linux/arch/s390/include/asm/atomic.h
Martin Schwidefsky 39475179d4 [S390] Improve code generated by atomic operations.
Git commit ea43546750 changed the
definition of atomic_t and atomic64_t for s390 by adding the volatile
modifier to the counter field. This has an unfortunate side effect
with newer versions of the gcc. The typeof operator now picks up the
volatile modifier from the expression. This causes the compiler to
think that it has to store the two temporary variable old_val and
new_val in the __CS_LOOP for the different atomic operations to the
stack as the variables are now volatile. Both stores are superfluous.

The hack to replace typeof(ptr->counter) with int in __CS_LOOP and
and long long in __CSG_LOOP avoids the two stores. A better solution
would be to drop the volatile from the counter field of the atomic_t
and atomic64_t definition. But that is a touchy subject ..

Cc: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2009-12-07 12:51:36 +01:00

373 lines
8.5 KiB
C

#ifndef __ARCH_S390_ATOMIC__
#define __ARCH_S390_ATOMIC__
/*
* Copyright 1999,2009 IBM Corp.
* Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>,
* Denis Joseph Barrow,
* Arnd Bergmann <arndb@de.ibm.com>,
*
* Atomic operations that C can't guarantee us.
* Useful for resource counting etc.
* s390 uses 'Compare And Swap' for atomicity in SMP enviroment.
*
*/
#include <linux/compiler.h>
#include <linux/types.h>
#define ATOMIC_INIT(i) { (i) }
#if __GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ > 2)
#define __CS_LOOP(ptr, op_val, op_string) ({ \
int old_val, new_val; \
asm volatile( \
" l %0,%2\n" \
"0: lr %1,%0\n" \
op_string " %1,%3\n" \
" cs %0,%1,%2\n" \
" jl 0b" \
: "=&d" (old_val), "=&d" (new_val), \
"=Q" (((atomic_t *)(ptr))->counter) \
: "d" (op_val), "Q" (((atomic_t *)(ptr))->counter) \
: "cc", "memory"); \
new_val; \
})
#else /* __GNUC__ */
#define __CS_LOOP(ptr, op_val, op_string) ({ \
int old_val, new_val; \
asm volatile( \
" l %0,0(%3)\n" \
"0: lr %1,%0\n" \
op_string " %1,%4\n" \
" cs %0,%1,0(%3)\n" \
" jl 0b" \
: "=&d" (old_val), "=&d" (new_val), \
"=m" (((atomic_t *)(ptr))->counter) \
: "a" (ptr), "d" (op_val), \
"m" (((atomic_t *)(ptr))->counter) \
: "cc", "memory"); \
new_val; \
})
#endif /* __GNUC__ */
static inline int atomic_read(const atomic_t *v)
{
barrier();
return v->counter;
}
static inline void atomic_set(atomic_t *v, int i)
{
v->counter = i;
barrier();
}
static inline int atomic_add_return(int i, atomic_t *v)
{
return __CS_LOOP(v, i, "ar");
}
#define atomic_add(_i, _v) atomic_add_return(_i, _v)
#define atomic_add_negative(_i, _v) (atomic_add_return(_i, _v) < 0)
#define atomic_inc(_v) atomic_add_return(1, _v)
#define atomic_inc_return(_v) atomic_add_return(1, _v)
#define atomic_inc_and_test(_v) (atomic_add_return(1, _v) == 0)
static inline int atomic_sub_return(int i, atomic_t *v)
{
return __CS_LOOP(v, i, "sr");
}
#define atomic_sub(_i, _v) atomic_sub_return(_i, _v)
#define atomic_sub_and_test(_i, _v) (atomic_sub_return(_i, _v) == 0)
#define atomic_dec(_v) atomic_sub_return(1, _v)
