[haiku-commits] haiku: hrev47741 - src/system/libroot/os/arch/arm src/system/libroot/os/arch/generic src/system/libroot/os/arch/x86_64 headers/os/support headers/private/kernel/arch/x86/64
- From: pdziepak@xxxxxxxxxxx
- To: haiku-commits@xxxxxxxxxxxxx
- Date: Mon, 25 Aug 2014 23:37:39 +0200 (CEST)
hrev47741 adds 6 changesets to branch 'master'
old head: ae18461ec1ad87c67a29310349b47ffd17f475f2
new head: 721a07ac24df63af87afd8d52f5e08ec6566c9e0
overview: http://cgit.haiku-os.org/haiku/log/?qt=range&q=721a07a+%5Eae18461
----------------------------------------------------------------------------
54b314f: add-ons/kernel: add casts when calling atomic_*()
Soon GCC is going to be a bit stricter about the type of pointer passed
to atomic_*() functions.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
2e2c9bd: os/support: implement atomic_*() using GCC builtin helpers
If GCC knows what these functions are actually doing the resulting
code can be optimized better what is especially noticeable in case of
invocations of atomic_{or,and}() that ignore the result. Obviously,
everything is inlined what also improves performance.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
aa58f7e: os/arch: implement all atomic_*() using std::atomic<>
Time to get rid of some asm code. Surprisingly, it appears that
on x86[_64] the emitted code for atomic_test_and_get() isn't as efficient
as it could be, even with -O2, but cmpxchg is so expensive that this slight
difference shouldn't matter much.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
d3b1caa: kernel, libroot: use C++11 atomics in atomic_*()
The less assembler in our sources the better. These functions wouldn't
be used very much since SupportDef.h inlines them, but the symbols should
be available.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
a4cdc60: build: remove B_USE_BUILTIN_ATOMIC_FUNCTIONS
No reason not to use GCC atomic support on non-x86 archs.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
721a07a: libroot: remove ATOMIC_FUNCS_ARE_SYSCALLS
GCC knows whether these functions need to be implemented using syscalls
(or more clever solutions like in Linux) and calls libgcc in such case.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
[ Paweł Dziepak <pdziepak@xxxxxxxxxxx> ]
----------------------------------------------------------------------------
16 files changed, 268 insertions(+), 431 deletions(-)
headers/os/support/SupportDefs.h | 158 +++++++++++++++----
headers/private/kernel/arch/x86/32/atomic.h | 14 +-
headers/private/kernel/arch/x86/64/atomic.h | 118 --------------
src/add-ons/kernel/busses/ata/ide_isa/ide_isa.c | 2 +-
.../drivers/disk/floppy/pc_floppy/floppy.c | 2 +-
src/build/libroot/atomic.cpp | 6 +
src/system/boot/Jamfile | 7 +-
src/system/kernel/lib/arch/arm/Jamfile | 1 -
src/system/kernel/lib/arch/x86_64/Jamfile | 3 +-
src/system/libroot/os/arch/arm/Jamfile | 1 -
src/system/libroot/os/arch/arm/atomic.S | 113 -------------
.../libroot/os/arch/generic/generic_atomic.cpp | 114 ++++++++++++-
src/system/libroot/os/arch/x86_64/Jamfile | 5 +-
src/system/libroot/os/arch/x86_64/atomic.S | 153 ------------------
src/system/runtime_loader/arch/arm/Jamfile | 1 -
src/system/runtime_loader/arch/x86_64/Jamfile | 1 -
############################################################################
Commit: 54b314f15fa73d8e8c8425c4490d82b13754f364
URL: http://cgit.haiku-os.org/haiku/commit/?id=54b314f
Author: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
Date: Mon Aug 25 12:54:24 2014 UTC
add-ons/kernel: add casts when calling atomic_*()
Soon GCC is going to be a bit stricter about the type of pointer passed
to atomic_*() functions.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
----------------------------------------------------------------------------
diff --git a/src/add-ons/kernel/busses/ata/ide_isa/ide_isa.c
b/src/add-ons/kernel/busses/ata/ide_isa/ide_isa.c
index 591d764..80e1d40 100644
--- a/src/add-ons/kernel/busses/ata/ide_isa/ide_isa.c
+++ b/src/add-ons/kernel/busses/ata/ide_isa/ide_isa.c
@@ -409,7 +409,7 @@ channel_removed(void *cookie)
TRACE("channel_removed()\n");
// disable access instantly
- atomic_or(&channel->lost, 1);
+ atomic_or((int32*)&channel->lost, 1);
}
diff --git a/src/add-ons/kernel/drivers/disk/floppy/pc_floppy/floppy.c
b/src/add-ons/kernel/drivers/disk/floppy/pc_floppy/floppy.c
index b78cda2..90e29d2 100644
--- a/src/add-ons/kernel/drivers/disk/floppy/pc_floppy/floppy.c
+++ b/src/add-ons/kernel/drivers/disk/floppy/pc_floppy/floppy.c
