x86-32: Rewrite 32-bit atomic64 functions in assembly

This patch replaces atomic64_32.c with two assembly implementations,
one for 386/486 machines using pushf/cli/popf and one for 586+ machines
using cmpxchg8b.

The cmpxchg8b implementation provides the following advantages over the
current one:

1. Implements atomic64_add_unless, atomic64_dec_if_positive and
   atomic64_inc_not_zero

2. Uses the ZF flag changed by cmpxchg8b instead of doing a comparison

3. Uses custom register calling conventions that reduce or eliminate
   register moves to suit cmpxchg8b

4. Reads the initial value instead of using cmpxchg8b to do that.
   Currently we use lock xaddl and movl, which seems the fastest.

5. Does not use the lock prefix for atomic64_set
   64-bit writes are already atomic, so we don't need that.
   We still need it for atomic64_read to avoid restoring a value
   changed in the meantime.

6. Allocates registers as well or better than gcc

The 386 implementation provides support for 386 and 486 machines.
386/486 SMP is not supported (we dropped it), but such support can be
added easily if desired.

A pure assembly implementation is required due to the custom calling
conventions, and desire to use %ebp in atomic64_add_return (we need
7 registers...), as well as the ability to use pushf/popf in the 386
code without an intermediate pop/push.

The parameter names are changed to match the convention in atomic_64.h

Changes in v3 (due to rebasing to tip/x86/asm):
- Patches atomic64_32.h instead of atomic_32.h
- Uses the CALL alternative mechanism from commit
  1b1d9258181bae199dc940f4bd0298126b9a73d9

Changes in v2:
- Merged 386 and cx8 support in the same patch
- 386 support now done in assembly, C code no longer used at all
- cmpxchg64 is used for atomic64_cmpxchg
- stop using macros, use one-line inline functions instead
- miscellanous changes and improvements

Signed-off-by: Luca Barbieri <luca@luca-barbieri.com>
LKML-Reference: <1267005265-27958-5-git-send-email-luca@luca-barbieri.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>

