Clean up BUSY_WAIT_NOP and atomic_delay.

This patch combines BUSY_WAIT_NOP and atomic_delay into a new
atomic_spin_nop function and adjusts all clients.  The new function is
put into atomic.h because what is best done in a spin loop is
architecture-specific, and atomics must be used for spinning.  The
function name is meant to tell users that this has no effect on
synchronization semantics but is a performance aid for spinning.

9 files changed, 10 insertions, 21 deletions

diff --git a/sysdeps/i386/i486/bits/atomic.h b/sysdeps/i386/i486/bits/atomic.h
index 59165bec94..59f3d34871 100644
--- a/sysdeps/i386/i486/bits/atomic.h
+++ b/sysdeps/i386/i486/bits/atomic.h
@@ -479,7 +479,7 @@ typedef uintmax_t uatomic_max_t;
      __result; })
 
 
-#define atomic_delay() asm ("rep; nop")
+#define atomic_spin_nop() asm ("rep; nop")
 
 
 #define __arch_and_body(lock, mem, mask) \
diff --git a/sysdeps/nacl/lll_timedwait_tid.c b/sysdeps/nacl/lll_timedwait_tid.c
index ecaf0b113a..ef544cf84f 100644
--- a/sysdeps/nacl/lll_timedwait_tid.c
+++ b/sysdeps/nacl/lll_timedwait_tid.c
@@ -40,7 +40,7 @@ __lll_timedwait_tid (int *tidp, const struct timespec *abstime)
 	   finish quick enough that the timeout doesn't matter.  If any
 	   thread ever stays in this state for long, there is something
 	   catastrophically wrong.  */
-	BUSY_WAIT_NOP;
+	atomic_spin_nop ();
       else
 	{
 	  assert (tid > 0);
diff --git a/sysdeps/nacl/lowlevellock.h b/sysdeps/nacl/lowlevellock.h
index 0b85d8d317..3634f1959a 100644
--- a/sysdeps/nacl/lowlevellock.h
+++ b/sysdeps/nacl/lowlevellock.h
@@ -21,10 +21,6 @@
 /* Everything except the exit handling is the same as the generic code.  */
 # include <sysdeps/nptl/lowlevellock.h>
 
-# ifndef BUSY_WAIT_NOP
-#  define BUSY_WAIT_NOP		__sync_synchronize ()
-# endif
-
 /* See exit-thread.h for details.  */
 # define NACL_EXITING_TID	1
 
@@ -36,7 +32,7 @@
     while ((__tid = atomic_load_relaxed (__tidp)) != 0) \
       {							\
 	if (__tid == NACL_EXITING_TID)			\
-	  BUSY_WAIT_NOP;				\
+	  atomic_spin_nop ();				\
 	else						\
 	  lll_futex_wait (__tidp, __tid, LLL_PRIVATE);	\
       }							\
diff --git a/sysdeps/sparc/sparc32/sparcv9/bits/atomic.h b/sysdeps/sparc/sparc32/sparcv9/bits/atomic.h
index 317be62ccb..2122afbb09 100644
--- a/sysdeps/sparc/sparc32/sparcv9/bits/atomic.h
+++ b/sysdeps/sparc/sparc32/sparcv9/bits/atomic.h
@@ -100,3 +100,6 @@ typedef uintmax_t uatomic_max_t;
   __asm __volatile ("membar #LoadLoad | #LoadStore" : : : "memory")
 #define atomic_write_barrier() \
   __asm __volatile ("membar #LoadStore | #StoreStore" : : : "memory")
+
+extern void __cpu_relax (void);
+#define atomic_spin_nop () __cpu_relax ()
diff --git a/sysdeps/sparc/sparc64/bits/atomic.h b/sysdeps/sparc/sparc64/bits/atomic.h
index 35804a8e14..48b7fd6216 100644
--- a/sysdeps/sparc/sparc64/bits/atomic.h
+++ b/sysdeps/sparc/sparc64/bits/atomic.h
@@ -121,3 +121,6 @@ typedef uintmax_t uatomic_max_t;
   __asm __volatile ("membar #LoadLoad | #LoadStore" : : : "memory")
 #define atomic_write_barrier() \
   __asm __volatile ("membar #LoadStore | #StoreStore" : : : "memory")
+
+extern void __cpu_relax (void);
+#define atomic_spin_nop () __cpu_relax ()
diff --git a/sysdeps/unix/sysv/linux/i386/lowlevellock.h b/sysdeps/unix/sysv/linux/i386/lowlevellock.h
index f57afc6e2a..58f5638e37 100644
--- a/sysdeps/unix/sysv/linux/i386/lowlevellock.h
+++ b/sysdeps/unix/sysv/linux/i386/lowlevellock.h
@@ -58,10 +58,6 @@
 #define LLL_LOCK_INITIALIZER_WAITERS	(2)
 
 
-/* Delay in spinlock loop.  */
-#define BUSY_WAIT_NOP	asm ("rep; nop")
-
-
 /* NB: in the lll_trylock macro we simply return the value in %eax
    after the cmpxchg instruction.  In case the operation succeded this
    value is zero.  In case the operation failed, the cmpxchg instruction
diff --git a/sysdeps/unix/sysv/linux/sparc/lowlevellock.h b/sysdeps/unix/sysv/linux/sparc/lowlevellock.h
index 9aefd9eb59..7608c01d17 100644
--- a/sysdeps/unix/sysv/linux/sparc/lowlevellock.h
+++ b/sysdeps/unix/sysv/linux/sparc/lowlevellock.h
@@ -25,12 +25,6 @@
 #include <atomic.h>
 #include <kernel-features.h>
 
-#ifndef __sparc32_atomic_do_lock
-/* Delay in spinlock loop.  */
-extern void __cpu_relax (void);
-#define BUSY_WAIT_NOP	__cpu_relax ()
-#endif
-
 #include <lowlevellock-futex.h>
 
 static inline int
diff --git a/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h b/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
index 573b48c4fe..de525cd4c7 100644
--- a/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
+++ b/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
@@ -57,9 +57,6 @@
 #define LLL_LOCK_INITIALIZER_LOCKED	(1)
 #define LLL_LOCK_INITIALIZER_WAITERS	(2)
 
-/* Delay in spinlock loop.  */
-#define BUSY_WAIT_NOP	  asm ("rep; nop")
-
 
 /* NB: in the lll_trylock macro we simply return the value in %eax
    after the cmpxchg instruction.  In case the operation succeded this
diff --git a/sysdeps/x86_64/bits/atomic.h b/sysdeps/x86_64/bits/atomic.h
index 203d92c20d..337b334db1 100644
--- a/sysdeps/x86_64/bits/atomic.h
+++ b/sysdeps/x86_64/bits/atomic.h
@@ -410,7 +410,7 @@ typedef uintmax_t uatomic_max_t;
      __result; })
 
 
-#define atomic_delay() asm ("rep; nop")
+#define atomic_spin_nop() asm ("rep; nop")
 
 
 #define __arch_and_body(lock, mem, mask) \