linux_dsm_epyc7002/arch/arm/include/asm/atomic.h

542 lines
13 KiB
C
Raw Normal View History

/*
* arch/arm/include/asm/atomic.h
*
* Copyright (C) 1996 Russell King.
* Copyright (C) 2002 Deep Blue Solutions Ltd.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
#ifndef __ASM_ARM_ATOMIC_H
#define __ASM_ARM_ATOMIC_H
#include <linux/compiler.h>
#include <linux/prefetch.h>
#include <linux/types.h>
#include <linux/irqflags.h>
#include <asm/barrier.h>
#include <asm/cmpxchg.h>
#define ATOMIC_INIT(i) { (i) }
#ifdef __KERNEL__
/*
* On ARM, ordinary assignment (str instruction) doesn't clear the local
* strex/ldrex monitor on some implementations. The reason we can use it for
* atomic_set() is the clrex or dummy strex done on every exception return.
*/
#define atomic_read(v) READ_ONCE((v)->counter)
#define atomic_set(v,i) WRITE_ONCE(((v)->counter), (i))
#if __LINUX_ARM_ARCH__ >= 6
/*
* ARMv6 UP and SMP safe atomic ops. We use load exclusive and
* store exclusive to ensure that these are atomic. We may loop
* to ensure that the update happens.
*/
#define ATOMIC_OP(op, c_op, asm_op) \
static inline void atomic_##op(int i, atomic_t *v) \
{ \
unsigned long tmp; \
int result; \
\
prefetchw(&v->counter); \
__asm__ __volatile__("@ atomic_" #op "\n" \
"1: ldrex %0, [%3]\n" \
" " #asm_op " %0, %0, %4\n" \
" strex %1, %0, [%3]\n" \
" teq %1, #0\n" \
" bne 1b" \
: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter) \
: "r" (&v->counter), "Ir" (i) \
: "cc"); \
} \
#define ATOMIC_OP_RETURN(op, c_op, asm_op) \
static inline int atomic_##op##_return_relaxed(int i, atomic_t *v) \
{ \
unsigned long tmp; \
int result; \
\
prefetchw(&v->counter); \
\
__asm__ __volatile__("@ atomic_" #op "_return\n" \
"1: ldrex %0, [%3]\n" \
" " #asm_op " %0, %0, %4\n" \
" strex %1, %0, [%3]\n" \
" teq %1, #0\n" \
" bne 1b" \
: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter) \
: "r" (&v->counter), "Ir" (i) \
: "cc"); \
\
return result; \
}
#define ATOMIC_FETCH_OP(op, c_op, asm_op) \
static inline int atomic_fetch_##op##_relaxed(int i, atomic_t *v) \
{ \
unsigned long tmp; \
int result, val; \
\
prefetchw(&v->counter); \
\
__asm__ __volatile__("@ atomic_fetch_" #op "\n" \
"1: ldrex %0, [%4]\n" \
" " #asm_op " %1, %0, %5\n" \
" strex %2, %1, [%4]\n" \
" teq %2, #0\n" \
" bne 1b" \
: "=&r" (result), "=&r" (val), "=&r" (tmp), "+Qo" (v->counter) \
: "r" (&v->counter), "Ir" (i) \
: "cc"); \
\
return result; \
}
#define atomic_add_return_relaxed atomic_add_return_relaxed
#define atomic_sub_return_relaxed atomic_sub_return_relaxed
#define atomic_fetch_add_relaxed atomic_fetch_add_relaxed
#define atomic_fetch_sub_relaxed atomic_fetch_sub_relaxed
#define atomic_fetch_and_relaxed atomic_fetch_and_relaxed
#define atomic_fetch_andnot_relaxed atomic_fetch_andnot_relaxed
#define atomic_fetch_or_relaxed atomic_fetch_or_relaxed
#define atomic_fetch_xor_relaxed atomic_fetch_xor_relaxed
static inline int atomic_cmpxchg_relaxed(atomic_t *ptr, int old, int new)
{
int oldval;
unsigned long res;
prefetchw(&ptr->counter);
do {