1 files changed, 225 insertions, 0 deletions

diff --git a/arch/x86/lib/atomic64_cx8_32.S b/arch/x86/lib/atomic64_cx8_32.S
new file mode 100644
index 000000000000..e49c4ebca9f4
--- /dev/null
+++ b/arch/x86/lib/atomic64_cx8_32.S
@@ -0,0 +1,225 @@
+/*
+ * atomic64_t for 586+
+ *
+ * Copyright © 2010  Luca Barbieri
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ */
+
+#include <linux/linkage.h>
+#include <asm/alternative-asm.h>
+#include <asm/dwarf2.h>
+
+.macro SAVE reg
+	pushl %\reg
+	CFI_ADJUST_CFA_OFFSET 4
+	CFI_REL_OFFSET \reg, 0
+.endm
+
+.macro RESTORE reg
+	popl %\reg
+	CFI_ADJUST_CFA_OFFSET -4
+	CFI_RESTORE \reg
+.endm
+
+.macro read64 reg
+	movl %ebx, %eax
+	movl %ecx, %edx
+/* we need LOCK_PREFIX since otherwise cmpxchg8b always does the write */
+	LOCK_PREFIX
+	cmpxchg8b (\reg)
+.endm
+
+ENTRY(atomic64_read_cx8)
+	CFI_STARTPROC
+
+	read64 %ecx
+	ret
+	CFI_ENDPROC
+ENDPROC(atomic64_read_cx8)
+
+ENTRY(atomic64_set_cx8)
+	CFI_STARTPROC
+
+1:
+/* we don't need LOCK_PREFIX since aligned 64-bit writes
+ * are atomic on 586 and newer */
+	cmpxchg8b (%esi)
+	jne 1b
+
+	ret
+	CFI_ENDPROC
+ENDPROC(atomic64_set_cx8)
+
+ENTRY(atomic64_xchg_cx8)
+	CFI_STARTPROC
+
+	movl %ebx, %eax
+	movl %ecx, %edx
+1:
+	LOCK_PREFIX
+	cmpxchg8b (%esi)
+	jne 1b
+
+	ret
+	CFI_ENDPROC
+ENDPROC(atomic64_xchg_cx8)
+
+.macro addsub_return func ins insc
+ENTRY(atomic64_\func\()_return_cx8)
+	CFI_STARTPROC
+	SAVE ebp
+	SAVE ebx
+	SAVE esi
+	SAVE edi
+
+	movl %eax, %esi
+	movl %edx, %edi
+	movl %ecx, %ebp
+
+	read64 %ebp
+1:
+	movl %eax, %ebx
+	movl %edx, %ecx
+	\ins\()l %esi, %ebx
+	\insc\()l %edi, %ecx
+	LOCK_PREFIX
+	cmpxchg8b (%ebp)
+	jne 1b
+
+10:
+	movl %ebx, %eax
+	movl %ecx, %edx
+	RESTORE edi
+	RESTORE esi
+	RESTORE ebx
+	RESTORE ebp
+	ret
+	CFI_ENDPROC
+ENDPROC(atomic64_\func\()_return_cx8)
+.endm
+
+addsub_return add add adc
+addsub_return sub sub sbb
+
+.macro incdec_return func ins insc
+ENTRY(atomic64_\func\()_return_cx8)
+	CFI_STARTPROC
+	SAVE ebx
+
+	read64 %esi
+1:
+	movl %eax, %ebx
+	movl %edx, %ecx
+	\ins\()l $1, %ebx
+	\insc\()l $0, %ecx
+	LOCK_PREFIX
+	cmpxchg8b (%esi)
+	jne 1b
+
+10:
+	movl %ebx, %eax
+	movl %ecx, %edx
+	RESTORE ebx
+	ret
+	CFI_ENDPROC
+ENDPROC(atomic64_\func\()_return_cx8)
+.endm
+
+incdec_return inc add adc
+incdec_return dec sub sbb
+
+ENTRY(atomic64_dec_if_positive_cx8)
+	CFI_STARTPROC
+	SAVE ebx
+
+	read64 %esi
+1:
+	movl %eax, %ebx
+	movl %edx, %ecx
+	subl $1, %ebx
+	sbb $0, %ecx
+	js 2f
+	LOCK_PREFIX
+	cmpxchg8b (%esi)
+	jne 1b
+
+2:
+	movl %ebx, %eax
+	movl %ecx, %edx
+	RESTORE ebx
+	ret
+	CFI_ENDPROC
+ENDPROC(atomic64_dec_if_positive_cx8)
+
+ENTRY(atomic64_add_unless_cx8)
+	CFI_STARTPROC
+	SAVE ebp
+	SAVE ebx
+/* these just push these two parameters on the stack */
+	SAVE edi
+	SAVE esi
+
+	movl %ecx, %ebp
+	movl %eax, %esi
+	movl %edx, %edi
+
+	read64 %ebp
+1:
+	cmpl %eax, 0(%esp)
+	je 4f
+2:
+	movl %eax, %ebx
+	movl %edx, %ecx
+	addl %esi, %ebx
+	adcl %edi, %ecx
+	LOCK_PREFIX
+	cmpxchg8b (%ebp)
+	jne 1b
+
+	xorl %eax, %eax
+3:
+	addl $8, %esp
+	CFI_ADJUST_CFA_OFFSET -8
+	RESTORE ebx
+	RESTORE ebp
+	ret
+4:
+	cmpl %edx, 4(%esp)
+	jne 2b
+	movl $1, %eax
+	jmp 3b
+	CFI_ENDPROC
+ENDPROC(atomic64_add_unless_cx8)
+
+ENTRY(atomic64_inc_not_zero_cx8)
+	CFI_STARTPROC
+	SAVE ebx
+
+	read64 %esi
+1:
+	testl %eax, %eax
+	je 4f
+2:
+	movl %eax, %ebx
+	movl %edx, %ecx
+	addl $1, %ebx
+	adcl $0, %ecx
+	LOCK_PREFIX
+	cmpxchg8b (%esi)
+	jne 1b
+
+	xorl %eax, %eax
+3:
+	RESTORE ebx
+	ret
+4:
+	testl %edx, %edx
+	jne 2b
+	movl $1, %eax
+	jmp 3b
+	CFI_ENDPROC
+ENDPROC(atomic64_inc_not_zero_cx8)