@@ -545,7 +545,7 @@ motor_off_daemon(void *t, int tim)
for (i = 0; i < MAX_FLOPPIES; i++) {
if (floppies[i].iobase && !floppies[i].pending_cmd &&
floppies[i].motor_timeout > 0) {
TRACE("floppies[%d].motor_timeout = %ld\n", i,
floppies[i].motor_timeout);
- if (atomic_add(&floppies[i].motor_timeout, -500000) <=
500000) {
+ if (atomic_add((int32*)&floppies[i].motor_timeout,
-500000) <= 500000) {
dprintf("turning off motor for drive %d\n",
floppies[i].drive_num);
turn_off_motor(&floppies[i]);
floppies[i].motor_timeout = 0;
############################################################################
Commit: 2e2c9bd3d058dac3609103ca67011eaee14e9333
URL: http://cgit.haiku-os.org/haiku/commit/?id=2e2c9bd
Author: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
Date: Mon Aug 25 12:55:55 2014 UTC
os/support: implement atomic_*() using GCC builtin helpers
If GCC knows what these functions are actually doing the resulting
code can be optimized better what is especially noticeable in case of
invocations of atomic_{or,and}() that ignore the result. Obviously,
everything is inlined what also improves performance.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
----------------------------------------------------------------------------
diff --git a/headers/os/support/SupportDefs.h b/headers/os/support/SupportDefs.h
index 1f413d3..b863334 100644
--- a/headers/os/support/SupportDefs.h
+++ b/headers/os/support/SupportDefs.h
@@ -195,23 +195,6 @@ extern const char *B_EMPTY_STRING;
extern "C" {
#endif
-/* Atomic functions; previous value is returned */
-extern void atomic_set(int32* value, int32 newValue);
-extern int32 atomic_get_and_set(int32* value, int32 newValue);
-extern int32 atomic_test_and_set(int32 *value, int32 newValue, int32
testAgainst);
-extern int32 atomic_add(int32 *value, int32 addValue);
-extern int32 atomic_and(int32 *value, int32 andValue);
-extern int32 atomic_or(int32 *value, int32 orValue);
-extern int32 atomic_get(int32 *value);
-
-extern void atomic_set64(int64* value, int64 newValue);
-extern int64 atomic_get_and_set64(int64* value, int64 newValue);
-extern int64 atomic_test_and_set64(int64 *value, int64 newValue, int64
testAgainst);
-extern int64 atomic_add64(int64 *value, int64 addValue);
-extern int64 atomic_and64(int64 *value, int64 andValue);
-extern int64 atomic_or64(int64 *value, int64 orValue);
-extern int64 atomic_get64(int64 *value);
-
/* Other stuff */
extern void* get_stack_frame(void);
@@ -232,18 +215,141 @@ extern void* get_stack_frame(void);
/* Use the built-in atomic functions, if requested and available. */
-#if defined(B_USE_BUILTIN_ATOMIC_FUNCTIONS) && __GNUC__ >= 4
+#if defined(B_USE_BUILTIN_ATOMIC_FUNCTIONS) && __GNUC__ > 4 \
+ || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7)
+
+
+static inline void
+atomic_set(int32* value, int32 newValue)
+{
+ __atomic_store_n(value, newValue, __ATOMIC_RELEASE);
+}
+
+
+static inline int32
+atomic_get_and_set(int32* value, int32 newValue)
+{
+ return __atomic_exchange_n(value, newValue, __ATOMIC_SEQ_CST);
+}
+
+
+static inline int32
+atomic_test_and_set(int32* value, int32 newValue, int32 testAgainst)
+{
+ __atomic_compare_exchange_n(value, &testAgainst, newValue, 1,
+ __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST);
+ return testAgainst;
+}
+
+
+static inline int32
+atomic_add(int32* value, int32 addValue)
+{
+ return __atomic_fetch_add(value, addValue, __ATOMIC_SEQ_CST);
+}
+
+
+static inline int32
+atomic_and(int32* value, int32 andValue)
+{
+ return __atomic_fetch_and(value, andValue, __ATOMIC_SEQ_CST);
+}
+
+
+static inline int32
+atomic_or(int32* value, int32 orValue)
+{
+ return __atomic_fetch_or(value, orValue, __ATOMIC_SEQ_CST);
+}
+
+
+static inline int32
+atomic_get(int32* value)
+{
+ return __atomic_load_n(value, __ATOMIC_ACQUIRE);
+}
+
+
+static inline void
+atomic_set64(int64* value, int64 newValue)
+{
+ __atomic_store_n(value, newValue, __ATOMIC_RELEASE);
+}
+
+
+static inline int64
+atomic_get_and_set64(int64* value, int64 newValue)
+{
+ return __atomic_exchange_n(value, newValue, __ATOMIC_SEQ_CST);
+}
+
+
+static inline int64
+atomic_test_and_set64(int64* value, int64 newValue, int64 testAgainst)
+{
+ __atomic_compare_exchange_n(value, &testAgainst, newValue, 1,
+ __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST);
+ return testAgainst;
+}
+
+
+static inline int64
+atomic_add64(int64* value, int64 addValue)
+{
+ return __atomic_fetch_add(value, addValue, __ATOMIC_SEQ_CST);
+}
+
+
+static inline int64