__asm__ __volatile__("@ atomic_cmpxchg\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
"ldrex %1, [%3]\n"
"mov %0, #0\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
"teq %1, %4\n"
"strexeq %0, %5, [%3]\n"
: "=&r" (res), "=&r" (oldval), "+Qo" (ptr->counter)
: "r" (&ptr->counter), "Ir" (old), "r" (new)
: "cc");
} while (res);
return oldval;
}
#define atomic_cmpxchg_relaxed atomic_cmpxchg_relaxed
static inline int __atomic_add_unless(atomic_t *v, int a, int u)
{
int oldval, newval;
unsigned long tmp;
smp_mb();
prefetchw(&v->counter);
__asm__ __volatile__ ("@ atomic_add_unless\n"
"1: ldrex %0, [%4]\n"
" teq %0, %5\n"
" beq 2f\n"
" add %1, %0, %6\n"
" strex %2, %1, [%4]\n"
" teq %2, #0\n"
" bne 1b\n"
"2:"
: "=&r" (oldval), "=&r" (newval), "=&r" (tmp), "+Qo" (v->counter)
: "r" (&v->counter), "r" (u), "r" (a)
: "cc");
if (oldval != u)
smp_mb();
return oldval;
}
#else /* ARM_ARCH_6 */
#ifdef CONFIG_SMP
#error SMP not supported on pre-ARMv6 CPUs
#endif
#define ATOMIC_OP(op, c_op, asm_op) \
static inline void atomic_##op(int i, atomic_t *v) \
{ \
unsigned long flags; \
\
raw_local_irq_save(flags); \
v->counter c_op i; \
raw_local_irq_restore(flags); \
} \
#define ATOMIC_OP_RETURN(op, c_op, asm_op) \
static inline int atomic_##op##_return(int i, atomic_t *v) \
{ \
unsigned long flags; \
int val; \
\
raw_local_irq_save(flags); \
v->counter c_op i; \
val = v->counter; \
raw_local_irq_restore(flags); \
\
return val; \
}
#define ATOMIC_FETCH_OP(op, c_op, asm_op) \
static inline int atomic_fetch_##op(int i, atomic_t *v) \
{ \
unsigned long flags; \
int val; \
\
raw_local_irq_save(flags); \
val = v->counter; \
v->counter c_op i; \
raw_local_irq_restore(flags); \
\
return val; \
}
static inline int atomic_cmpxchg(atomic_t *v, int old, int new)
{
int ret;
unsigned long flags;
raw_local_irq_save(flags);
ret = v->counter;
if (likely(ret == old))
v->counter = new;
raw_local_irq_restore(flags);
return ret;
}
static inline int __atomic_add_unless(atomic_t *v, int a, int u)
{
int c, old;
c = atomic_read(v);
while (c != u && (old = atomic_cmpxchg((v), c, c + a)) != c)
c = old;
return c;
}
#endif /* __LINUX_ARM_ARCH__ */
#define ATOMIC_OPS(op, c_op, asm_op) \
ATOMIC_OP(op, c_op, asm_op) \
ATOMIC_OP_RETURN(op, c_op, asm_op) \
ATOMIC_FETCH_OP(op, c_op, asm_op)
ATOMIC_OPS(add, +=, add)
ATOMIC_OPS(sub, -=, sub)
#define atomic_andnot atomic_andnot
#undef ATOMIC_OPS
#define ATOMIC_OPS(op, c_op, asm_op) \
ATOMIC_OP(op, c_op, asm_op) \
ATOMIC_FETCH_OP(op, c_op, asm_op)
ATOMIC_OPS(and, &=, and)
ATOMIC_OPS(andnot, &= ~, bic)
ATOMIC_OPS(or, |=, orr)
ATOMIC_OPS(xor, ^=, eor)
#undef ATOMIC_OPS
#undef ATOMIC_FETCH_OP
#undef ATOMIC_OP_RETURN
#undef ATOMIC_OP
#define atomic_xchg(v, new) (xchg(&((v)->counter), new))
#define atomic_inc(v) atomic_add(1, v)
#define atomic_dec(v) atomic_sub(1, v)
#define atomic_inc_and_test(v) (atomic_add_return(1, v) == 0)
#define atomic_dec_and_test(v) (atomic_sub_return(1, v) == 0)
#define atomic_inc_return_relaxed(v) (atomic_add_return_relaxed(1, v))
