X-Git-Url: https://git.liburcu.org/?a=blobdiff_plain;f=include%2Furcu%2Fuatomic%2Fx86.h;h=efc67bb62af9689cdf23e219c7e7ccabaadc23a2;hb=81562423503c71c206a3303cf53e3b11f8ca034b;hp=129a2f5670385880870912235a14a63a6d0ea99e;hpb=6893800a4d1cc14dff0395ddcd660a5138db183d;p=urcu.git diff --git a/include/urcu/uatomic/x86.h b/include/urcu/uatomic/x86.h index 129a2f5..efc67bb 100644 --- a/include/urcu/uatomic/x86.h +++ b/include/urcu/uatomic/x86.h @@ -20,6 +20,7 @@ * Boehm-Demers-Weiser conservative garbage collector. */ +#include #include #include @@ -34,10 +35,21 @@ extern "C" { * Derived from AO_compare_and_swap() and AO_test_and_set_full(). */ -struct __uatomic_dummy { - unsigned long v[10]; -}; -#define __hp(x) ((struct __uatomic_dummy *)(x)) +/* + * The __hp() macro casts the void pointer @x to a pointer to a structure + * containing an array of char of the specified size. This allows passing the + * @addr arguments of the following inline functions as "m" and "+m" operands + * to the assembly. The @size parameter should be a constant to support + * compilers such as clang which do not support VLA. Create typedefs because + * C++ does not allow types be defined in casts. + */ + +typedef struct { char v[1]; } __hp_1; +typedef struct { char v[2]; } __hp_2; +typedef struct { char v[4]; } __hp_4; +typedef struct { char v[8]; } __hp_8; + +#define __hp(size, x) ((__hp_##size *)(x)) #define _uatomic_set(addr, v) ((void) CMM_STORE_SHARED(*(addr), (v))) @@ -54,7 +66,7 @@ unsigned long __uatomic_cmpxchg(void *addr, unsigned long old, __asm__ __volatile__( "lock; cmpxchgb %2, %1" - : "+a"(result), "+m"(*__hp(addr)) + : "+a"(result), "+m"(*__hp(1, addr)) : "q"((unsigned char)_new) : "memory"); return result; @@ -65,7 +77,7 @@ unsigned long __uatomic_cmpxchg(void *addr, unsigned long old, __asm__ __volatile__( "lock; cmpxchgw %2, %1" - : "+a"(result), "+m"(*__hp(addr)) + : "+a"(result), "+m"(*__hp(2, addr)) : "r"((unsigned short)_new) : "memory"); return result; @@ -76,7 +88,7 @@ unsigned long __uatomic_cmpxchg(void *addr, unsigned long old, __asm__ __volatile__( "lock; cmpxchgl %2, %1" - : "+a"(result), "+m"(*__hp(addr)) + : "+a"(result), "+m"(*__hp(4, addr)) : "r"((unsigned int)_new) : "memory"); return result; @@ -88,7 +100,7 @@ unsigned long __uatomic_cmpxchg(void *addr, unsigned long old, __asm__ __volatile__( "lock; cmpxchgq %2, %1" - : "+a"(result), "+m"(*__hp(addr)) + : "+a"(result), "+m"(*__hp(8, addr)) : "r"((unsigned long)_new) : "memory"); return result; @@ -121,7 +133,7 @@ unsigned long __uatomic_exchange(void *addr, unsigned long val, int len) unsigned char result; __asm__ __volatile__( "xchgb %0, %1" - : "=q"(result), "+m"(*__hp(addr)) + : "=q"(result), "+m"(*__hp(1, addr)) : "0" ((unsigned char)val) : "memory"); return result; @@ -131,7 +143,7 @@ unsigned long __uatomic_exchange(void *addr, unsigned long val, int len) unsigned short result; __asm__ __volatile__( "xchgw %0, %1" - : "=r"(result), "+m"(*__hp(addr)) + : "=r"(result), "+m"(*__hp(2, addr)) : "0" ((unsigned short)val) : "memory"); return result; @@ -141,7 +153,7 @@ unsigned long __uatomic_exchange(void *addr, unsigned long val, int len) unsigned int result; __asm__ __volatile__( "xchgl %0, %1" - : "=r"(result), "+m"(*__hp(addr)) + : "=r"(result), "+m"(*__hp(4, addr)) : "0" ((unsigned int)val) : "memory"); return result; @@ -152,7 +164,7 @@ unsigned long __uatomic_exchange(void *addr, unsigned long val, int len) unsigned long result; __asm__ __volatile__( "xchgq %0, %1" - : "=r"(result), "+m"(*__hp(addr)) + : "=r"(result), "+m"(*__hp(8, addr)) : "0" ((unsigned long)val) : "memory"); return result; @@ -185,7 +197,7 @@ unsigned long __uatomic_add_return(void *addr, unsigned long val, __asm__ __volatile__( "lock; xaddb %1, %0" - : "+m"(*__hp(addr)), "+q" (result) + : "+m"(*__hp(1, addr)), "+q" (result) : : "memory"); return result + (unsigned char)val; @@ -196,7 +208,7 @@ unsigned long __uatomic_add_return(void *addr, unsigned long val, __asm__ __volatile__( "lock; xaddw %1, %0" - : "+m"(*__hp(addr)), "+r" (result) + : "+m"(*__hp(2, addr)), "+r" (result) : : "memory"); return result + (unsigned short)val; @@ -207,7 +219,7 @@ unsigned long __uatomic_add_return(void *addr, unsigned long val, __asm__ __volatile__( "lock; xaddl %1, %0" - : "+m"(*__hp(addr)), "+r" (result) + : "+m"(*__hp(4, addr)), "+r" (result) : : "memory"); return