Commit | Line | Data |
---|---|---|
2af7f59e AV |
1 | /* |
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 | ||
12 | /** | |
13 | * __mutex_fastpath_lock - try to take the lock by moving the count | |
14 | * from 1 to a 0 value | |
15 | * @count: pointer of type atomic_t | |
16 | * @fn: function to call if the original value was not 1 | |
17 | * | |
18 | * Change the count from 1 to a value lower than 1, and call <fn> if it | |
19 | * wasn't 1 originally. This function MUST leave the value lower than 1 | |
20 | * even when the "1" assertion wasn't true. | |
21 | */ | |
22 | #define __mutex_fastpath_lock(count, fail_fn) \ | |
23 | do { \ | |
24 | unsigned int dummy; \ | |
25 | \ | |
26 | typecheck(atomic_t *, count); \ | |
27 | typecheck_fn(fastcall void (*)(atomic_t *), fail_fn); \ | |
28 | \ | |
29 | __asm__ __volatile__( \ | |
30 | LOCK " decl (%%eax) \n" \ | |
31 | " js "#fail_fn" \n" \ | |
32 | \ | |
33 | :"=a" (dummy) \ | |
34 | : "a" (count) \ | |
35 | : "memory", "ecx", "edx"); \ | |
36 | } while (0) | |
37 | ||
38 | ||
39 | /** | |
40 | * __mutex_fastpath_lock_retval - try to take the lock by moving the count | |
41 | * from 1 to a 0 value | |
42 | * @count: pointer of type atomic_t | |
43 | * @fail_fn: function to call if the original value was not 1 | |
44 | * | |
45 | * Change the count from 1 to a value lower than 1, and call <fail_fn> if it | |
46 | * wasn't 1 originally. This function returns 0 if the fastpath succeeds, | |
47 | * or anything the slow path function returns | |
48 | */ | |
49 | static inline int | |
50 | __mutex_fastpath_lock_retval(atomic_t *count, | |
51 | int fastcall (*fail_fn)(atomic_t *)) | |
52 | { | |
53 | if (unlikely(atomic_dec_return(count) < 0)) | |
54 | return fail_fn(count); | |
55 | else | |
56 | return 0; | |
57 | } | |
58 | ||
59 | /** | |
60 | * __mutex_fastpath_unlock - try to promote the mutex from 0 to 1 | |
61 | * @count: pointer of type atomic_t | |
62 | * @fail_fn: function to call if the original value was not 0 | |
63 | * | |
64 | * try to promote the mutex from 0 to 1. if it wasn't 0, call <fail_fn>. | |
65 | * In the failure case, this function is allowed to either set the value | |
66 | * to 1, or to set it to a value lower than 1. | |
67 | * | |
68 | * If the implementation sets it to a value of lower than 1, the | |
69 | * __mutex_slowpath_needs_to_unlock() macro needs to return 1, it needs | |
70 | * to return 0 otherwise. | |
71 | */ | |
72 | #define __mutex_fastpath_unlock(count, fail_fn) \ | |
73 | do { \ | |
74 | unsigned int dummy; \ | |
75 | \ | |
76 | typecheck(atomic_t *, count); \ | |
77 | typecheck_fn(fastcall void (*)(atomic_t *), fail_fn); \ | |
78 | \ | |
79 | __asm__ __volatile__( \ | |
80 | LOCK " incl (%%eax) \n" \ | |
81 | " jle "#fail_fn" \n" \ | |
82 | \ | |
83 | :"=a" (dummy) \ | |
84 | : "a" (count) \ | |
85 | : "memory", "ecx", "edx"); \ | |
86 | } while (0) | |
87 | ||
88 | #define __mutex_slowpath_needs_to_unlock() 1 | |
89 | ||
90 | /** | |
91 | * __mutex_fastpath_trylock - try to acquire the mutex, without waiting | |
92 | * | |
93 | * @count: pointer of type atomic_t | |
94 | * @fail_fn: fallback function | |
95 | * | |
96 | * Change the count from 1 to a value lower than 1, and return 0 (failure) | |
97 | * if it wasn't 1 originally, or return 1 (success) otherwise. This function | |
98 | * MUST leave the value lower than 1 even when the "1" assertion wasn't true. | |
99 | * Additionally, if the value was < 0 originally, this function must not leave | |
100 | * it to 0 on failure. | |
101 | */ | |
102 | static inline int | |
103 | __mutex_fastpath_trylock(atomic_t *count, int (*fail_fn)(atomic_t *)) | |
104 | { | |
105 | /* | |
106 | * We have two variants here. The cmpxchg based one is the best one | |
107 | * because it never induce a false contention state. It is included | |
108 | * here because architectures using the inc/dec algorithms over the | |
109 | * xchg ones are much more likely to support cmpxchg natively. | |
110 | * | |
111 | * If not we fall back to the spinlock based variant - that is | |
112 | * just as efficient (and simpler) as a 'destructive' probing of | |
113 | * the mutex state would be. | |
114 | */ | |
115 | #ifdef __HAVE_ARCH_CMPXCHG | |
116 | if (likely(atomic_cmpxchg(count, 1, 0)) == 1) | |
117 | return 1; | |
118 | return 0; | |
119 | #else | |
120 | return fail_fn(count); | |
121 | #endif | |
122 | } | |
123 | ||
124 | #endif |