libc: provide atomic operations will full barriers for NDK apps.

__atomic_cmpxchg and other related atomic operations did not
provide memory barriers, which can be a problem for non-platform
code that links against them when it runs on multi-core devices.

This patch does two things to fix this:

- It modifies the existing implementation of the functions
  that are exported by the C library to always provide
  full memory barriers. We need to keep them exported by
  the C library to prevent breaking existing application
  machine code.

- It also modifies <sys/atomics.h> to only export
  always-inlined versions of the functions, to ensure that
  any application code compiled against the new header will
  not rely on the platform version of the functions.

  This ensure that said machine code will run properly on
  all multi-core devices.

This is based on the GCC built-in sync primitives.

The end result should be only slightly slower than the
previous implementation.

Note that the platform code does not use these functions
at all. A previous patch completely removed their usage in
the pthread and libstdc++ code.

+ rename arch-arm/bionic/atomics_arm.S to futex_arm.S
+ rename arch-x86/bionic/atomics_x86.S to futex_x86.S
+ remove arch-x86/include/sys/atomics.h which already
  provided inlined functions to the x86 platform.

Change-Id: I752a594475090cf37fa926bb38209c2175dda539
This commit is contained in:
David 'Digit' Turner 2011-11-16 17:37:15 +01:00
parent b385229837
commit 0fec6b9d88
7 changed files with 244 additions and 307 deletions

View File

@ -346,10 +346,11 @@ libc_common_src_files += \
arch-arm/bionic/__get_sp.S \
arch-arm/bionic/_exit_with_stack_teardown.S \
arch-arm/bionic/_setjmp.S \
arch-arm/bionic/atomics_arm.S \
arch-arm/bionic/atomics_arm.c \
arch-arm/bionic/clone.S \
arch-arm/bionic/eabi.c \
arch-arm/bionic/ffs.S \
arch-arm/bionic/futex_arm.S \
arch-arm/bionic/kill.S \
arch-arm/bionic/libgcc_compat.c \
arch-arm/bionic/tkill.S \
@ -394,9 +395,9 @@ libc_common_src_files += \
arch-x86/bionic/__get_sp.S \
arch-x86/bionic/__get_tls.c \
arch-x86/bionic/__set_tls.c \
arch-x86/bionic/atomics_x86.S \
arch-x86/bionic/clone.S \
arch-x86/bionic/_exit_with_stack_teardown.S \
arch-x86/bionic/futex_x86.S \
arch-x86/bionic/setjmp.S \
arch-x86/bionic/_setjmp.S \
arch-x86/bionic/sigsetjmp.S \

