diff options
author | Matthew Wilcox <matthew@wil.cx> | 2008-03-07 21:55:58 -0500 |
---|---|---|
committer | Matthew Wilcox <willy@linux.intel.com> | 2008-04-17 10:42:34 -0400 |
commit | 64ac24e738823161693bf791f87adc802cf529ff (patch) | |
tree | 19c0b0cf314d4394ca580c05b86cdf874ce0a167 /arch/arm | |
parent | e48b3deee475134585eed03e7afebe4bf9e0dba9 (diff) |
Generic semaphore implementation
Semaphores are no longer performance-critical, so a generic C
implementation is better for maintainability, debuggability and
extensibility. Thanks to Peter Zijlstra for fixing the lockdep
warning. Thanks to Harvey Harrison for pointing out that the
unlikely() was unnecessary.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'arch/arm')
-rw-r--r-- | arch/arm/kernel/Makefile | 2 | ||||
-rw-r--r-- | arch/arm/kernel/semaphore.c | 221 |
2 files changed, 1 insertions, 222 deletions
diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile index 00d44c6fbfe9..6235f72a14f0 100644 --- a/arch/arm/kernel/Makefile +++ b/arch/arm/kernel/Makefile | |||
@@ -7,7 +7,7 @@ AFLAGS_head.o := -DTEXT_OFFSET=$(TEXT_OFFSET) | |||
7 | # Object file lists. | 7 | # Object file lists. |
8 | 8 | ||
9 | obj-y := compat.o entry-armv.o entry-common.o irq.o \ | 9 | obj-y := compat.o entry-armv.o entry-common.o irq.o \ |
10 | process.o ptrace.o semaphore.o setup.o signal.o \ | 10 | process.o ptrace.o setup.o signal.o \ |
11 | sys_arm.o stacktrace.o time.o traps.o | 11 | sys_arm.o stacktrace.o time.o traps.o |
12 | 12 | ||
13 | obj-$(CONFIG_ISA_DMA_API) += dma.o | 13 | obj-$(CONFIG_ISA_DMA_API) += dma.o |
diff --git a/arch/arm/kernel/semaphore.c b/arch/arm/kernel/semaphore.c deleted file mode 100644 index 981fe5c6ccbe..000000000000 --- a/arch/arm/kernel/semaphore.c +++ /dev/null | |||
@@ -1,221 +0,0 @@ | |||
1 | /* | ||
2 | * ARM semaphore implementation, taken from | ||
3 | * | ||
4 | * i386 semaphore implementation. | ||
5 | * | ||
6 | * (C) Copyright 1999 Linus Torvalds | ||
7 | * | ||
8 | * Modified for ARM by Russell King | ||
9 | * | ||
10 | * This program is free software; you can redistribute it and/or modify | ||
11 | * it under the terms of the GNU General Public License version 2 as | ||
12 | * published by the Free Software Foundation. | ||
13 | */ | ||
14 | #include <linux/module.h> | ||
15 | #include <linux/sched.h> | ||
16 | #include <linux/errno.h> | ||
17 | #include <linux/init.h> | ||
18 | |||
19 | #include <asm/semaphore.h> | ||
20 | |||
21 | /* | ||
22 | * Semaphores are implemented using a two-way counter: | ||
23 | * The "count" variable is decremented for each process | ||
24 | * that tries to acquire the semaphore, while the "sleeping" | ||
25 | * variable is a count of such acquires. | ||
26 | * | ||
27 | * Notably, the inline "up()" and "down()" functions can | ||
28 | * efficiently test if they need to do any extra work (up | ||
29 | * needs to do something only if count was negative before | ||
30 | * the increment operation. | ||
31 | * | ||
32 | * "sleeping" and the contention routine ordering is | ||
33 | * protected by the semaphore spinlock. | ||
34 | * | ||
35 | * Note that these functions are only called when there is | ||
36 | * contention on the lock, and as such all this is the | ||