#define atomic_dec_return_relaxed(v) (atomic_sub_return_relaxed(1, v))
#define atomic_sub_and_test(i, v) (atomic_sub_return(i, v) == 0)
#define atomic_add_negative(i,v) (atomic_add_return(i, v) < 0)
#ifndef CONFIG_GENERIC_ATOMIC64
typedef struct {
long long counter;
} atomic64_t;
#define ATOMIC64_INIT(i) { (i) }
#ifdef CONFIG_ARM_LPAE
static inline long long atomic64_read(const atomic64_t *v)
{
long long result;
__asm__ __volatile__("@ atomic64_read\n"
" ldrd %0, %H0, [%1]"
: "=&r" (result)
: "r" (&v->counter), "Qo" (v->counter)
);
return result;
}
static inline void atomic64_set(atomic64_t *v, long long i)
{
__asm__ __volatile__("@ atomic64_set\n"
" strd %2, %H2, [%1]"
: "=Qo" (v->counter)
: "r" (&v->counter), "r" (i)
);
}
#else
static inline long long atomic64_read(const atomic64_t *v)
{
long long result;
__asm__ __volatile__("@ atomic64_read\n"
" ldrexd %0, %H0, [%1]"
: "=&r" (result)
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
: "r" (&v->counter), "Qo" (v->counter)
);
return result;
}
static inline void atomic64_set(atomic64_t *v, long long i)
{
long long tmp;
prefetchw(&v->counter);
__asm__ __volatile__("@ atomic64_set\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
"1: ldrexd %0, %H0, [%2]\n"
" strexd %0, %3, %H3, [%2]\n"
" teq %0, #0\n"
" bne 1b"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
: "=&r" (tmp), "=Qo" (v->counter)
: "r" (&v->counter), "r" (i)
: "cc");
}
#endif
#define ATOMIC64_OP(op, op1, op2) \
static inline void atomic64_##op(long long i, atomic64_t *v) \
{ \
long long result; \
unsigned long tmp; \
\
prefetchw(&v->counter); \
__asm__ __volatile__("@ atomic64_" #op "\n" \
"1: ldrexd %0, %H0, [%3]\n" \
" " #op1 " %Q0, %Q0, %Q4\n" \
" " #op2 " %R0, %R0, %R4\n" \
" strexd %1, %0, %H0, [%3]\n" \
" teq %1, #0\n" \
" bne 1b" \
: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter) \
: "r" (&v->counter), "r" (i) \
: "cc"); \
} \
#define ATOMIC64_OP_RETURN(op, op1, op2) \
static inline long long \
atomic64_##op##_return_relaxed(long long i, atomic64_t *v) \
{ \
long long result; \
unsigned long tmp; \
\
prefetchw(&v->counter); \
\
__asm__ __volatile__("@ atomic64_" #op "_return\n" \
"1: ldrexd %0, %H0, [%3]\n" \
" " #op1 " %Q0, %Q0, %Q4\n" \
" " #op2 " %R0, %R0, %R4\n" \
" strexd %1, %0, %H0, [%3]\n" \
" teq %1, #0\n" \
" bne 1b" \
: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter) \
: "r" (&v->counter), "r" (i) \
: "cc"); \
\
return result; \
}
#define ATOMIC64_FETCH_OP(op, op1, op2) \
static inline long long \
atomic64_fetch_##op##_relaxed(long long i, atomic64_t *v) \
{ \
long long result, val; \
unsigned long tmp; \
\
prefetchw(&v->counter); \
\
__asm__ __volatile__("@ atomic64_fetch_" #op "\n" \
"1: ldrexd %0, %H0, [%4]\n" \
" " #op1 " %Q1, %Q0, %Q5\n" \
" " #op2 " %R1, %R0, %R5\n" \
" strexd %2, %1, %H1, [%4]\n" \
" teq %2, #0\n" \
" bne 1b" \
: "=&r" (result), "=&r" (val), "=&r" (tmp), "+Qo" (v->counter) \
: "r" (&v->counter), "r" (i) \
: "cc"); \
\
return result; \
}
#define ATOMIC64_OPS(op, op1, op2) \
ATOMIC64_OP(op, op1, op2) \
ATOMIC64_OP_RETURN(op, op1, op2) \