+atomic_and64(int64* value, int64 andValue)
+{
+ return __atomic_fetch_and(value, andValue, __ATOMIC_SEQ_CST);
+}
+
-#define atomic_test_and_set(valuePointer, newValue, testAgainst) \
- __sync_val_compare_and_swap(valuePointer, testAgainst, newValue)
-#define atomic_add(valuePointer, addValue) \
- __sync_fetch_and_add(valuePointer, addValue)
-#define atomic_and(valuePointer, andValue) \
- __sync_fetch_and_and(valuePointer, andValue)
-#define atomic_or(valuePointer, orValue) \
- __sync_fetch_and_or(valuePointer, orValue)
+static inline int64
+atomic_or64(int64* value, int64 orValue)
+{
+ return __atomic_fetch_or(value, orValue, __ATOMIC_SEQ_CST);
+}
+
+
+static inline int64
+atomic_get64(int64* value)
+{
+ return __atomic_load_n(value, __ATOMIC_ACQUIRE);
+}
+
+
+#else // B_USE_BUILTIN_ATOMIC_FUNCTIONS && __GNUC__ > 4
+ // || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7)
-#endif // B_USE_BUILTIN_ATOMIC_FUNCTIONS && __GNUC__ >= 4
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Atomic functions; previous value is returned */
+extern void atomic_set(int32* value, int32 newValue);
+extern int32 atomic_get_and_set(int32* value, int32 newValue);
+extern int32 atomic_test_and_set(int32 *value, int32 newValue, int32
testAgainst);
+extern int32 atomic_add(int32 *value, int32 addValue);
+extern int32 atomic_and(int32 *value, int32 andValue);
+extern int32 atomic_or(int32 *value, int32 orValue);
+extern int32 atomic_get(int32 *value);
+
+extern void atomic_set64(int64* value, int64 newValue);
+extern int64 atomic_get_and_set64(int64* value, int64 newValue);
+extern int64 atomic_test_and_set64(int64 *value, int64 newValue, int64
testAgainst);
+extern int64 atomic_add64(int64 *value, int64 addValue);
+extern int64 atomic_and64(int64 *value, int64 andValue);
+extern int64 atomic_or64(int64 *value, int64 orValue);
+extern int64 atomic_get64(int64 *value);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
#endif /* _SUPPORT_DEFS_H */
diff --git a/headers/private/kernel/arch/x86/32/atomic.h
b/headers/private/kernel/arch/x86/32/atomic.h
index adec9c0..eb2996c 100644
--- a/headers/private/kernel/arch/x86/32/atomic.h
+++ b/headers/private/kernel/arch/x86/32/atomic.h
@@ -32,6 +32,9 @@ memory_full_barrier_inline(void)
#define memory_full_barrier memory_full_barrier_inline
+#if __GNUC__ < 4
+
+
static inline void
atomic_set_inline(int32* value, int32 newValue)
{
@@ -84,14 +87,13 @@ atomic_get_inline(int32* value)
#define atomic_set atomic_set_inline
#define atomic_get_and_set atomic_get_and_set_inline
-#ifndef atomic_test_and_set
-# define atomic_test_and_set atomic_test_and_set_inline
-#endif
-#ifndef atomic_add
-# define atomic_add atomic_add_inline
-#endif
+#define atomic_test_and_set atomic_test_and_set_inline
+#define atomic_add atomic_add_inline
#define atomic_get atomic_get_inline
+#endif // dark ages
+
+
#endif // _KERNEL_ARCH_X86_32_ATOMIC_H
diff --git a/headers/private/kernel/arch/x86/64/atomic.h
b/headers/private/kernel/arch/x86/64/atomic.h
index fd56731..ed3245e 100644
--- a/headers/private/kernel/arch/x86/64/atomic.h
+++ b/headers/private/kernel/arch/x86/64/atomic.h
@@ -31,123 +31,5 @@ memory_full_barrier_inline(void)
#define memory_write_barrier memory_write_barrier_inline
#define memory_full_barrier memory_full_barrier_inline
-
-static inline void
-atomic_set_inline(int32* value, int32 newValue)
-{
- memory_write_barrier();
- *(volatile int32*)value = newValue;
-}
-
-
-static inline int32
-atomic_get_and_set_inline(int32* value, int32 newValue)
-{
- asm volatile("xchg %0, (%1)"
- : "+r" (newValue)
- : "r" (value)
- : "memory");
- return newValue;
-}
-
-
-static inline int32
-atomic_test_and_set_inline(int32* value, int32 newValue, int32 testAgainst)
-{
- asm volatile("lock; cmpxchgl %2, (%3)"
- : "=a" (newValue)
- : "0" (testAgainst), "r" (newValue), "r" (value)
- : "memory");
- return newValue;
-}
-
-
-static inline int32
-atomic_add_inline(int32* value, int32 newValue)
-{
- asm volatile("lock; xaddl %0, (%1)"
- : "+r" (newValue)
- : "r" (value)
- : "memory");
- return newValue;
-}
-
-
-static inline int32
-atomic_get_inline(int32* value)
-{
- int32 newValue = *(volatile int32*)value;
- memory_read_barrier();
- return newValue;
-}
-
-
-static inline void
-atomic_set64_inline(int64* value, int64 newValue)
-{
- memory_write_barrier();
- *(volatile int64*)value = newValue;
-}
-
-
-static inline int64
-atomic_get_and_set64_inline(int64* value, int64 newValue)
-{
- asm volatile("xchgq %0, (%1)"
- : "+r" (newValue)