result + (unsigned int)val; @@ -219,7 +231,7 @@ unsigned long __uatomic_add_return(void *addr, unsigned long val, __asm__ __volatile__( "lock; xaddq %1, %0" - : "+m"(*__hp(addr)), "+r" (result) + : "+m"(*__hp(8, addr)), "+r" (result) : : "memory"); return result + (unsigned long)val; @@ -249,7 +261,7 @@ void __uatomic_and(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; andb %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(1, addr)) : "iq" ((unsigned char)val) : "memory"); return; @@ -258,7 +270,7 @@ void __uatomic_and(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; andw %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(2, addr)) : "ir" ((unsigned short)val) : "memory"); return; @@ -267,7 +279,7 @@ void __uatomic_and(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; andl %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(4, addr)) : "ir" ((unsigned int)val) : "memory"); return; @@ -277,7 +289,7 @@ void __uatomic_and(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; andq %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(8, addr)) : "er" ((unsigned long)val) : "memory"); return; @@ -305,7 +317,7 @@ void __uatomic_or(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; orb %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(1, addr)) : "iq" ((unsigned char)val) : "memory"); return; @@ -314,7 +326,7 @@ void __uatomic_or(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; orw %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(2, addr)) : "ir" ((unsigned short)val) : "memory"); return; @@ -323,7 +335,7 @@ void __uatomic_or(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; orl %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(4, addr)) : "ir" ((unsigned int)val) : "memory"); return; @@ -333,7 +345,7 @@ void __uatomic_or(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; orq %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(8, addr)) : "er" ((unsigned long)val) : "memory"); return; @@ -361,7 +373,7 @@ void __uatomic_add(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; addb %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(1, addr)) : "iq" ((unsigned char)val) : "memory"); return; @@ -370,7 +382,7 @@ void __uatomic_add(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; addw %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(2, addr)) : "ir" ((unsigned short)val) : "memory"); return; @@ -379,7 +391,7 @@ void __uatomic_add(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; addl %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(4, addr)) : "ir" ((unsigned int)val) : "memory"); return; @@ -389,7 +401,7 @@ void __uatomic_add(void *addr, unsigned long val, int len) { __asm__ __volatile__( "lock; addq %1, %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(8, addr)) : "er" ((unsigned long)val) : "memory"); return; @@ -418,7 +430,7 @@ void __uatomic_inc(void *addr, int len) { __asm__ __volatile__( "lock; incb %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(1, addr)) : : "memory"); return; @@ -427,7 +439,7 @@ void __uatomic_inc(void *addr, int len) { __asm__ __volatile__( "lock; incw %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(2, addr)) : : "memory"); return; @@ -436,7 +448,7 @@ void __uatomic_inc(void *addr, int len) { __asm__ __volatile__( "lock; incl %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(4, addr)) : : "memory"); return; @@ -446,7 +458,7 @@ void __uatomic_inc(void *addr, int len) { __asm__ __volatile__( "lock; incq %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(8, addr)) : : "memory"); return; @@ -471,7 +483,7 @@ void __uatomic_dec(void *addr, int len) { __asm__ __volatile__( "lock; decb %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(1, addr)) : : "memory"); return; @@ -480,7 +492,7 @@ void __uatomic_dec(void *addr, int len) { __asm__ __volatile__( "lock; decw %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(2, addr)) : : "memory"); return; @@ -489,7 +501,7 @@ void __uatomic_dec(void *addr, int len) { __asm__ __volatile__( "lock; decl %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(4, addr)) : : "memory"); return; @@ -499,7 +511,7 @@ void __uatomic_dec(void *addr, int len) { __asm__ __volatile__( "lock; decq %0" - : "=m"(*__hp(addr)) + : "=m"(*__hp(8, addr)) : : "memory"); return;