ATOMIC64_FETCH_OP(op, op1, op2)
ATOMIC64_OPS(add, adds, adc)
ATOMIC64_OPS(sub, subs, sbc)
#define atomic64_add_return_relaxed atomic64_add_return_relaxed
#define atomic64_sub_return_relaxed atomic64_sub_return_relaxed
#define atomic64_fetch_add_relaxed atomic64_fetch_add_relaxed
#define atomic64_fetch_sub_relaxed atomic64_fetch_sub_relaxed
#undef ATOMIC64_OPS
#define ATOMIC64_OPS(op, op1, op2) \
ATOMIC64_OP(op, op1, op2) \
ATOMIC64_FETCH_OP(op, op1, op2)
#define atomic64_andnot atomic64_andnot
ATOMIC64_OPS(and, and, and)
ATOMIC64_OPS(andnot, bic, bic)
ATOMIC64_OPS(or, orr, orr)
ATOMIC64_OPS(xor, eor, eor)
#define atomic64_fetch_and_relaxed atomic64_fetch_and_relaxed
#define atomic64_fetch_andnot_relaxed atomic64_fetch_andnot_relaxed
#define atomic64_fetch_or_relaxed atomic64_fetch_or_relaxed
#define atomic64_fetch_xor_relaxed atomic64_fetch_xor_relaxed
#undef ATOMIC64_OPS
#undef ATOMIC64_FETCH_OP
#undef ATOMIC64_OP_RETURN
#undef ATOMIC64_OP
static inline long long
atomic64_cmpxchg_relaxed(atomic64_t *ptr, long long old, long long new)
{
long long oldval;
unsigned long res;
prefetchw(&ptr->counter);
do {
__asm__ __volatile__("@ atomic64_cmpxchg\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
"ldrexd %1, %H1, [%3]\n"
"mov %0, #0\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
"teq %1, %4\n"
"teqeq %H1, %H4\n"
"strexdeq %0, %5, %H5, [%3]"
: "=&r" (res), "=&r" (oldval), "+Qo" (ptr->counter)
: "r" (&ptr->counter), "r" (old), "r" (new)
: "cc");
} while (res);
return oldval;
}
#define atomic64_cmpxchg_relaxed atomic64_cmpxchg_relaxed
static inline long long atomic64_xchg_relaxed(atomic64_t *ptr, long long new)
{
long long result;
unsigned long tmp;
prefetchw(&ptr->counter);
__asm__ __volatile__("@ atomic64_xchg\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
"1: ldrexd %0, %H0, [%3]\n"
" strexd %1, %4, %H4, [%3]\n"
" teq %1, #0\n"
" bne 1b"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
: "=&r" (result), "=&r" (tmp), "+Qo" (ptr->counter)
: "r" (&ptr->counter), "r" (new)
: "cc");
return result;
}
#define atomic64_xchg_relaxed atomic64_xchg_relaxed
static inline long long atomic64_dec_if_positive(atomic64_t *v)
{
long long result;
unsigned long tmp;
smp_mb();
prefetchw(&v->counter);
__asm__ __volatile__("@ atomic64_dec_if_positive\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
"1: ldrexd %0, %H0, [%3]\n"
" subs %Q0, %Q0, #1\n"
" sbc %R0, %R0, #0\n"
" teq %R0, #0\n"
" bmi 2f\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
" strexd %1, %0, %H0, [%3]\n"
" teq %1, #0\n"
" bne 1b\n"
"2:"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter)
: "r" (&v->counter)
: "cc");
smp_mb();
return result;
}
static inline int atomic64_add_unless(atomic64_t *v, long long a, long long u)
{
long long val;
unsigned long tmp;
int ret = 1;
smp_mb();
prefetchw(&v->counter);
__asm__ __volatile__("@ atomic64_add_unless\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
"1: ldrexd %0, %H0, [%4]\n"
" teq %0, %5\n"
" teqeq %H0, %H5\n"
" moveq %1, #0\n"
" beq 2f\n"