- : "r" (value)
- : "memory");
- return newValue;
-}
-
-
-static inline int64
-atomic_test_and_set64_inline(int64* value, int64 newValue, int64 testAgainst)
-{
- asm volatile("lock; cmpxchgq %2, (%3)"
- : "=a" (newValue)
- : "0" (testAgainst), "r" (newValue), "r" (value)
- : "memory");
- return newValue;
-}
-
-
-static inline int64
-atomic_add64_inline(int64* value, int64 newValue)
-{
- asm volatile("lock; xaddq %0, (%1)"
- : "+r" (newValue)
- : "r" (value)
- : "memory");
- return newValue;
-}
-
-
-static inline int64
-atomic_get64_inline(int64* value)
-{
- int64 newValue = *(volatile int64*)value;
- memory_read_barrier();
- return newValue;
-}
-
-
-#define atomic_set atomic_set_inline
-#define atomic_get_and_set atomic_get_and_set_inline
-#ifndef atomic_test_and_set
-# define atomic_test_and_set atomic_test_and_set_inline
-#endif
-#ifndef atomic_add
-# define atomic_add atomic_add_inline
-#endif
-#define atomic_get atomic_get_inline
-
-#define atomic_set64 atomic_set64_inline
-#define atomic_get_and_set64 atomic_get_and_set64_inline
-#define atomic_test_and_set64 atomic_test_and_set64_inline
-#define atomic_add64 atomic_add64_inline
-#define atomic_get64 atomic_get64_inline
-
-
#endif // _KERNEL_ARCH_X86_64_ATOMIC_H
diff --git a/src/build/libroot/atomic.cpp b/src/build/libroot/atomic.cpp
index 38aed02..0e1fed4 100644
--- a/src/build/libroot/atomic.cpp
+++ b/src/build/libroot/atomic.cpp
@@ -6,6 +6,9 @@
#include <SupportDefs.h>
+#if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 7)
+
+
void
atomic_set(int32 *value, int32 newValue)
{
@@ -120,3 +123,6 @@ atomic_get64(int64 *value)
return *value;
}
+
+#endif // __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 7)
+
############################################################################
Commit: aa58f7e431a5b295d9d83c59418d97bec7180fa8
URL: http://cgit.haiku-os.org/haiku/commit/?id=aa58f7e
Author: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
Date: Mon Aug 25 13:01:44 2014 UTC
os/arch: implement all atomic_*() using std::atomic<>
Time to get rid of some asm code. Surprisingly, it appears that
on x86[_64] the emitted code for atomic_test_and_get() isn't as efficient
as it could be, even with -O2, but cmpxchg is so expensive that this slight
difference shouldn't matter much.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
----------------------------------------------------------------------------
diff --git a/src/system/libroot/os/arch/generic/generic_atomic.cpp
b/src/system/libroot/os/arch/generic/generic_atomic.cpp
index 73b5c56..dce3a5b 100644
--- a/src/system/libroot/os/arch/generic/generic_atomic.cpp
+++ b/src/system/libroot/os/arch/generic/generic_atomic.cpp
@@ -9,11 +9,119 @@
#ifndef ATOMIC_FUNCS_ARE_SYSCALLS
-extern "C" int32_t
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] void
+atomic_set(int32_t* ptr, int32_t value)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int32_t>*>(ptr);
+ obj.store(value, std::memory_order_release);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int32_t
atomic_get_and_set(int32_t* ptr, int32_t value)
{
- auto& obj = *reinterpret_cast<std::atomic<int32_t>*>(ptr);
- return obj.exchange(value);
+ auto& obj = *reinterpret_cast<std::atomic<int32_t>*>(ptr);
+ return obj.exchange(value);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int32_t
+atomic_test_and_set(int32_t* ptr, int32_t desired, int32_t expected)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int32_t>*>(ptr);
+ obj.compare_exchange_strong(expected, desired);
+ return expected;
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int32_t
+atomic_add(int32_t* ptr, int32_t value)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int32_t>*>(ptr);
+ return obj.fetch_add(value);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int32_t
+atomic_and(int32_t* ptr, int32_t value)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int32_t>*>(ptr);
+ return obj.fetch_and(value);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int32_t
+atomic_or(int32_t* ptr, int32_t value)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int32_t>*>(ptr);
+ return obj.fetch_or(value);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int32_t
+atomic_get(int32_t* ptr)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int32_t>*>(ptr);
+ return obj.load(std::memory_order_acquire);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] void
+atomic_set64(int64_t* ptr, int64_t value)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int64_t>*>(ptr);
