xref: /netbsd-src/sys/arch/x86/include/lock.h (revision 2b25048cdd284ab89064935296b4895e91e8a510)
1 /*	$NetBSD: lock.h,v 1.29 2022/02/12 17:17:54 riastradh Exp $	*/
2 
3 /*-
4  * Copyright (c) 2000, 2006 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe and Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Machine-dependent spin lock operations.
34  */
35 
36 #ifndef _X86_LOCK_H_
37 #define	_X86_LOCK_H_
38 
39 #include <sys/param.h>
40 
41 static __inline int
__SIMPLELOCK_LOCKED_P(const __cpu_simple_lock_t * __ptr)42 __SIMPLELOCK_LOCKED_P(const __cpu_simple_lock_t *__ptr)
43 {
44 	return *__ptr == __SIMPLELOCK_LOCKED;
45 }
46 
47 static __inline int
__SIMPLELOCK_UNLOCKED_P(const __cpu_simple_lock_t * __ptr)48 __SIMPLELOCK_UNLOCKED_P(const __cpu_simple_lock_t *__ptr)
49 {
50 	return *__ptr == __SIMPLELOCK_UNLOCKED;
51 }
52 
53 static __inline void
__cpu_simple_lock_set(__cpu_simple_lock_t * __ptr)54 __cpu_simple_lock_set(__cpu_simple_lock_t *__ptr)
55 {
56 
57 	*__ptr = __SIMPLELOCK_LOCKED;
58 }
59 
60 static __inline void
__cpu_simple_lock_clear(__cpu_simple_lock_t * __ptr)61 __cpu_simple_lock_clear(__cpu_simple_lock_t *__ptr)
62 {
63 
64 	*__ptr = __SIMPLELOCK_UNLOCKED;
65 }
66 
67 #ifdef _HARDKERNEL
68 # include <machine/cpufunc.h>
69 # define SPINLOCK_SPIN_HOOK	/* nothing */
70 # ifdef SPINLOCK_BACKOFF_HOOK
71 #  undef SPINLOCK_BACKOFF_HOOK
72 # endif
73 # define SPINLOCK_BACKOFF_HOOK	x86_pause()
74 # define SPINLOCK_INLINE
75 #else /* !_HARDKERNEL */
76 # define SPINLOCK_BODY
77 # define SPINLOCK_INLINE static __inline __unused
78 #endif /* _HARDKERNEL */
79 
80 SPINLOCK_INLINE void	__cpu_simple_lock_init(__cpu_simple_lock_t *);
81 SPINLOCK_INLINE void	__cpu_simple_lock(__cpu_simple_lock_t *);
82 SPINLOCK_INLINE int	__cpu_simple_lock_try(__cpu_simple_lock_t *);
83 SPINLOCK_INLINE void	__cpu_simple_unlock(__cpu_simple_lock_t *);
84 
85 #ifdef SPINLOCK_BODY
86 SPINLOCK_INLINE void
__cpu_simple_lock_init(__cpu_simple_lock_t * lockp)87 __cpu_simple_lock_init(__cpu_simple_lock_t *lockp)
88 {
89 
90 	*lockp = __SIMPLELOCK_UNLOCKED;
91 }
92 
93 SPINLOCK_INLINE int
__cpu_simple_lock_try(__cpu_simple_lock_t * lockp)94 __cpu_simple_lock_try(__cpu_simple_lock_t *lockp)
95 {
96 	uint8_t val;
97 
98 	val = __SIMPLELOCK_LOCKED;
99 	__asm volatile ("xchgb %0,(%2)" :
100 	    "=qQ" (val)
101 	    :"0" (val), "r" (lockp));
102 	__insn_barrier();
103 	return val == __SIMPLELOCK_UNLOCKED;
104 }
105 
106 SPINLOCK_INLINE void
__cpu_simple_lock(__cpu_simple_lock_t * lockp)107 __cpu_simple_lock(__cpu_simple_lock_t *lockp)
108 {
109 
110 	while (!__cpu_simple_lock_try(lockp))
111 		/* nothing */;
112 	__insn_barrier();
113 }
114 
115 /*
116  * Note on x86 memory ordering
117  *
118  * When releasing a lock we must ensure that no stores or loads from within
119  * the critical section are re-ordered by the CPU to occur outside of it:
120  * they must have completed and be visible to other processors once the lock
121  * has been released.
122  *
123  * NetBSD usually runs with the kernel mapped (via MTRR) in a WB (write
124  * back) memory region.  In that case, memory ordering on x86 platforms
125  * looks like this:
126  *
127  * i386		All loads/stores occur in instruction sequence.
128  *
129  * i486		All loads/stores occur in instruction sequence.  In
130  * Pentium	exceptional circumstances, loads can be re-ordered around
131  *		stores, but for the purposes of releasing a lock it does
132  *		not matter.  Stores may not be immediately visible to other
133  *		processors as they can be buffered.  However, since the
134  *		stores are buffered in order the lock release will always be
135  *		the last operation in the critical section that becomes
136  *		visible to other CPUs.
137  *
138  * Pentium Pro	The "Intel 64 and IA-32 Architectures Software Developer's
139  * onwards	Manual" volume 3A (order number 248966) says that (1) "Reads
140  *		can be carried out speculatively and in any order" and (2)
141  *		"Reads can pass buffered stores, but the processor is
142  *		self-consistent.".  This would be a problem for the below,
143  *		and would mandate a locked instruction cycle or load fence
144  *		before releasing the simple lock.
145  *
146  *		The "Intel Pentium 4 Processor Optimization" guide (order
147  *		number 253668-022US) says: "Loads can be moved before stores
148  *		that occurred earlier in the program if they are not
149  *		predicted to load from the same linear address.".  This is
150  *		not a problem since the only loads that can be re-ordered
151  *		take place once the lock has been released via a store.
152  *
153  *		The above two documents seem to contradict each other,
154  *		however with the exception of early steppings of the Pentium
155  *		Pro, the second document is closer to the truth: a store
156  *		will always act as a load fence for all loads that precede
157  *		the store in instruction order.
158  *
159  *		Again, note that stores can be buffered and will not always
160  *		become immediately visible to other CPUs: they are however
161  *		buffered in order.
162  *
163  * AMD64	Stores occur in order and are buffered.  Loads can be
164  *		reordered, however stores act as load fences, meaning that
165  *		loads can not be reordered around stores.
166  */
167 SPINLOCK_INLINE void
__cpu_simple_unlock(__cpu_simple_lock_t * lockp)168 __cpu_simple_unlock(__cpu_simple_lock_t *lockp)
169 {
170 
171 	__insn_barrier();
172 	*lockp = __SIMPLELOCK_UNLOCKED;
173 }
174 
175 #endif	/* SPINLOCK_BODY */
176 
177 #endif /* _X86_LOCK_H_ */
178