summaryrefslogtreecommitdiffstats
path: root/arch/arc/include/asm/cmpxchg.h
blob: 3ea4112c8302a8b71b3dc7e473fd7e78d3992b1d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
/*
 * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */

#ifndef __ASM_ARC_CMPXCHG_H
#define __ASM_ARC_CMPXCHG_H

#include <linux/types.h>

#include <asm/barrier.h>
#include <asm/smp.h>

#ifdef CONFIG_ARC_HAS_LLSC

static inline unsigned long
__cmpxchg(volatile void *ptr, unsigned long expected, unsigned long new)
{
	unsigned long prev;

	/*
	 * Explicit full memory barrier needed before/after as
	 * LLOCK/SCOND thmeselves don't provide any such semantics
	 */
	smp_mb();

	__asm__ __volatile__(
	"1:	llock   %0, [%1]	\n"
	"	brne    %0, %2, 2f	\n"
	"	scond   %3, [%1]	\n"
	"	bnz     1b		\n"
	"2:				\n"
	: "=&r"(prev)	/* Early clobber, to prevent reg reuse */
	: "r"(ptr),	/* Not "m": llock only supports reg direct addr mode */
	  "ir"(expected),
	  "r"(new)	/* can't be "ir". scond can't take LIMM for "b" */
	: "cc", "memory"); /* so that gcc knows memory is being written here */

	smp_mb();

	return prev;
}

#elif !defined(CONFIG_ARC_PLAT_EZNPS)

static inline unsigned long
__cmpxchg(volatile void *ptr, unsigned long expected, unsigned long new)
{
	unsigned long flags;
	int prev;
	volatile unsigned long *p = ptr;

	/*
	 * spin lock/unlock provide the needed smp_mb() before/after
	 */
	atomic_ops_lock(flags);
	prev = *p;
	if (prev == expected)
		*p = new;
	atomic_ops_unlock(flags);
	return prev;
}

#else /* CONFIG_ARC_PLAT_EZNPS */

static inline unsigned long
__cmpxchg(volatile void *ptr, unsigned long expected, unsigned long new)
{
	/*
	 * Explicit full memory barrier needed before/after
	 */
	smp_mb();

	write_aux_reg(CTOP_AUX_GPA1, expected);

	__asm__ __volatile__(
	"	mov r2, %0\n"
	"	mov r3, %1\n"
	"	.word %2\n"
	"	mov %0, r2"
	: "+r"(new)
	: "r"(ptr), "i"(CTOP_INST_EXC_DI_R2_R2_R3)
	: "r2", "r3", "memory");

	smp_mb();

	return new;
}

#endif /* CONFIG_ARC_HAS_LLSC */

#define cmpxchg(ptr, o, n) ({				\
	(typeof(*(ptr)))__cmpxchg((ptr),		\
				  (unsigned long)(o),	\
				  (unsigned long)(n));	\
})

/*
 * atomic_cmpxchg is same as cmpxchg
 *   LLSC: only different in data-type, semantics are exactly same
 *  !LLSC: cmpxchg() has to use an external lock atomic_ops_lock to guarantee
 *         semantics, and this lock also happens to be used by atomic_*()
 */
#define atomic_cmpxchg(v, o, n) ((int)cmpxchg(&((v)->counter), (o), (n)))


#ifndef CONFIG_ARC_PLAT_EZNPS

/*
 * xchg (reg with memory) based on "Native atomic" EX insn
 */
static inline unsigned long __xchg(unsigned long val, volatile void *ptr,
				   int size)
{
	extern unsigned long __xchg_bad_pointer(void);

	switch (size) {
	case 4:
		smp_mb();

		__asm__ __volatile__(
		"	ex  %0, [%1]	\n"
		: "+r"(val)
		: "r"(ptr)
		: "memory");

		smp_mb();

		return val;
	}
	return __xchg_bad_pointer();
}

#define _xchg(ptr, with) ((typeof(*(ptr)))__xchg((unsigned long)(with), (ptr), \
						 sizeof(*(ptr))))

/*
 * xchg() maps directly to ARC EX instruction which guarantees atomicity.
 * However in !LLSC config, it also needs to be use @atomic_ops_lock spinlock
 * due to a subtle reason:
 *  - For !LLSC, cmpxchg() needs to use that lock (see above) and there is lot
 *    of  kernel code which calls xchg()/cmpxchg() on same data (see llist.h)
 *    Hence xchg() needs to follow same locking rules.
 *
 * Technically the lock is also needed for UP (boils down to irq save/restore)
 * but we can cheat a bit since cmpxchg() atomic_ops_lock() would cause irqs to
 * be disabled thus can't possibly be interrpted/preempted/clobbered by xchg()
 * Other way around, xchg is one instruction anyways, so can't be interrupted
 * as such
 */

#if !defined(CONFIG_ARC_HAS_LLSC) && defined(CONFIG_SMP)

#define xchg(ptr, with)			\
({					\
	unsigned long flags;		\
	typeof(*(ptr)) old_val;		\
					\
	atomic_ops_lock(flags);		\
	old_val = _xchg(ptr, with);	\
	atomic_ops_unlock(flags);	\
	old_val;			\
})

#else

#define xchg(ptr, with)  _xchg(ptr, with)

#endif

#else /* CONFIG_ARC_PLAT_EZNPS */

static inline unsigned long __xchg(unsigned long val, volatile void *ptr,
				   int size)
{
	extern unsigned long __xchg_bad_pointer(void);

	switch (size) {
	case 4:
		/*
		 * Explicit full memory barrier needed before/after
		 */
		smp_mb();

		__asm__ __volatile__(
		"	mov r2, %0\n"
		"	mov r3, %1\n"
		"	.word %2\n"
		"	mov %0, r2\n"
		: "+r"(val)
		: "r"(ptr), "i"(CTOP_INST_XEX_DI_R2_R2_R3)
		: "r2", "r3", "memory");

		smp_mb();

		return val;
	}
	return __xchg_bad_pointer();
}

#define xchg(ptr, with) ({				\
	(typeof(*(ptr)))__xchg((unsigned long)(with),	\
			       (ptr),			\
			       sizeof(*(ptr)));		\
})

#endif /* CONFIG_ARC_PLAT_EZNPS */

/*
 * "atomic" variant of xchg()
 * REQ: It needs to follow the same serialization rules as other atomic_xxx()
 * Since xchg() doesn't always do that, it would seem that following defintion
 * is incorrect. But here's the rationale:
 *   SMP : Even xchg() takes the atomic_ops_lock, so OK.
 *   LLSC: atomic_ops_lock are not relevant at all (even if SMP, since LLSC
 *         is natively "SMP safe", no serialization required).
 *   UP  : other atomics disable IRQ, so no way a difft ctxt atomic_xchg()
 *         could clobber them. atomic_xchg() itself would be 1 insn, so it
 *         can't be clobbered by others. Thus no serialization required when
 *         atomic_xchg is involved.
 */
#define atomic_xchg(v, new) (xchg(&((v)->counter), new))

#endif