+ obj.store(value, std::memory_order_release);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int64_t
+atomic_get_and_set64(int64_t* ptr, int64_t value)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int64_t>*>(ptr);
+ return obj.exchange(value);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int64_t
+atomic_test_and_set64(int64_t* ptr, int64_t desired, int64_t expected)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int64_t>*>(ptr);
+ obj.compare_exchange_strong(expected, desired);
+ return expected;
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int64_t
+atomic_add64(int64_t* ptr, int64_t value)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int64_t>*>(ptr);
+ return obj.fetch_add(value);
}
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int64_t
+atomic_and64(int64_t* ptr, int64_t value)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int64_t>*>(ptr);
+ return obj.fetch_and(value);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int64_t
+atomic_or64(int64_t* ptr, int64_t value)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int64_t>*>(ptr);
+ return obj.fetch_or(value);
+}
+
+
+extern "C" [[gnu::optimize("omit-frame-pointer")]] int64_t
+atomic_get64(int64_t* ptr)
+{
+ auto& obj = *reinterpret_cast<std::atomic<int64_t>*>(ptr);
+ return obj.load(std::memory_order_acquire);
+}
+
+
#endif /* ATOMIC64_FUNCS_ARE_SYSCALLS */
############################################################################
Commit: d3b1caa62d83444c7c95a73cdc2094d2087fb818
URL: http://cgit.haiku-os.org/haiku/commit/?id=d3b1caa
Author: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
Date: Mon Aug 25 13:05:00 2014 UTC
kernel, libroot: use C++11 atomics in atomic_*()
The less assembler in our sources the better. These functions wouldn't
be used very much since SupportDef.h inlines them, but the symbols should
be available.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
----------------------------------------------------------------------------
diff --git a/src/system/boot/Jamfile b/src/system/boot/Jamfile
index 46aecb6..4070d0e 100644
--- a/src/system/boot/Jamfile
+++ b/src/system/boot/Jamfile
@@ -11,9 +11,13 @@ SEARCH_SOURCE += [ FDirName $(SUBDIR) $(DOTDOT) kernel lib ]
;
UsePrivateHeaders [ FDirName libroot locale ] ;
+local extraSources = ;
+if $(TARGET_GCC_VERSION_$(TARGET_PACKAGING_ARCH)[1]) = 2 {
+ extraSources += atomic.S ;
+}
+
BootMergeObject boot_libroot.o :
abs.c
- atomic.S
ctype.cpp
LocaleData.cpp
qsort.c
@@ -36,6 +40,7 @@ BootMergeObject boot_libroot.o :
strchr.c
strrchr.c
strtol.c
+ $(extraSources)
: -fno-pic
;
diff --git a/src/system/kernel/lib/arch/arm/Jamfile
b/src/system/kernel/lib/arch/arm/Jamfile
index 12b8c50..2b7c1cc 100644
--- a/src/system/kernel/lib/arch/arm/Jamfile
+++ b/src/system/kernel/lib/arch/arm/Jamfile
@@ -9,7 +9,6 @@ SEARCH_SOURCE += [ FDirName $(librootSources) os arch
$(TARGET_ARCH) ] ;
SEARCH_SOURCE += [ FDirName $(librootSources) os arch generic ] ;
KernelMergeObject kernel_os_arch_$(TARGET_ARCH).o :
- atomic.S
byteorder.S
generic_atomic.cpp
diff --git a/src/system/kernel/lib/arch/x86_64/Jamfile
b/src/system/kernel/lib/arch/x86_64/Jamfile
index 259708b..168f14c 100644
--- a/src/system/kernel/lib/arch/x86_64/Jamfile
+++ b/src/system/kernel/lib/arch/x86_64/Jamfile
@@ -10,13 +10,14 @@ local librootSources = [ FDirName $(HAIKU_TOP) src system
libroot ] ;
local posixSources = [ FDirName $(librootSources) posix ] ;
SEARCH_SOURCE += [ FDirName $(librootSources) os arch $(TARGET_ARCH) ] ;
+SEARCH_SOURCE += [ FDirName $(librootSources) os arch generic ] ;
KernelMergeObject kernel_os_arch_$(TARGET_ARCH).o :
- atomic.S
byteorder.S
system_time_asm.S
system_time.c
+ generic_atomic.cpp
: $(TARGET_KERNEL_PIC_CCFLAGS)
;
diff --git a/src/system/libroot/os/arch/arm/Jamfile
b/src/system/libroot/os/arch/arm/Jamfile
index 87acdea..d22221b 100644
--- a/src/system/libroot/os/arch/arm/Jamfile
+++ b/src/system/libroot/os/arch/arm/Jamfile
@@ -11,7 +11,6 @@ for architectureObject in [ MultiArchSubDirSetup arm ] {
SEARCH_SOURCE += [ FDirName $(SUBDIR) $(DOTDOT) generic ] ;
MergeObject <$(architecture)>os_arch_$(TARGET_ARCH).o :
- atomic.S
byteorder.S
system_time.c
stack_frame.c
diff --git a/src/system/libroot/os/arch/arm/atomic.S
b/src/system/libroot/os/arch/arm/atomic.S
deleted file mode 100644
index 56f1c2c..0000000
--- a/src/system/libroot/os/arch/arm/atomic.S
+++ /dev/null
@@ -1,113 +0,0 @@
-/*
-** Copyright 2003, Axel Dörfler, axeld@xxxxxxxxxxxxxxxx. All rights reserved.
