atomic_swap_ptr atomic_swap_64 #endif #undef __ATOMIC_REL #undef __ATOMIC_ACQ static __inline void atomic_thread_fence_acq(void) { powerpc_lwsync(); } static __inline void atomic_thread_fence_rel(void) { powerpc_lwsync(); } static __inline void atomic_thread_fence_acq_rel(void) { powerpc_lwsync(); } static __inline void atomic_thread_fence_seq_cst(void) { __asm __volatile("sync" : : : "memory"); } #endif /* ! _MACHINE_ATOMIC_H_ */ Index: stable/11/sys/riscv/include/atomic.h =================================================================== --- stable/11/sys/riscv/include/atomic.h (revision 315370) +++ stable/11/sys/riscv/include/atomic.h (revision 315371) @@ -1,461 +1,562 @@ /*- * Copyright (c) 2015 Ruslan Bukin * All rights reserved. * * Portions of this software were developed by SRI International and the * University of Cambridge Computer Laboratory under DARPA/AFRL contract * FA8750-10-C-0237 ("CTSRD"), as part of the DARPA CRASH research programme. * * Portions of this software were developed by the University of Cambridge * Computer Laboratory as part of the CTSRD Project, with support from the * UK Higher Education Innovation Fund (HEIF). * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_H_ #define _MACHINE_ATOMIC_H_ #define fence() __asm __volatile("fence" ::: "memory"); #define mb() fence() #define rmb() fence() #define wmb() fence() #define ATOMIC_ACQ_REL(NAME, WIDTH) \ static __inline void \ atomic_##NAME##_acq_##WIDTH(__volatile uint##WIDTH##_t *p, uint##WIDTH##_t v)\ { \ atomic_##NAME##_##WIDTH(p, v); \ fence(); \ } \ \ static __inline void \ atomic_##NAME##_rel_##WIDTH(__volatile uint##WIDTH##_t *p, uint##WIDTH##_t v)\ { \ fence(); \ atomic_##NAME##_##WIDTH(p, v); \ } static __inline void atomic_add_32(volatile uint32_t *p, uint32_t val) { __asm __volatile("amoadd.w zero, %1, %0" : "+A" (*p) : "r" (val) : "memory"); } static __inline void atomic_subtract_32(volatile uint32_t *p, uint32_t val) { __asm __volatile("amoadd.w zero, %1, %0" : "+A" (*p) : "r" (-val) : "memory"); } static __inline void atomic_set_32(volatile uint32_t *p, uint32_t val) { __asm __volatile("amoor.w zero, %1, %0" : "+A" (*p) : "r" (val) : "memory"); } static __inline void atomic_clear_32(volatile uint32_t *p, uint32_t val) { __asm __volatile("amoand.w zero, %1, %0" : "+A" (*p) : "r" (~val) : "memory"); } static __inline int atomic_cmpset_32(volatile uint32_t *p, uint32_t cmpval, uint32_t newval) { uint32_t tmp; int res; res = 0; __asm __volatile( "0:" "li %1, 1\n" /* Preset to fail */ "lr.w %0, %2\n" "bne %0, %z3, 1f\n" "sc.w %1, %z4, %2\n" "bnez %1, 0b\n" "1:" : "=&r" (tmp), "=&r" (res), "+A" (*p) : "rJ" (cmpval), "rJ" (newval) : "memory"); return (!res); } +static __inline int +atomic_fcmpset_32(volatile uint32_t *p, uint32_t *cmpval, uint32_t newval) +{ + uint32_t tmp; + int res; + + res = 0; + + __asm __volatile( + "0:" + "li %1, 1\n" /* Preset to fail */ + "lr.w %0, %2\n" /* Load old value */ + "bne %0, %z4, 1f\n" /* Compare */ + "sc.w %1, %z5, %2\n" /* Try to store new value */ + "j 2f\n" + "1:" + "sw %0, %3\n" /* Save old value */ + "2:" + : "=&r" (tmp), "=&r" (res), "+A" (*p), "+A" (*cmpval) + : "rJ" (*cmpval), "rJ" (newval) + : "memory"); + + return (!res); +} + static __inline uint32_t atomic_fetchadd_32(volatile uint32_t *p, uint32_t val) { uint32_t ret; __asm __volatile("amoadd.w %0, %2, %1" : "=&r" (ret), "+A" (*p) : "r" (val) : "memory"); return (ret); } static __inline uint32_t atomic_readandclear_32(volatile uint32_t *p) { uint32_t ret; uint32_t val; val = 0; __asm __volatile("amoswap.w %0, %2, %1" : "=&r"(ret), "+A" (*p) : "r" (val) : "memory"); return (ret); } #define atomic_add_int atomic_add_32 #define atomic_clear_int atomic_clear_32 #define atomic_cmpset_int atomic_cmpset_32 +#define atomic_fcmpset_int atomic_fcmpset_32 #define atomic_fetchadd_int atomic_fetchadd_32 #define atomic_readandclear_int atomic_readandclear_32 #define atomic_set_int atomic_set_32 #define atomic_subtract_int atomic_subtract_32 ATOMIC_ACQ_REL(set, 32) ATOMIC_ACQ_REL(clear, 32) ATOMIC_ACQ_REL(add, 32) ATOMIC_ACQ_REL(subtract, 32) static __inline int atomic_cmpset_acq_32(volatile uint32_t *p, uint32_t cmpval, uint32_t newval) { int res; res = atomic_cmpset_32(p, cmpval, newval); fence(); return (res); } static __inline int atomic_cmpset_rel_32(volatile uint32_t *p, uint32_t cmpval, uint32_t newval) { fence(); return (atomic_cmpset_32(p, cmpval, newval)); } +static __inline int +atomic_fcmpset_acq_32(volatile uint32_t *p, uint32_t *cmpval, uint32_t newval) +{ + int res; + + res = atomic_fcmpset_32(p, cmpval, newval); + + fence(); + + return (res); +} + +static __inline int +atomic_fcmpset_rel_32(volatile uint32_t *p, uint32_t *cmpval, uint32_t newval) +{ + + fence(); + + return (atomic_fcmpset_32(p, cmpval, newval)); +} + static __inline uint32_t atomic_load_acq_32(volatile uint32_t *p) { uint32_t ret; ret = *p; fence(); return (ret); } static __inline void atomic_store_rel_32(volatile uint32_t *p, uint32_t val) { fence(); *p = val; } #define atomic_add_acq_int atomic_add_acq_32 #define atomic_clear_acq_int atomic_clear_acq_32 #define atomic_cmpset_acq_int atomic_cmpset_acq_32 +#define atomic_fcmpset_acq_int atomic_fcmpset_acq_32 #define atomic_load_acq_int atomic_load_acq_32 #define atomic_set_acq_int atomic_set_acq_32 #define atomic_subtract_acq_int atomic_subtract_acq_32 #define atomic_add_rel_int atomic_add_rel_32 #define atomic_clear_rel_int atomic_add_rel_32 #define atomic_cmpset_rel_int atomic_cmpset_rel_32 +#define atomic_fcmpset_rel_int atomic_fcmpset_rel_32 #define atomic_set_rel_int atomic_set_rel_32 #define atomic_subtract_rel_int atomic_subtract_rel_32 #define atomic_store_rel_int atomic_store_rel_32 static __inline void atomic_add_64(volatile uint64_t *p, uint64_t val) { __asm __volatile("amoadd.d zero, %1, %0" : "+A" (*p) : "r" (val) : "memory"); } static __inline void atomic_subtract_64(volatile uint64_t *p, uint64_t val) { __asm __volatile("amoadd.d zero, %1, %0" : "+A" (*p) : "r" (-val) : "memory"); } static __inline void atomic_set_64(volatile uint64_t *p, uint64_t val) { __asm __volatile("amoor.d zero, %1, %0" : "+A" (*p) : "r" (val) : "memory"); } static __inline void atomic_clear_64(volatile uint64_t *p, uint64_t val) { __asm __volatile("amoand.d zero, %1, %0" : "+A" (*p) : "r" (~val) : "memory"); } static __inline int atomic_cmpset_64(volatile uint64_t *p, uint64_t cmpval, uint64_t newval) { uint64_t tmp; int res; res = 0; __asm __volatile( "0:" "li %1, 1\n" /* Preset to fail */ "lr.d %0, %2\n" "bne %0, %z3, 1f\n" "sc.d %1, %z4, %2\n" "bnez %1, 0b\n" "1:" : "=&r" (tmp), "=&r" (res), "+A" (*p) : "rJ" (cmpval), "rJ" (newval) : "memory"); return (!res); } +static __inline int +atomic_fcmpset_64(volatile