View File

@ -1,236 +0,0 @@
/*
* Copyright (C) 2008 The Android Open Source Project
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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 COPYRIGHT HOLDERS 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
* COPYRIGHT OWNER 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.
*/
#include <sys/linux-syscalls.h>
#include <machine/asm.h>
#include <machine/cpu-features.h>
#define FUTEX_WAIT 0
#define FUTEX_WAKE 1
#if defined(__ARM_HAVE_LDREX_STREX)
/*
* ===========================================================================
* ARMv6+ implementation
* ===========================================================================
*/
/* r0(addr) -> r0(old) */
ENTRY(__atomic_dec)
mov r1, r0 @ copy addr so we don't clobber it
1: ldrex r0, [r1] @ load current value into r0
sub r2, r0, #1 @ generate new value into r2
strex r3, r2, [r1] @ try to store new value; result in r3
cmp r3, #0 @ success?
bxeq lr @ yes, return
b 1b @ no, retry
END(__atomic_dec)
/* r0(addr) -> r0(old) */
ENTRY(__atomic_inc)
mov r1, r0
1: ldrex r0, [r1]
add r2, r0, #1
strex r3, r2, [r1]
cmp r3, #0
bxeq lr
b 1b
END(__atomic_inc)
/* r0(old) r1(new) r2(addr) -> r0(zero_if_succeeded) */
ENTRY(__atomic_cmpxchg)
1: mov ip, #2 @ ip=2 means "new != old"
ldrex r3, [r2] @ load current value into r3
teq r0, r3 @ new == old?
strexeq ip, r1, [r2] @ yes, try store, set ip to 0 or 1
teq ip, #1 @ strex failure?
beq 1b @ yes, retry
mov r0, ip @ return 0 on success, 2 on failure
bx lr
END(__atomic_cmpxchg)
/* r0(new) r1(addr) -> r0(old) */
ENTRY(__atomic_swap)
1: ldrex r2, [r1]
strex r3, r0, [r1]
teq r3, #0
bne 1b
mov r0, r2
bx lr
END(__atomic_swap)
#else /*not defined __ARM_HAVE_LDREX_STREX*/
/*
* ===========================================================================
* Pre-ARMv6 implementation
* ===========================================================================
*/
/* int __kernel_cmpxchg(int oldval, int newval, int* ptr) */
.equ kernel_cmpxchg, 0xFFFF0FC0
.equ kernel_atomic_base, 0xFFFF0FFF
/* r0(addr) -> r0(old) */
ENTRY(__atomic_dec)
.save {r4, lr}
stmdb sp!, {r4, lr}
mov r2, r0
1: @ atomic_dec
ldr r0, [r2]
mov r3, #kernel_atomic_base
add lr, pc, #4
sub r1, r0, #1
add pc, r3, #(kernel_cmpxchg - kernel_atomic_base)
bcc 1b
add r0, r1, #1
ldmia sp!, {r4, lr}
bx lr
END(__atomic_dec)
/* r0(addr) -> r0(old) */
ENTRY(__atomic_inc)
.save {r4, lr}
stmdb sp!, {r4, lr}
mov r2, r0
1: @ atomic_inc
ldr r0, [r2]
mov r3, #kernel_atomic_base
add lr, pc, #4
add r1, r0, #1
add pc, r3, #(kernel_cmpxchg - kernel_atomic_base)
bcc 1b
sub r0, r1, #1
ldmia sp!, {r4, lr}
bx lr
END(__atomic_inc)
/* r0(old) r1(new) r2(addr) -> r0(zero_if_succeeded) */
ENTRY(__atomic_cmpxchg)
.save {r4, lr}
stmdb sp!, {r4, lr}
mov r4, r0 /* r4 = save oldvalue */
1: @ atomic_cmpxchg
mov r3, #kernel_atomic_base
add lr, pc, #4
mov r0, r4 /* r0 = oldvalue */
add pc, r3, #(kernel_cmpxchg - kernel_atomic_base)
bcs 2f /* swap was made. we're good, return. */
ldr r3, [r2] /* swap not made, see if it's because *ptr!=oldvalue */
cmp r3, r4
beq 1b
2: @ atomic_cmpxchg
ldmia sp!, {r4, lr}
bx lr
END(__atomic_cmpxchg)
/* r0(new) r1(addr) -> r0(old) */
ENTRY(__atomic_swap)
swp r0, r0, [r1]
bx lr
END(__atomic_swap)
#endif /*not defined __ARM_HAVE_LDREX_STREX*/
/* __futex_wait(*ftx, val, *timespec) */
/* __futex_wake(*ftx, counter) */
/* __futex_syscall3(*ftx, op, val) */
/* __futex_syscall4(*ftx, op, val, *timespec) */
.global __futex_wait
.type __futex_wait, %function
.global __futex_wake
.type __futex_wake, %function
.global __futex_syscall3
.type __futex_syscall3, %function
.global __futex_syscall4
.type __futex_syscall4, %function
#if __ARM_EABI__
ENTRY(__futex_syscall3)
stmdb sp!, {r4, r7}
.save {r4, r7}
ldr r7, =__NR_futex
swi #0
ldmia sp!, {r4, r7}
bx lr
END(__futex_syscall3)
ENTRY(__futex_wait)
stmdb sp!, {r4, r7}
.save {r4, r7}
mov r3, r2
mov r2, r1
mov r1, #FUTEX_WAIT
ldr r7, =__NR_futex
swi #0
ldmia sp!, {r4, r7}
bx lr
END(__futex_wait)
ENTRY(__futex_wake)
.save {r4, r7}
stmdb sp!, {r4, r7}
mov r2, r1
mov r1, #FUTEX_WAKE
ldr r7, =__NR_futex
swi #0
ldmia sp!, {r4, r7}
bx lr
END(__futex_wake)
#else
ENTRY(__futex_syscall3)
swi #__NR_futex
bx lr
END(__futex_syscall3)
ENTRY(__futex_wait)
mov r3, r2
mov r2, r1
mov r1, #FUTEX_WAIT
swi #__NR_futex
bx lr
END(__futex_wait)
ENTRY(__futex_wake)
mov r2, r1
mov r1, #FUTEX_WAKE
swi #__NR_futex
bx lr
END(__futex_wake)
#endif
ENTRY(__futex_syscall4)
b __futex_syscall3
END(__futex_syscall4)

