blob: 05a5385312294582a082592838f600c4b9ab7a32 [file] [log] [blame]
Arjan van de Ven2af7f592006-01-09 15:59:18 -08001/*
2 * Assembly implementation of the mutex fastpath, based on atomic
3 * decrement/increment.
4 *
5 * started by Ingo Molnar:
6 *
7 * Copyright (C) 2004, 2005, 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
8 */
9#ifndef _ASM_MUTEX_H
10#define _ASM_MUTEX_H
11
Gerd Hoffmann9a0b5812006-03-23 02:59:32 -080012#include "asm/alternative.h"
13
Arjan van de Ven2af7f592006-01-09 15:59:18 -080014/**
15 * __mutex_fastpath_lock - try to take the lock by moving the count
16 * from 1 to a 0 value
17 * @count: pointer of type atomic_t
18 * @fn: function to call if the original value was not 1
19 *
20 * Change the count from 1 to a value lower than 1, and call <fn> if it
21 * wasn't 1 originally. This function MUST leave the value lower than 1
22 * even when the "1" assertion wasn't true.
23 */
24#define __mutex_fastpath_lock(count, fail_fn) \
25do { \
26 unsigned int dummy; \
27 \
28 typecheck(atomic_t *, count); \
29 typecheck_fn(fastcall void (*)(atomic_t *), fail_fn); \
30 \
31 __asm__ __volatile__( \
Gerd Hoffmann9a0b5812006-03-23 02:59:32 -080032 LOCK_PREFIX " decl (%%eax) \n" \
Ingo Molnar73165b82006-01-10 22:07:44 +010033 " js 2f \n" \
34 "1: \n" \
35 \
36 LOCK_SECTION_START("") \
37 "2: call "#fail_fn" \n" \
38 " jmp 1b \n" \
39 LOCK_SECTION_END \
Arjan van de Ven2af7f592006-01-09 15:59:18 -080040 \
41 :"=a" (dummy) \
42 : "a" (count) \
43 : "memory", "ecx", "edx"); \
44} while (0)
45
46
47/**
48 * __mutex_fastpath_lock_retval - try to take the lock by moving the count
49 * from 1 to a 0 value
50 * @count: pointer of type atomic_t
51 * @fail_fn: function to call if the original value was not 1
52 *
53 * Change the count from 1 to a value lower than 1, and call <fail_fn> if it
54 * wasn't 1 originally. This function returns 0 if the fastpath succeeds,
55 * or anything the slow path function returns
56 */
57static inline int
58__mutex_fastpath_lock_retval(atomic_t *count,
59 int fastcall (*fail_fn)(atomic_t *))
60{
61 if (unlikely(atomic_dec_return(count) < 0))
62 return fail_fn(count);
63 else
64 return 0;
65}
66
67/**
68 * __mutex_fastpath_unlock - try to promote the mutex from 0 to 1
69 * @count: pointer of type atomic_t
70 * @fail_fn: function to call if the original value was not 0
71 *
72 * try to promote the mutex from 0 to 1. if it wasn't 0, call <fail_fn>.
73 * In the failure case, this function is allowed to either set the value
74 * to 1, or to set it to a value lower than 1.
75 *
76 * If the implementation sets it to a value of lower than 1, the
77 * __mutex_slowpath_needs_to_unlock() macro needs to return 1, it needs
78 * to return 0 otherwise.
79 */
80#define __mutex_fastpath_unlock(count, fail_fn) \
81do { \
82 unsigned int dummy; \
83 \
84 typecheck(atomic_t *, count); \
85 typecheck_fn(fastcall void (*)(atomic_t *), fail_fn); \
86 \
87 __asm__ __volatile__( \
Gerd Hoffmann9a0b5812006-03-23 02:59:32 -080088 LOCK_PREFIX " incl (%%eax) \n" \
Ingo Molnar73165b82006-01-10 22:07:44 +010089 " jle 2f \n" \
90 "1: \n" \
91 \
92 LOCK_SECTION_START("") \
93 "2: call "#fail_fn" \n" \
94 " jmp 1b \n" \
95 LOCK_SECTION_END \
Arjan van de Ven2af7f592006-01-09 15:59:18 -080096 \
97 :"=a" (dummy) \
98 : "a" (count) \
99 : "memory", "ecx", "edx"); \
100} while (0)
101
102#define __mutex_slowpath_needs_to_unlock() 1
103
104/**
105 * __mutex_fastpath_trylock - try to acquire the mutex, without waiting
106 *
107 * @count: pointer of type atomic_t
108 * @fail_fn: fallback function
109 *
110 * Change the count from 1 to a value lower than 1, and return 0 (failure)
111 * if it wasn't 1 originally, or return 1 (success) otherwise. This function
112 * MUST leave the value lower than 1 even when the "1" assertion wasn't true.
113 * Additionally, if the value was < 0 originally, this function must not leave
114 * it to 0 on failure.
115 */
116static inline int
117__mutex_fastpath_trylock(atomic_t *count, int (*fail_fn)(atomic_t *))
118{
119 /*
120 * We have two variants here. The cmpxchg based one is the best one
121 * because it never induce a false contention state. It is included
122 * here because architectures using the inc/dec algorithms over the
123 * xchg ones are much more likely to support cmpxchg natively.
124 *
125 * If not we fall back to the spinlock based variant - that is
126 * just as efficient (and simpler) as a 'destructive' probing of
127 * the mutex state would be.
128 */
129#ifdef __HAVE_ARCH_CMPXCHG
Linus Torvalds4cec8732006-01-11 15:50:47 -0800130 if (likely(atomic_cmpxchg(count, 1, 0) == 1))
Arjan van de Ven2af7f592006-01-09 15:59:18 -0800131 return 1;
132 return 0;
133#else
134 return fail_fn(count);
135#endif
136}
137
138#endif