uint64_t *p, uint64_t *cmpval, uint64_t newval) +{ + uint64_t tmp; + int res; + + res = 0; + + __asm __volatile( + "0:" + "li %1, 1\n" /* Preset to fail */ + "lr.d %0, %2\n" /* Load old value */ + "bne %0, %z4, 1f\n" /* Compare */ + "sc.d %1, %z5, %2\n" /* Try to store new value */ + "j 2f\n" + "1:" + "sd %0, %3\n" /* Save old value */ + "2:" + : "=&r" (tmp), "=&r" (res), "+A" (*p), "+A" (*cmpval) + : "rJ" (*cmpval), "rJ" (newval) + : "memory"); + + return (!res); +} + static __inline uint64_t atomic_fetchadd_64(volatile uint64_t *p, uint64_t val) { uint64_t ret; __asm __volatile("amoadd.d %0, %2, %1" : "=&r" (ret), "+A" (*p) : "r" (val) : "memory"); return (ret); } static __inline uint64_t atomic_readandclear_64(volatile uint64_t *p) { uint64_t ret; uint64_t val; val = 0; __asm __volatile("amoswap.d %0, %2, %1" : "=&r"(ret), "+A" (*p) : "r" (val) : "memory"); return (ret); } static __inline uint32_t atomic_swap_32(volatile uint32_t *p, uint32_t val) { uint32_t old; __asm __volatile("amoswap.w %0, %2, %1" : "=&r"(old), "+A" (*p) : "r" (val) : "memory"); return (old); } static __inline uint64_t atomic_swap_64(volatile uint64_t *p, uint64_t val) { uint64_t old; __asm __volatile("amoswap.d %0, %2, %1" : "=&r"(old), "+A" (*p) : "r" (val) : "memory"); return (old); } #define atomic_add_long atomic_add_64 #define atomic_clear_long atomic_clear_64 #define atomic_cmpset_long atomic_cmpset_64 +#define atomic_fcmpset_long atomic_fcmpset_64 #define atomic_fetchadd_long atomic_fetchadd_64 #define atomic_readandclear_long atomic_readandclear_64 #define atomic_set_long atomic_set_64 #define atomic_subtract_long atomic_subtract_64 #define atomic_add_ptr atomic_add_64 #define atomic_clear_ptr atomic_clear_64 #define atomic_cmpset_ptr atomic_cmpset_64 +#define atomic_fcmpset_ptr atomic_fcmpset_64 #define atomic_fetchadd_ptr atomic_fetchadd_64 #define atomic_readandclear_ptr atomic_readandclear_64 #define atomic_set_ptr atomic_set_64 #define atomic_subtract_ptr atomic_subtract_64 ATOMIC_ACQ_REL(set, 64) ATOMIC_ACQ_REL(clear, 64) ATOMIC_ACQ_REL(add, 64) ATOMIC_ACQ_REL(subtract, 64) static __inline int atomic_cmpset_acq_64(volatile uint64_t *p, uint64_t cmpval, uint64_t newval) { int res; res = atomic_cmpset_64(p, cmpval, newval); fence(); return (res); } static __inline int atomic_cmpset_rel_64(volatile uint64_t *p, uint64_t cmpval, uint64_t newval) { fence(); return (atomic_cmpset_64(p, cmpval, newval)); } +static __inline int +atomic_fcmpset_acq_64(volatile uint64_t *p, uint64_t *cmpval, uint64_t newval) +{ + int res; + + res = atomic_fcmpset_64(p, cmpval, newval); + + fence(); + + return (res); +} + +static __inline int +atomic_fcmpset_rel_64(volatile uint64_t *p, uint64_t *cmpval, uint64_t newval) +{ + + fence(); + + return (atomic_fcmpset_64(p, cmpval, newval)); +} + static __inline uint64_t atomic_load_acq_64(volatile uint64_t *p) { uint64_t ret; ret = *p; fence(); return (ret); } static __inline void atomic_store_rel_64(volatile uint64_t *p, uint64_t val) { fence(); *p = val; } #define atomic_add_acq_long atomic_add_acq_64 #define atomic_clear_acq_long atomic_add_acq_64 #define atomic_cmpset_acq_long atomic_cmpset_acq_64 +#define atomic_fcmpset_acq_long atomic_fcmpset_acq_64 #define atomic_load_acq_long atomic_load_acq_64 #define atomic_set_acq_long atomic_set_acq_64 #define atomic_subtract_acq_long