37 | * "non-critical" part of the whole semaphore business. The | ||
38 | * critical part is the inline stuff in <asm/semaphore.h> | ||
39 | * where we want to avoid any extra jumps and calls. | ||
40 | */ | ||
41 | |||
42 | /* | ||
43 | * Logic: | ||
44 | * - only on a boundary condition do we need to care. When we go | ||
45 | * from a negative count to a non-negative, we wake people up. | ||
46 | * - when we go from a non-negative count to a negative do we | ||
47 | * (a) synchronize with the "sleeper" count and (b) make sure | ||
48 | * that we're on the wakeup list before we synchronize so that | ||
49 | * we cannot lose wakeup events. | ||
50 | */ | ||
51 | |||
52 | void __up(struct semaphore *sem) | ||
53 | { | ||
54 | wake_up(&sem->wait); | ||
55 | } | ||
56 | |||
57 | static DEFINE_SPINLOCK(semaphore_lock); | ||
58 | |||
59 | void __sched __down(struct semaphore * sem) | ||
60 | { | ||
61 | struct task_struct *tsk = current; | ||
62 | DECLARE_WAITQUEUE(wait, tsk); | ||
63 | tsk->state = TASK_UNINTERRUPTIBLE; | ||
64 | add_wait_queue_exclusive(&sem->wait, &wait); | ||
65 | |||
66 | spin_lock_irq(&semaphore_lock); | ||
67 | sem->sleepers++; | ||
68 | for (;;) { | ||
69 | int sleepers = sem->sleepers; | ||
70 | |||
71 | /* | ||
72 | * Add "everybody else" into it. They aren't | ||
73 | * playing, because we own the spinlock. | ||
74 | */ | ||
75 | if (!atomic_add_negative(sleepers - 1, &sem->count)) { | ||
76 | sem->sleepers = 0; | ||
77 | break; | ||
78 | } | ||
79 | sem->sleepers = 1; /* us - see -1 above */ | ||
80 | spin_unlock_irq(&semaphore_lock); | ||
81 | |||
82 | schedule(); | ||
83 | tsk->state = TASK_UNINTERRUPTIBLE; | ||
84 | spin_lock_irq(&semaphore_lock); | ||
85 | } | ||
86 | spin_unlock_irq(&semaphore_lock); | ||
87 | remove_wait_queue(&sem->wait, &wait); | ||
88 | tsk->state = TASK_RUNNING; | ||
89 | wake_up(&sem->wait); | ||
90 | } | ||
91 | |||
92 | int __sched __down_interruptible(struct semaphore * sem) | ||
93 | { | ||
94 | int retval = 0; | ||
95 | struct task_struct *tsk = current; | ||
96 | DECLARE_WAITQUEUE(wait, tsk); | ||
97 | tsk->state = TASK_INTERRUPTIBLE; | ||
98 | add_wait_queue_exclusive(&sem->wait, &wait); | ||
99 | |||
100 | spin_lock_irq(&semaphore_lock); | ||
101 | sem->sleepers ++; | ||
102 | for (;;) { | ||
103 | int sleepers = sem->sleepers; | ||
104 | |||
105 | /* | ||
106 | * With signals pending, this turns into | ||
107 | * the trylock failure case - we won't be | ||
108 | * sleeping, and we* can't get the lock as | ||
109 | * it has contention. Just correct the count | ||
110 | * and exit. | ||
111 | */ | ||
112 | if (signal_pending(current)) { | ||
113 | retval = -EINTR; | ||
114 | sem->sleepers = 0; | ||
115 | atomic_add(sleepers, &sem->count); | ||
116 | break; | ||
117 | } | ||
118 | |||
119 | /* | ||
120 | * Add "everybody else" into it. They aren't | ||
121 | * playing, because we own the spinlock. The | ||
122 | * "-1" is because we're still hoping to get | ||
123 | * the lock. | ||
124 | */ | ||
125 | if (!atomic_add_negative(sleepers - 1, &sem->count)) { | ||
126 | sem->sleepers = 0; | ||
127 | break; | ||
128 | } | ||
129 | sem->sleepers = 1; /* us - see -1 above */ | ||