View File

@ -0,0 +1,87 @@
/*
* Copyright (C) 2011 The Android Open Source Project
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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 COPYRIGHT HOLDERS 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
* COPYRIGHT OWNER 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.
*/
/* The purpose of this file is to export a small set of atomic-related
* functions from the C library, to ensure binary ABI compatibility for
* the NDK.
*
* These functions were initially exposed by the NDK through <sys/atomics.h>,
* which was unfortunate because their implementation didn't provide any
* memory barriers at all.
*
* This wasn't a problem for the platform code that used them, because it
* used explicit barrier instructions around them. On the other hand, it means
* that any NDK-generated machine code that linked against them would not
* perform correctly when running on multi-core devices.
*
* To fix this, the platform code was first modified to not use any of these
* functions (everything is now inlined through assembly statements, see
* libc/private/bionic_arm_inline.h and the headers it includes.
*
* The functions here are thus only for the benefit of NDK applications,
* and now includes full memory barriers to prevent any random memory ordering
* issue from cropping.
*
* Note that we also provide an updated <sys/atomics.h> header that defines
* always_inlined versions of the functions that use the GCC builtin
* intrinsics to perform the same thing.
*
* NOTE: There is no need for a similar file for non-ARM platforms.
*/
/* DO NOT INCLUDE <sys/atomics.h> HERE ! */
int
__android_cmpxchg(int old, int _new, volatile int *ptr)
{
/* We must return 0 on success */
return __sync_val_compare_and_swap(ptr, old, _new) != old;
}
int
__atomic_swap(int _new, volatile int *ptr)
{
int prev;
do {
prev = *ptr;
} while (__sync_val_compare_and_swap(ptr, prev, _new) != prev);
return prev;
}
int
__atomic_dec(volatile int *ptr)
{
return __sync_fetch_and_sub (ptr, 1);
}
int
__atomic_inc(volatile int *ptr)
{
return __sync_fetch_and_add (ptr, 1);
}

View File

@ -0,0 +1,112 @@
/*
* Copyright (C) 2008 The Android Open Source Project
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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 COPYRIGHT HOLDERS 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
* COPYRIGHT OWNER 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.
*/
#include <sys/linux-syscalls.h>
#include <machine/asm.h>
#include <machine/cpu-features.h>
#define FUTEX_WAIT 0
#define FUTEX_WAKE 1
/* __futex_wait(*ftx, val, *timespec) */
/* __futex_wake(*ftx, counter) */
/* __futex_syscall3(*ftx, op, val) */
/* __futex_syscall4(*ftx, op, val, *timespec) */
.global __futex_wait
.type __futex_wait, %function
.global __futex_wake
.type __futex_wake, %function
.global __futex_syscall3
.type __futex_syscall3, %function
.global __futex_syscall4
.type __futex_syscall4, %function
#if __ARM_EABI__
ENTRY(__futex_syscall3)
stmdb sp!, {r4, r7}
.save {r4, r7}
ldr r7, =__NR_futex
swi #0
ldmia sp!, {r4, r7}
bx lr
END(__futex_syscall3)
ENTRY(__futex_wait)
stmdb sp!, {r4, r7}
.save {r4, r7}
mov r3, r2
mov r2, r1
mov r1, #FUTEX_WAIT
ldr r7, =__NR_futex
swi #0
ldmia sp!, {r4, r7}
bx lr
END(__futex_wait)
ENTRY(__futex_wake)
.save {r4, r7}
stmdb sp!, {r4, r7}
mov r2, r1
mov r1, #FUTEX_WAKE
ldr r7, =__NR_futex
swi #0
ldmia sp!, {r4, r7}
bx lr
END(__futex_wake)
#else
ENTRY(__futex_syscall3)
swi #__NR_futex
bx lr
END(__futex_syscall3)
ENTRY(__futex_wait)
mov r3, r2
mov r2, r1
mov r1, #FUTEX_WAIT
swi #__NR_futex
bx lr
END(__futex_wait)
ENTRY(__futex_wake)
mov r2, r1
mov r1, #FUTEX_WAKE
swi #__NR_futex
bx lr
END(__futex_wake)
#endif
ENTRY(__futex_syscall4)
b __futex_syscall3
END(__futex_syscall4)