#define atomic_dec_return(_v) atomic_sub_return(1, _v)
#define atomic_dec_and_test(_v) (atomic_sub_return(1, _v) == 0)
static inline void atomic_clear_mask(unsigned long mask, atomic_t *v)
{
__CS_LOOP(v, ~mask, "nr");
}
static inline void atomic_set_mask(unsigned long mask, atomic_t *v)
{
__CS_LOOP(v, mask, "or");
}
#define atomic_xchg(v, new) (xchg(&((v)->counter), new))
static inline int atomic_cmpxchg(atomic_t *v, int old, int new)
{
#if __GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ > 2)
asm volatile(
" cs %0,%2,%1"
: "+d" (old), "=Q" (v->counter)
: "d" (new), "Q" (v->counter)
: "cc", "memory");
#else /* __GNUC__ */
asm volatile(
" cs %0,%3,0(%2)"
: "+d" (old), "=m" (v->counter)
: "a" (v), "d" (new), "m" (v->counter)
: "cc", "memory");
#endif /* __GNUC__ */
return old;
}
static inline int atomic_add_unless(atomic_t *v, int a, int u)
{
int c, old;
c = atomic_read(v);
for (;;) {
if (unlikely(c == u))
break;
old = atomic_cmpxchg(v, c, c + a);
if (likely(old == c))
break;
c = old;
}
return c != u;
}
#define atomic_inc_not_zero(v) atomic_add_unless((v), 1, 0)
#undef __CS_LOOP
#define ATOMIC64_INIT(i) { (i) }
#ifdef CONFIG_64BIT
#if __GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ > 2)
#define __CSG_LOOP(ptr, op_val, op_string) ({ \
long long old_val, new_val; \
asm volatile( \
" lg %0,%2\n" \
"0: lgr %1,%0\n" \
op_string " %1,%3\n" \
" csg %0,%1,%2\n" \
" jl 0b" \
: "=&d" (old_val), "=&d" (new_val), \
"=Q" (((atomic_t *)(ptr))->counter) \
: "d" (op_val), "Q" (((atomic_t *)(ptr))->counter) \
: "cc", "memory"); \
new_val; \
})
#else /* __GNUC__ */
#define __CSG_LOOP(ptr, op_val, op_string) ({ \
long long old_val, new_val; \
asm volatile( \
" lg %0,0(%3)\n" \
"0: lgr %1,%0\n" \
op_string " %1,%4\n" \
" csg %0,%1,0(%3)\n" \
" jl 0b" \
: "=&d" (old_val), "=&d" (new_val), \
"=m" (((atomic_t *)(ptr))->counter) \
: "a" (ptr), "d" (op_val), \
"m" (((atomic_t *)(ptr))->counter) \
: "cc", "memory"); \
new_val; \
})
#endif /* __GNUC__ */
static inline long long atomic64_read(const atomic64_t *v)
{
barrier();
return v->counter;
}
static inline void atomic64_set(atomic64_t *v, long long i)
{
v->counter = i;
barrier();
}
static inline long long atomic64_add_return(long long i, atomic64_t *v)
{
return __CSG_LOOP(v, i, "agr");
}
static inline long long atomic64_sub_return(long long i, atomic64_t *v)
{
return __CSG_LOOP(v, i, "sgr");
}
static inline void atomic64_clear_mask(unsigned long mask, atomic64_t *v)
{
__CSG_LOOP(v, ~mask, "ngr");
}
static inline void atomic64_set_mask(unsigned long mask, atomic64_t *v)
{
__CSG_LOOP(v, mask, "ogr");
}
#define atomic64_xchg(v, new) (xchg(&((v)->counter), new))
static inline long long atomic64_cmpxchg(atomic64_t *v,
long long old, long long new)
{
#if __GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ > 2)
asm volatile(
" csg %0,%2,%1"
: "+d" (old), "=Q" (v->counter)
: "d" (new), "Q" (v->counter)
: "cc", "memory");
#else /* __GNUC__ */
asm volatile(
" csg %0,%3,0(%2)"
: "+d" (old), "=m" (v->counter)
: "a" (v), "d" (new), "m" (v->counter)
: "cc", "memory");
#endif /* __GNUC__ */
return old;