130 | spin_unlock_irq(&semaphore_lock); | ||
131 | |||
132 | schedule(); | ||
133 | tsk->state = TASK_INTERRUPTIBLE; | ||
134 | spin_lock_irq(&semaphore_lock); | ||
135 | } | ||
136 | spin_unlock_irq(&semaphore_lock); | ||
137 | tsk->state = TASK_RUNNING; | ||
138 | remove_wait_queue(&sem->wait, &wait); | ||
139 | wake_up(&sem->wait); | ||
140 | return retval; | ||
141 | } | ||
142 | |||
143 | /* | ||
144 | * Trylock failed - make sure we correct for | ||
145 | * having decremented the count. | ||
146 | * | ||
147 | * We could have done the trylock with a | ||
148 | * single "cmpxchg" without failure cases, | ||
149 | * but then it wouldn't work on a 386. | ||
150 | */ | ||
151 | int __down_trylock(struct semaphore * sem) | ||
152 | { | ||
153 | int sleepers; | ||
154 | unsigned long flags; | ||
155 | |||
156 | spin_lock_irqsave(&semaphore_lock, flags); | ||
157 | sleepers = sem->sleepers + 1; | ||
158 | sem->sleepers = 0; | ||
159 | |||
160 | /* | ||
161 | * Add "everybody else" and us into it. They aren't | ||
162 | * playing, because we own the spinlock. | ||
163 | */ | ||
164 | if (!atomic_add_negative(sleepers, &sem->count)) | ||
165 | wake_up(&sem->wait); | ||
166 | |||
167 | spin_unlock_irqrestore(&semaphore_lock, flags); | ||
168 | return 1; | ||
169 | } | ||
170 | |||
171 | /* | ||
172 | * The semaphore operations have a special calling sequence that | ||
173 | * allow us to do a simpler in-line version of them. These routines | ||
174 | * need to convert that sequence back into the C sequence when | ||
175 | * there is contention on the semaphore. | ||
176 | * | ||
177 | * ip contains the semaphore pointer on entry. Save the C-clobbered | ||
178 | * registers (r0 to r3 and lr), but not ip, as we use it as a return | ||
179 | * value in some cases.. | ||
180 | * To remain AAPCS compliant (64-bit stack align) we save r4 as well. | ||
181 | */ | ||
182 | asm(" .section .sched.text,\"ax\",%progbits \n\ | ||
183 | .align 5 \n\ | ||
184 | .globl __down_failed \n\ | ||
185 | __down_failed: \n\ | ||
186 | stmfd sp!, {r0 - r4, lr} \n\ | ||
187 | mov r0, ip \n\ | ||
188 | bl __down \n\ | ||
189 | ldmfd sp!, {r0 - r4, pc} \n\ | ||
190 | \n\ | ||
191 | .align 5 \n\ | ||
192 | .globl __down_interruptible_failed \n\ | ||
193 | __down_interruptible_failed: \n\ | ||
194 | stmfd sp!, {r0 - r4, lr} \n\ | ||
195 | mov r0, ip \n\ | ||
196 | bl __down_interruptible \n\ | ||
197 | mov ip, r0 \n\ | ||
198 | ldmfd sp!, {r0 - r4, pc} \n\ | ||
199 | \n\ | ||
200 | .align 5 \n\ | ||
201 | .globl __down_trylock_failed \n\ | ||
202 | __down_trylock_failed: \n\ | ||
203 | stmfd sp!, {r0 - r4, lr} \n\ | ||
204 | mov r0, ip \n\ | ||
205 | bl __down_trylock \n\ | ||
206 | mov ip, r0 \n\ | ||
207 | ldmfd sp!, {r0 - r4, pc} \n\ | ||
208 | \n\ | ||
209 | .align 5 \n\ | ||
210 | .globl __up_wakeup \n\ | ||
211 | __up_wakeup: \n\ | ||
212 | stmfd sp!, {r0 - r4, lr} \n\ | ||
213 | mov r0, ip \n\ | ||
214 | bl __up \n\ | ||
215 | ldmfd sp!, {r0 - r4, pc} \n\ | ||
216 | "); | ||
217 | |||
218 | EXPORT_SYMBOL(__down_failed); | ||
219 | EXPORT_SYMBOL(__down_interruptible_failed); | ||
220 | EXPORT_SYMBOL(__down_trylock_failed); | ||
221 | EXPORT_SYMBOL(__up_wakeup); | ||