atomic_subtract_acq_64 #define atomic_add_acq_ptr atomic_add_acq_64 #define atomic_clear_acq_ptr atomic_add_acq_64 #define atomic_cmpset_acq_ptr atomic_cmpset_acq_64 +#define atomic_fcmpset_acq_ptr atomic_fcmpset_acq_64 #define atomic_load_acq_ptr atomic_load_acq_64 #define atomic_set_acq_ptr atomic_set_acq_64 #define atomic_subtract_acq_ptr atomic_subtract_acq_64 static __inline void atomic_thread_fence_acq(void) { fence(); } static __inline void atomic_thread_fence_rel(void) { fence(); } static __inline void atomic_thread_fence_acq_rel(void) { fence(); } static __inline void atomic_thread_fence_seq_cst(void) { fence(); } #define atomic_add_rel_long atomic_add_rel_64 #define atomic_clear_rel_long atomic_clear_rel_64 #define atomic_add_rel_long atomic_add_rel_64 #define atomic_clear_rel_long atomic_clear_rel_64 #define atomic_cmpset_rel_long atomic_cmpset_rel_64 +#define atomic_fcmpset_rel_long atomic_fcmpset_rel_64 #define atomic_set_rel_long atomic_set_rel_64 #define atomic_subtract_rel_long atomic_subtract_rel_64 #define atomic_store_rel_long atomic_store_rel_64 #define atomic_add_rel_ptr atomic_add_rel_64 #define atomic_clear_rel_ptr atomic_clear_rel_64 #define atomic_cmpset_rel_ptr atomic_cmpset_rel_64 +#define atomic_fcmpset_rel_ptr atomic_fcmpset_rel_64 #define atomic_set_rel_ptr atomic_set_rel_64 #define atomic_subtract_rel_ptr atomic_subtract_rel_64 #define atomic_store_rel_ptr atomic_store_rel_64 #endif /* _MACHINE_ATOMIC_H_ */ Index: stable/11/sys/sparc64/include/atomic.h =================================================================== --- stable/11/sys/sparc64/include/atomic.h (revision 315370) +++ stable/11/sys/sparc64/include/atomic.h (revision 315371) @@ -1,337 +1,371 @@ /*- * Copyright (c) 1998 Doug Rabson. * Copyright (c) 2001 Jake Burkholder. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: FreeBSD: src/sys/i386/include/atomic.h,v 1.20 2001/02/11 * $FreeBSD$ */ #ifndef _MACHINE_ATOMIC_H_ #define _MACHINE_ATOMIC_H_ #include #define mb() __asm__ __volatile__ ("membar #MemIssue": : :"memory") #define wmb() mb() #define rmb() mb() /* Userland needs different ASI's. */ #ifdef _KERNEL #define __ASI_ATOMIC ASI_N #else #define __ASI_ATOMIC ASI_P #endif /* * Various simple arithmetic on memory which is atomic in the presence * of interrupts and multiple processors. See atomic(9) for details. * Note that efficient hardware support exists only for the 32 and 64 * bit variants; the 8 and 16 bit versions are not provided and should * not be used in MI code. * * This implementation takes advantage of the fact that the sparc64 * cas instruction is both a load and a store. The loop is often coded * as follows: * * do { * expect = *p; * new = expect + 1; * } while (cas(p, expect, new) != expect); * * which performs an unnnecessary load on each iteration that the cas * operation fails. Modified as follows: * * expect = *p; * for (;;) { * new = expect + 1; * result = cas(p, expect, new); * if (result == expect) * break; * expect = result; * } * * the return value of cas is used to avoid the extra reload. * * We only include a memory barrier in the rel variants as in total store * order which we use for running the kernel and all of the userland atomic * loads and stores behave as if the were followed by a membar with a mask * of #LoadLoad | #LoadStore | #StoreStore. In order to be also sufficient * for use of relaxed memory ordering, the atomic_cas() in the acq variants * additionally would have to be followed by a membar #LoadLoad | #LoadStore. * Due to the suggested assembly syntax of the membar operands containing a * # character, they cannot be used in macros. The cmask and mmask bits thus * are hard coded in machine/cpufunc.h and used here through macros. * Hopefully the bit numbers won't change in the future. */ #define itype(sz) uint ## sz ## _t #define atomic_cas_32(p, e, s) casa((p), (e), (s), __ASI_ATOMIC) #define atomic_cas_64(p, e, s) casxa((p), (e), (s), __ASI_ATOMIC) #define atomic_cas(p, e, s, sz) \ atomic_cas_ ## sz((p), (e), (s)) #define atomic_cas_acq(p, e, s, sz) ({ \ itype(sz) v; \ v = atomic_cas((p), (e), (s), sz); \ __compiler_membar(); \ v; \ }) #define atomic_cas_rel(p, e, s, sz) ({ \ itype(sz) v; \ membar(LoadStore | StoreStore); \ v = atomic_cas((p), (e), (s), sz); \ v; \ }) #define atomic_op(p, op, v, sz) ({ \ itype(sz) e, r, s; \ for (e = *(volatile itype(sz) *)(p);; e = r) { \ s = e op (v); \ r = atomic_cas_ ## sz((p), e, s); \ if (r == e) \ break; \ } \ e; \ }) #define atomic_op_acq(p, op, v, sz) ({ \ itype(sz) t; \ t = atomic_op((p), op, (v), sz); \ __compiler_membar(); \ t; \ }) #define atomic_op_rel(p, op, v, sz) ({ \ itype(sz) t; \ membar(LoadStore | StoreStore); \ t = atomic_op((p), op, (v), sz); \ t; \ }) #define atomic_ld_acq(p, sz) ({ \ itype(sz) v; \ v = atomic_cas((p), 0, 0, sz); \ __compiler_membar(); \ v; \ }) #define atomic_ld_clear(p, sz) ({ \ itype(sz) e, r; \ for (e = *(volatile itype(sz) *)(p);; e = r) { \ r = atomic_cas((p), e, 0, sz); \ if (r == e) \ break; \ } \ e; \ }) #define atomic_st(p, v, sz) do { \ itype(sz) e, r; \ for (e = *(volatile itype(sz) *)(p);; e = r) { \ r = atomic_cas((p), e, (v), sz); \ if (r == e) \ break; \ } \ } while (0) #define atomic_st_acq(p, v, sz) do { \ atomic_st((p), (v), sz); \ __compiler_membar(); \ } while (0) #define atomic_st_rel(p, v, sz) do { \ membar(LoadStore | StoreStore); \ atomic_st((p), (v), sz); \ } while (0) #define ATOMIC_GEN(name, ptype, vtype, atype, sz) \ \ static __inline vtype \ atomic_add_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op((p), +, (v), sz)); \ } \ static __inline vtype \ atomic_add_acq_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op_acq((p), +, (v), sz)); \ } \ static __inline vtype \ atomic_add_rel_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op_rel((p), +, (v), sz)); \ } \ \ static __inline vtype \ atomic_clear_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op((p), &, ~(v), sz)); \ } \ static __inline vtype \ atomic_clear_acq_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op_acq((p), &, ~(v), sz)); \ } \ static __inline vtype \ atomic_clear_rel_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op_rel((p), &, ~(v), sz)); \ } \ \ static __inline int \ atomic_cmpset_ ## name(volatile ptype p, vtype e, vtype s) \ { \ return (((vtype)atomic_cas((p), (e), (s), sz)) == (e)); \ } \ static __inline int \ atomic_cmpset_acq_ ## name(volatile ptype p, vtype e, vtype s) \ { \ return (((vtype)atomic_cas_acq((p), (e), (s), sz)) == (e)); \ } \ static __inline int \ atomic_cmpset_rel_ ## name(volatile ptype p, vtype e, vtype s) \ { \ return (((vtype)atomic_cas_rel((p), (e), (s), sz)) == (e)); \ } \ \ +static __inline int \ +atomic_fcmpset_ ## name(volatile