-** Distributed under the terms of the OpenBeOS License.
-*/
-
-#include <asm_defs.h>
-#include <arch_config.h>
-
-
-.text
-
-#ifndef ATOMIC_FUNCS_ARE_SYSCALLS
-
-/* int atomic_add(int *value, int increment)
- */
-FUNCTION(atomic_add):
-miss1: ldrex r12, [r0]
- add r2, r12, r1
- strex r3, r2, [r0]
- teq r3, #0
- bne miss1
- mov r0, r12
- bx lr
-FUNCTION_END(atomic_add)
-
-/* int atomic_and(int *value, int andValue)
- */
-FUNCTION(atomic_and):
-miss2: ldrex r12, [r0]
- and r2, r12, r1
- strex r3, r2, [r0]
- teq r3, #0
- bne miss2
- mov r0, r12
- bx lr
-FUNCTION_END(atomic_and)
-
-/* int atomic_or(int *value, int orValue)
- */
-FUNCTION(atomic_or):
-miss3: ldrex r12, [r0]
- eor r2, r12, r1
- strex r3, r2, [r0]
- teq r3, #0
- bne miss3
- mov r0, r12
- bx lr
-FUNCTION_END(atomic_or)
-
-/* void atomic_set(int *value, int setTo)
- */
-FUNCTION(atomic_set):
- str r1, [r0]
- bx lr
-FUNCTION_END(atomic_set)
-
-/* int atomic_test_and_set(int *value, int setTo, int testValue)
- */
-FUNCTION(atomic_test_and_set):
-miss5: ldrex r12, [r0] @ load from the address
and mark it exclusive
- cmp r12, r2 @ compare the value with
the comperand(r2)
- strexeq r3, r1, [r0] @ if they were equal,
attempt to store the new value (r1)
- bne differ @ if they were not equal
jump to (differ) which clears the exclusive tag on the address and returns<
- cmp r3, #1 @ check the status of the
store (returned in r3)
- beq miss5 @ go back to the start if
it failed (0=success, 1=failure)
- bne same @ if it succeeded, jump to
(same) and return. there is no need to clrex if strex succeeded
-differ: clrex @ clrex
-same: mov r0, r12
- bx lr
-FUNCTION_END(atomic_test_and_set)
-
-/* int atomic_get(int *value)
- */
-FUNCTION(atomic_get):
- ldr r0, [r0]
- bx lr
-FUNCTION_END(atomic_get)
-
-#endif /* ATOMIC_FUNCS_ARE_SYSCALLS */
-
-#ifndef ATOMIC64_FUNCS_ARE_SYSCALLS
-
-/* int64 atomic_add64(vint64 *value, int64 addValue) */
-FUNCTION(atomic_add64):
- bx lr
-FUNCTION_END(atomic_add64)
-
-/* int64 atomic_and64(vint64 *value, int64 andValue) */
-FUNCTION(atomic_and64):
- bx lr
-FUNCTION_END(atomic_and64)
-
-/* int64 atomic_or64(vint64 *value, int64 orValue) */
-FUNCTION(atomic_or64):
- bx lr
-FUNCTION_END(atomic_or64)
-
-/* int64 atomic_set64(vint64 *value, int64 newValue) */
-FUNCTION(atomic_set64):
- bx lr
-FUNCTION_END(atomic_set64)
-
-/* int64 atomic_test_and_set64(vint64 *value, int64 newValue, int64
testAgainst) */
-FUNCTION(atomic_test_and_set64):
- bx lr
-FUNCTION_END(atomic_test_and_set64)
-
-/* int64 atomic_get64(vint64 *value) */
-FUNCTION(atomic_get64):
- bx lr
-FUNCTION_END(atomic_get64)