View File

@ -1,65 +0,0 @@
/*
* Copyright (C) 2011 The Android Open Source Project
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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 COPYRIGHT HOLDERS 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
* COPYRIGHT OWNER 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.
*/
#ifndef _SYS_ATOMICS_H
#define _SYS_ATOMICS_H
#include <sys/cdefs.h>
#include <sys/time.h>
__BEGIN_DECLS
static inline __attribute__((always_inline)) int
__atomic_cmpxchg(int old, int _new, volatile int *ptr)
{
return !__sync_bool_compare_and_swap (ptr, old, _new);
}
static inline __attribute__((always_inline)) int
__atomic_swap(int _new, volatile int *ptr)
{
return __sync_lock_test_and_set(ptr, _new);
}
static inline __attribute__((always_inline)) int
__atomic_dec(volatile int *ptr)
{
return __sync_fetch_and_sub (ptr, 1);
}
static inline __attribute__((always_inline)) int
__atomic_inc(volatile int *ptr)
{
return __sync_fetch_and_add (ptr, 1);
}
int __futex_wait(volatile void *ftx, int val, const struct timespec *timeout);
int __futex_wake(volatile void *ftx, int count);
__END_DECLS
#endif /* _SYS_ATOMICS_H */

View File

@ -33,10 +33,48 @@
__BEGIN_DECLS
extern int __atomic_cmpxchg(int old, int _new, volatile int *ptr);
extern int __atomic_swap(int _new, volatile int *ptr);
extern int __atomic_dec(volatile int *ptr);
extern int __atomic_inc(volatile int *ptr);
/* Note: atomic operations that were exported by the C library didn't
* provide any memory barriers, which created potential issues on
* multi-core devices. We now define them as inlined calls to
* GCC sync builtins, which always provide a full barrier.
*
* NOTE: The C library still exports atomic functions by the same
* name to ensure ABI stability for existing NDK machine code.
*
* If you are an NDK developer, we encourage you to rebuild your
* unmodified sources against this header as soon as possible.
*/
#define __ATOMIC_INLINE__ static __inline__ __attribute__((always_inline))
__ATOMIC_INLINE__ int
__atomic_cmpxchg(int old, int _new, volatile int *ptr)
{
/* We must return 0 on success */
return __sync_val_compare_and_swap(ptr, old, _new) != old;
}
__ATOMIC_INLINE__ int
__atomic_swap(int _new, volatile int *ptr)
{
int prev;
do {
prev = *ptr;
} while (__sync_val_compare_and_swap(ptr, prev, _new) != prev);
return prev;
}
__ATOMIC_INLINE__ int
__atomic_dec(volatile int *ptr)
{
return __sync_fetch_and_sub (ptr, 1);
}
__ATOMIC_INLINE__ int
__atomic_inc(volatile int *ptr)
{
return __sync_fetch_and_add (ptr, 1);
}
int __futex_wait(volatile void *ftx, int val, const struct timespec *timeout);
int __futex_wake(volatile void *ftx, int count);