}
#undef __CSG_LOOP
#else /* CONFIG_64BIT */
typedef struct {
long long counter;
} atomic64_t;
static inline long long atomic64_read(const atomic64_t *v)
{
register_pair rp;
asm volatile(
" lm %0,%N0,0(%1)"
: "=&d" (rp)
: "a" (&v->counter), "m" (v->counter)
);
return rp.pair;
}
static inline void atomic64_set(atomic64_t *v, long long i)
{
register_pair rp = {.pair = i};
asm volatile(
" stm %1,%N1,0(%2)"
: "=m" (v->counter)
: "d" (rp), "a" (&v->counter)
);
}
static inline long long atomic64_xchg(atomic64_t *v, long long new)
{
register_pair rp_new = {.pair = new};
register_pair rp_old;
asm volatile(
" lm %0,%N0,0(%2)\n"
"0: cds %0,%3,0(%2)\n"
" jl 0b\n"
: "=&d" (rp_old), "+m" (v->counter)
: "a" (&v->counter), "d" (rp_new)
: "cc");
return rp_old.pair;
}
static inline long long atomic64_cmpxchg(atomic64_t *v,
long long old, long long new)
{
register_pair rp_old = {.pair = old};
register_pair rp_new = {.pair = new};
asm volatile(
" cds %0,%3,0(%2)"
: "+&d" (rp_old), "+m" (v->counter)
: "a" (&v->counter), "d" (rp_new)
: "cc");
return rp_old.pair;
}
static inline long long atomic64_add_return(long long i, atomic64_t *v)
{
long long old, new;
do {
old = atomic64_read(v);
new = old + i;
} while (atomic64_cmpxchg(v, old, new) != old);
return new;
}
static inline long long atomic64_sub_return(long long i, atomic64_t *v)
{
long long old, new;
do {
old = atomic64_read(v);
new = old - i;
} while (atomic64_cmpxchg(v, old, new) != old);
return new;
}
static inline void atomic64_set_mask(unsigned long long mask, atomic64_t *v)
{
long long old, new;
do {
old = atomic64_read(v);
new = old | mask;
} while (atomic64_cmpxchg(v, old, new) != old);
}
static inline void atomic64_clear_mask(unsigned long long mask, atomic64_t *v)
{
long long old, new;
do {
old = atomic64_read(v);
new = old & mask;
} while (atomic64_cmpxchg(v, old, new) != old);
}
#endif /* CONFIG_64BIT */
static inline int atomic64_add_unless(atomic64_t *v, long long a, long long u)
{
long long c, old;
c = atomic64_read(v);
for (;;) {
if (unlikely(c == u))
break;
old = atomic64_cmpxchg(v, c, c + a);
if (likely(old == c))
break;
c = old;
}
return c != u;
}
#define atomic64_add(_i, _v) atomic64_add_return(_i, _v)
#define atomic64_add_negative(_i, _v) (atomic64_add_return(_i, _v) < 0)
#define atomic64_inc(_v) atomic64_add_return(1, _v)
#define atomic64_inc_return(_v) atomic64_add_return(1, _v)
#define atomic64_inc_and_test(_v) (atomic64_add_return(1, _v) == 0)
#define atomic64_sub(_i, _v) atomic64_sub_return(_i, _v)
#define atomic64_sub_and_test(_i, _v) (atomic64_sub_return(_i, _v) == 0)
#define atomic64_dec(_v) atomic64_sub_return(1, _v)
#define atomic64_dec_return(_v) atomic64_sub_return(1, _v)
#define atomic64_dec_and_test(_v) (atomic64_sub_return(1, _v) == 0)
#define atomic64_inc_not_zero(v) atomic64_add_unless((v), 1, 0)
#define smp_mb__before_atomic_dec() smp_mb()
#define smp_mb__after_atomic_dec() smp_mb()
#define smp_mb__before_atomic_inc() smp_mb()
#define smp_mb__after_atomic_inc() smp_mb()
#include <asm-generic/atomic-long.h>
#endif /* __ARCH_S390_ATOMIC__ */