-
-#endif /* ATOMIC64_FUNCS_ARE_SYSCALLS */
diff --git a/src/system/libroot/os/arch/x86_64/Jamfile
b/src/system/libroot/os/arch/x86_64/Jamfile
index 26b8c27..5aad005 100644
--- a/src/system/libroot/os/arch/x86_64/Jamfile
+++ b/src/system/libroot/os/arch/x86_64/Jamfile
@@ -12,8 +12,9 @@ for architectureObject in [ MultiArchSubDirSetup x86_64 ] {
# TODO in time.c!
UsePrivateSystemHeaders ;
+ SEARCH_SOURCE += [ FDirName $(SUBDIR) $(DOTDOT) generic ] ;
+
MergeObject <$(architecture)>os_arch_$(TARGET_ARCH).o :
- atomic.S
byteorder.S
get_stack_frame.S
system_info.cpp
@@ -21,6 +22,8 @@ for architectureObject in [ MultiArchSubDirSetup x86_64 ] {
thread.cpp
time.cpp
tls.cpp
+
+ generic_atomic.cpp
;
}
}
diff --git a/src/system/libroot/os/arch/x86_64/atomic.S
b/src/system/libroot/os/arch/x86_64/atomic.S
deleted file mode 100644
index b88af32..0000000
--- a/src/system/libroot/os/arch/x86_64/atomic.S
+++ /dev/null
@@ -1,153 +0,0 @@
-/*
- * Copyright 2013, Paweł Dziepak, pdziepak@xxxxxxxxxxx.
- * Copyright 2012, Alex Smith, alex@xxxxxxxxxxxxxxxx.
- * Distributed under the terms of the MIT License.
- */
-
-
-#include <asm_defs.h>
-
-
-.text
-
-
-/* int32 atomic_set(int32* value, int32 newValue) */
-FUNCTION(atomic_set):
- sfence
- movl %esi, (%rdi)
- ret
-FUNCTION_END(atomic_set)
-
-
-/* int32 atomic_get_and_set(int32* value, int32 newValue) */
-FUNCTION(atomic_get_and_set):
- movl %esi, %eax
- xchgl %eax, (%rdi)
- ret
-FUNCTION_END(atomic_get_and_set)
-
-
-/* int32 atomic_test_and_set(int32* value, int32 newValue, int32 testAgainst)
*/
-FUNCTION(atomic_test_and_set):
- movl %edx, %eax
- lock
- cmpxchgl %esi, (%rdi)
- ret
-FUNCTION_END(atomic_test_and_set)
-
-
-/* int32 atomic_add(int32* value, int32 addValue) */
-FUNCTION(atomic_add):
- movl %esi, %eax
- lock
- xaddl %eax, (%rdi)
- ret
-FUNCTION_END(atomic_add)
-
-
-/* int32 atomic_and(int32* value, int32 andValue) */
-FUNCTION(atomic_and):
- movl (%rdi), %eax
-1: movl %eax, %edx
- movl %eax, %ecx
- andl %esi, %edx
- lock
- cmpxchgl %edx, (%rdi)
- jnz 1b
- movl %ecx, %eax
- ret
-FUNCTION_END(atomic_and)
-
-
-/* int32 atomic_or(int32* value, int32 orValue) */
-FUNCTION(atomic_or):
- movl (%rdi), %eax
-1: movl %eax, %edx
- movl %eax, %ecx
- orl %esi, %edx
- lock
- cmpxchgl %edx, (%rdi)
- jnz 1b
- movl %ecx, %eax
- ret
-FUNCTION_END(atomic_or)
-
-
-/* int32 atomic_get(int32* value) */
-FUNCTION(atomic_get):
- movl (%rdi), %eax
- lfence
- ret
-FUNCTION_END(atomic_get)
-
-
-/* int64 atomic_set64(int64* value, int64 newValue) */
-FUNCTION(atomic_set64):
- sfence
- movq %rsi, (%rdi)
- ret
-FUNCTION_END(atomic_set64)
-
-
-/* int64 atomic_get_and_set64(int64* value, int64 newValue) */
-FUNCTION(atomic_get_and_set64):
- movq %rsi, %rax
- xchgq %rax, (%rdi)
- ret
-FUNCTION_END(atomic_get_and_set64)
-
-
-/* int64 atomic_test_and_set64(int64* value, int64 newValue,
- int64 testAgainst) */
-FUNCTION(atomic_test_and_set64):
- movq %rdx, %rax
- lock
- cmpxchgq %rsi, (%rdi)
- ret
-FUNCTION_END(atomic_test_and_set64)
-
-
-/* int64 atomic_add64(int64* value, int64 addValue) */
-FUNCTION(atomic_add64):
- movq %rsi, %rax
- lock
- xaddq %rax, (%rdi)
- ret
-FUNCTION_END(atomic_add64)
-
-
-/* int64 atomic_and64(int64* value, int64 andValue) */
-FUNCTION(atomic_and64):
- movq (%rdi), %rax
-1: movq %rax, %rdx
- movq %rax, %rcx
- andq %rsi, %rdx
- lock
- cmpxchgq %rdx, (%rdi)
- jnz 1b
- movq %rcx, %rax
- ret
-FUNCTION_END(atomic_and64)
-
-
-/* int64 atomic_or64(int64* value, int64 orValue) */
-FUNCTION(atomic_or64):
- movq (%rdi), %rax
-1: movq %rax, %rdx
- movq %rax, %rcx
- orq %rsi, %rdx
- lock
- cmpxchgq %rdx, (%rdi)
- jnz 1b
- movq %rcx, %rax
- ret
-FUNCTION_END(atomic_or64)
-
-
-/* int64 atomic_get64(int64* value) */
-FUNCTION(atomic_get64):
- movq (%rdi), %rax
- lfence
- ret
-FUNCTION_END(atomic_get64)
-
diff --git a/src/system/runtime_loader/arch/arm/Jamfile