ptype p, vtype *ep, vtype s) \ +{ \ + vtype t; \ + \ + t = (vtype)atomic_cas((p), (*ep), (s), sz); \ + if (t == (*ep)) \ + return (1); \ + *ep = t; \ + return (0); \ +} \ +static __inline int \ +atomic_fcmpset_acq_ ## name(volatile ptype p, vtype *ep, vtype s) \ +{ \ + vtype t; \ + \ + t = (vtype)atomic_cas_acq((p), (*ep), (s), sz); \ + if (t == (*ep)) \ + return (1); \ + *ep = t; \ + return (0); \ +} \ +static __inline int \ +atomic_fcmpset_rel_ ## name(volatile ptype p, vtype *ep, vtype s) \ +{ \ + vtype t; \ + \ + t = (vtype)atomic_cas_rel((p), (*ep), (s), sz); \ + if (t == (*ep)) \ + return (1); \ + *ep = t; \ + return (0); \ +} \ + \ static __inline vtype \ atomic_load_ ## name(volatile ptype p) \ { \ return ((vtype)atomic_cas((p), 0, 0, sz)); \ } \ static __inline vtype \ atomic_load_acq_ ## name(volatile ptype p) \ { \ return ((vtype)atomic_cas_acq((p), 0, 0, sz)); \ } \ \ static __inline vtype \ atomic_readandclear_ ## name(volatile ptype p) \ { \ return ((vtype)atomic_ld_clear((p), sz)); \ } \ \ static __inline vtype \ atomic_set_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op((p), |, (v), sz)); \ } \ static __inline vtype \ atomic_set_acq_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op_acq((p), |, (v), sz)); \ } \ static __inline vtype \ atomic_set_rel_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op_rel((p), |, (v), sz)); \ } \ \ static __inline vtype \ atomic_subtract_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op((p), -, (v), sz)); \ } \ static __inline vtype \ atomic_subtract_acq_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op_acq((p), -, (v), sz)); \ } \ static __inline vtype \ atomic_subtract_rel_ ## name(volatile ptype p, atype v) \ { \ return ((vtype)atomic_op_rel((p), -, (v), sz)); \ } \ \ static __inline void \ atomic_store_acq_ ## name(volatile ptype p, vtype v) \ { \ atomic_st_acq((p), (v), sz); \ } \ static __inline void \ atomic_store_rel_ ## name(volatile ptype p, vtype v) \ { \ atomic_st_rel((p), (v), sz); \ } static __inline void atomic_thread_fence_acq(void) { __compiler_membar(); } static __inline void atomic_thread_fence_rel(void) { __compiler_membar(); } static __inline void atomic_thread_fence_acq_rel(void) { __compiler_membar(); } static __inline void atomic_thread_fence_seq_cst(void) { membar(LoadLoad | LoadStore | StoreStore | StoreLoad); } ATOMIC_GEN(int, u_int *, u_int, u_int, 32); ATOMIC_GEN(32, uint32_t *, uint32_t, uint32_t, 32); ATOMIC_GEN(long, u_long *, u_long, u_long, 64); ATOMIC_GEN(64, uint64_t *, uint64_t, uint64_t, 64); ATOMIC_GEN(ptr, uintptr_t *, uintptr_t, uintptr_t, 64); #define atomic_fetchadd_int atomic_add_int #define atomic_fetchadd_32 atomic_add_32 #define atomic_fetchadd_long atomic_add_long #define atomic_fetchadd_64 atomic_add_64 #undef ATOMIC_GEN #undef atomic_cas #undef atomic_cas_acq #undef atomic_cas_rel #undef atomic_op #undef atomic_op_acq #undef atomic_op_rel #undef atomic_ld_acq #undef atomic_ld_clear #undef atomic_st #undef atomic_st_acq #undef atomic_st_rel #endif /* !_MACHINE_ATOMIC_H_ */ Index: stable/11 =================================================================== --- stable/11 (revision 315370) +++ stable/11 (revision 315371) Property changes on: stable/11 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r311169,311898,312925,312973,312975,313007,313040,313080,313254,313341