" adds %Q0, %Q0, %Q6\n"
" adc %R0, %R0, %R6\n"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
" strexd %2, %0, %H0, [%4]\n"
" teq %2, #0\n"
" bne 1b\n"
"2:"
ARM: 6212/1: atomic ops: add memory constraints to inline asm Currently, the 32-bit and 64-bit atomic operations on ARM do not include memory constraints in the inline assembly blocks. In the case of barrier-less operations [for example, atomic_add], this means that the compiler may constant fold values which have actually been modified by a call to an atomic operation. This issue can be observed in the atomic64_test routine in <kernel root>/lib/atomic64_test.c: 00000000 <test_atomic64>: 0: e1a0c00d mov ip, sp 4: e92dd830 push {r4, r5, fp, ip, lr, pc} 8: e24cb004 sub fp, ip, #4 c: e24dd008 sub sp, sp, #8 10: e24b3014 sub r3, fp, #20 14: e30d000d movw r0, #53261 ; 0xd00d 18: e3011337 movw r1, #4919 ; 0x1337 1c: e34c0001 movt r0, #49153 ; 0xc001 20: e34a1aa3 movt r1, #43683 ; 0xaaa3 24: e16300f8 strd r0, [r3, #-8]! 28: e30c0afe movw r0, #51966 ; 0xcafe 2c: e30b1eef movw r1, #48879 ; 0xbeef 30: e34d0eaf movt r0, #57007 ; 0xdeaf 34: e34d1ead movt r1, #57005 ; 0xdead 38: e1b34f9f ldrexd r4, [r3] 3c: e1a34f90 strexd r4, r0, [r3] 40: e3340000 teq r4, #0 44: 1afffffb bne 38 <test_atomic64+0x38> 48: e59f0004 ldr r0, [pc, #4] ; 54 <test_atomic64+0x54> 4c: e3a0101e mov r1, #30 50: ebfffffe bl 0 <__bug> 54: 00000000 .word 0x00000000 The atomic64_set (0x38-0x44) writes to the atomic64_t, but the compiler doesn't see this, assumes the test condition is always false and generates an unconditional branch to __bug. The rest of the test is optimised away. This patch adds suitable memory constraints to the atomic operations on ARM to ensure that the compiler is informed of the correct data hazards. We have to use the "Qo" constraints to avoid hitting the GCC anomaly described at http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44492 , where the compiler makes assumptions about the writeback in the addressing mode used by the inline assembly. These constraints forbid the use of auto{inc,dec} addressing modes, so it doesn't matter if we don't use the operand exactly once. Cc: stable@kernel.org Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-07-08 16:59:16 +07:00
: "=&r" (val), "+r" (ret), "=&r" (tmp), "+Qo" (v->counter)
: "r" (&v->counter), "r" (u), "r" (a)
: "cc");
if (ret)
smp_mb();
return ret;
}
#define atomic64_add_negative(a, v) (atomic64_add_return((a), (v)) < 0)
#define atomic64_inc(v) atomic64_add(1LL, (v))
#define atomic64_inc_return_relaxed(v) atomic64_add_return_relaxed(1LL, (v))
#define atomic64_inc_and_test(v) (atomic64_inc_return(v) == 0)
#define atomic64_sub_and_test(a, v) (atomic64_sub_return((a), (v)) == 0)
#define atomic64_dec(v) atomic64_sub(1LL, (v))
#define atomic64_dec_return_relaxed(v) atomic64_sub_return_relaxed(1LL, (v))
#define atomic64_dec_and_test(v) (atomic64_dec_return((v)) == 0)
#define atomic64_inc_not_zero(v) atomic64_add_unless((v), 1LL, 0LL)
#endif /* !CONFIG_GENERIC_ATOMIC64 */
#endif
#endif