b/src/system/runtime_loader/arch/arm/Jamfile
index afc33d3..86d9d2b 100644
--- a/src/system/runtime_loader/arch/arm/Jamfile
+++ b/src/system/runtime_loader/arch/arm/Jamfile
@@ -10,7 +10,6 @@ SubDirHdrs [ FDirName $(SUBDIR) $(DOTDOT) $(DOTDOT) ] ;
StaticLibrary libruntime_loader_$(TARGET_ARCH).a :
arch_relocate.cpp
:
- <src!system!libroot!os!arch!$(TARGET_ARCH)!$(architecture)>atomic.o
<src!system!libroot!os!arch!$(TARGET_ARCH)!$(architecture)>thread.o
<src!system!libroot!posix!string!arch!$(TARGET_ARCH)!$(architecture)>arch_string.o
diff --git a/src/system/runtime_loader/arch/x86_64/Jamfile
b/src/system/runtime_loader/arch/x86_64/Jamfile
index b9baa79..42e72d4 100644
--- a/src/system/runtime_loader/arch/x86_64/Jamfile
+++ b/src/system/runtime_loader/arch/x86_64/Jamfile
@@ -10,7 +10,6 @@ SubDirHdrs [ FDirName $(SUBDIR) $(DOTDOT) $(DOTDOT) ] ;
StaticLibrary libruntime_loader_$(TARGET_ARCH).a :
arch_relocate.cpp
:
- <src!system!libroot!os!arch!$(TARGET_ARCH)!$(architecture)>atomic.o
<src!system!libroot!os!arch!$(TARGET_ARCH)!$(architecture)>thread.o
<src!system!libroot!posix!string!arch!$(TARGET_ARCH)!$(architecture)>arch_string.o
;
############################################################################
Commit: a4cdc6072cafccc8bd7d3c4a4458bf72ff68272c
URL: http://cgit.haiku-os.org/haiku/commit/?id=a4cdc60
Author: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
Date: Mon Aug 25 13:44:12 2014 UTC
build: remove B_USE_BUILTIN_ATOMIC_FUNCTIONS
No reason not to use GCC atomic support on non-x86 archs.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
----------------------------------------------------------------------------
diff --git a/headers/os/support/SupportDefs.h b/headers/os/support/SupportDefs.h
index b863334..7e47018 100644
--- a/headers/os/support/SupportDefs.h
+++ b/headers/os/support/SupportDefs.h
@@ -215,8 +215,7 @@ extern void* get_stack_frame(void);
/* Use the built-in atomic functions, if requested and available. */
-#if defined(B_USE_BUILTIN_ATOMIC_FUNCTIONS) && __GNUC__ > 4 \
- || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7)
+#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7)
static inline void
@@ -321,8 +320,7 @@ atomic_get64(int64* value)
}
-#else // B_USE_BUILTIN_ATOMIC_FUNCTIONS && __GNUC__ > 4
- // || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7)
+#else // __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7)
#ifdef __cplusplus
extern "C" {
############################################################################
Revision: hrev47741
Commit: 721a07ac24df63af87afd8d52f5e08ec6566c9e0
URL: http://cgit.haiku-os.org/haiku/commit/?id=721a07a
Author: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
Date: Mon Aug 25 13:45:26 2014 UTC
libroot: remove ATOMIC_FUNCS_ARE_SYSCALLS
GCC knows whether these functions need to be implemented using syscalls
(or more clever solutions like in Linux) and calls libgcc in such case.
Signed-off-by: Paweł Dziepak <pdziepak@xxxxxxxxxxx>
----------------------------------------------------------------------------
diff --git a/src/system/libroot/os/arch/generic/generic_atomic.cpp
b/src/system/libroot/os/arch/generic/generic_atomic.cpp
index dce3a5b..cc21f1d 100644
--- a/src/system/libroot/os/arch/generic/generic_atomic.cpp
+++ b/src/system/libroot/os/arch/generic/generic_atomic.cpp
@@ -7,8 +7,6 @@
#include <atomic>
#include <sys/types.h>
-#ifndef ATOMIC_FUNCS_ARE_SYSCALLS
-
extern "C" [[gnu::optimize("omit-frame-pointer")]] void
atomic_set(int32_t* ptr, int32_t value)
@@ -123,5 +121,3 @@ atomic_get64(int64_t* ptr)
return obj.load(std::memory_order_acquire);
}
-
-#endif /* ATOMIC64_FUNCS_ARE_SYSCALLS */
Other related posts:
- » [haiku-commits] haiku: hrev47741 - src/system/libroot/os/arch/arm src/system/libroot/os/arch/generic src/system/libroot/os/arch/x86_64 headers/os/support headers/private/kernel/arch/x86/64 - pdziepak
