x86: Remove arch-specific low level lock implementation

This patch removes the arch-specific x86 assembly implementation for
low level locking and consolidate both 64 bits and 32 bits in a
single implementation.

Different than other architectures, x86 lll_trylock, lll_lock, and
lll_unlock implements a single-thread optimization to avoid atomic
operation, using cmpxchgl instead.  This patch implements by using
the new single-thread.h definitions in a generic way, although using
the previous semantic.

The lll_cond_trylock, lll_cond_lock, and lll_timedlock just use
atomic operations plus calls to lll_lock_wait*.

For __lll_lock_wait_private and __lll_lock_wait the generic implemtation
there is no indication that assembly implementation is required
performance-wise.

Checked on x86_64-linux-gnu and i686-linux-gnu.

	* sysdeps/nptl/lowlevellock.h (__lll_trylock): New macro.
	(lll_trylock): Call __lll_trylock.
	* sysdeps/unix/sysv/linux/i386/libc-lowlevellock.S: Remove file.
	* sysdeps/unix/sysv/linux/i386/lll_timedlock_wait.c: Likewise.
	* sysdeps/unix/sysv/linux/i386/lowlevellock.S: Likewise.
	* sysdeps/unix/sysv/linux/i386/lowlevellock.h: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/libc-lowlevellock.S: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/lll_timedlock_wait.c: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/lowlevellock.S: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/lowlevellock.h: Likewise.
	* sysdeps/unix/sysv/linux/x86/lowlevellock.h: New file.
	* sysdeps/unix/sysv/linux/x86_64/cancellation.S: Include
	lowlevellock-futex.h.

1 files changed, 110 insertions, 0 deletions

diff --git a/sysdeps/unix/sysv/linux/x86/lowlevellock.h b/sysdeps/unix/sysv/linux/x86/lowlevellock.h
new file mode 100644
index 0000000000..8a78dcff47
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/x86/lowlevellock.h
@@ -0,0 +1,110 @@
+/* Low-level lock implementation, x86 version.
+   Copyright (C) 2019 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _X86_64_LOWLEVELLOCK_H
+#define _X86_64_LOWLEVELLOCK_H	1
+
+#ifndef __ASSEMBLER__
+#include <sysdeps/nptl/lowlevellock.h>
+#include <single-thread.h>
+
+/* The lll_trylock, lll_lock, and lll_unlock implements a single-thread
+   optimization using the cmpxchgl instruction.  It checks if the process
+   is single thread and avoid a more expensive atomic instruction.  */
+
+/* The single-thread optimization only works for libc itself, we need
+   atomicity for libpthread in case of shared futexes.  */
+#if !IS_IN(libc)
+# define is_single_thread 0
+#else
+# define is_single_thread SINGLE_THREAD_P
+#endif
+
+/* In the __lllc_as we simply return the value in %eax after the cmpxchg
+   instruction.  In case the operation succeeded this value is zero.  In case
+   the operation failed, the cmpxchg instruction has loaded the current value
+   of the memory work which is guaranteed to be nonzero.  */
+static inline int
+__attribute__ ((always_inline))
+__lll_cas_lock (int *futex)
+{
+  int ret;
+  asm volatile ("cmpxchgl %2, %1"
+		: "=a" (ret), "=m" (*futex)
+		: "r" (1), "m" (*futex), "0" (0)
+		: "memory");
+  return ret;
+}
+
+#undef lll_trylock
+#define lll_trylock(lock)						     \
+  ({									     \
+    int __ret;								     \
+    if (is_single_thread)						     \
+      __ret = __lll_cas_lock (&(lock));					     \
+    else								     \
+      __ret = __lll_trylock (&(lock));					     \
+    __ret;								     \
+  })
+
+#undef lll_lock
+#define lll_lock(lock, private)						     \
+  ((void)								     \
+   ({									     \
+     if (is_single_thread)						     \
+       __lll_cas_lock (&(lock));					     \
+     else								     \
+       __lll_lock (&(lock), private);					     \
+   }))
+
+#undef lll_unlock
+#define lll_unlock(lock, private)					     \
+  ((void)								     \
+   ({									     \
+     if (is_single_thread)						     \
+       (lock)--;							     \
+     else								     \
+       __lll_unlock (&(lock), private);					     \
+   }))
+
+extern int __lll_timedlock_elision (int *futex, short *adapt_count,
+					 const struct timespec *timeout,
+					 int private) attribute_hidden;
+
+#define lll_timedlock_elision(futex, adapt_count, timeout, private)	\
+  __lll_timedlock_elision(&(futex), &(adapt_count), timeout, private)
+
+extern int __lll_lock_elision (int *futex, short *adapt_count, int private)
+  attribute_hidden;
+
+extern int __lll_unlock_elision (int *lock, int private)
+  attribute_hidden;
+
+extern int __lll_trylock_elision (int *lock, short *adapt_count)
+  attribute_hidden;
+
+#define lll_lock_elision(futex, adapt_count, private) \
+  __lll_lock_elision (&(futex), &(adapt_count), private)
+#define lll_unlock_elision(futex, adapt_count, private) \
+  __lll_unlock_elision (&(futex), private)
+#define lll_trylock_elision(futex, adapt_count) \
+  __lll_trylock_elision (&(futex), &(adapt_count))
+
+#endif  /* !__ASSEMBLER__ */
+
+#endif	/* lowlevellock.h */