xref: /linux/arch/riscv/include/asm/barrier.h (revision a4a755c422242c27cb0f7900ac00cf33ac17b1ce)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Based on arch/arm/include/asm/barrier.h
4  *
5  * Copyright (C) 2012 ARM Ltd.
6  * Copyright (C) 2013 Regents of the University of California
7  * Copyright (C) 2017 SiFive
8  */
9 
10 #ifndef _ASM_RISCV_BARRIER_H
11 #define _ASM_RISCV_BARRIER_H
12 
13 #ifndef __ASSEMBLY__
14 #include <asm/fence.h>
15 
16 #define nop()		__asm__ __volatile__ ("nop")
17 #define __nops(n)	".rept	" #n "\nnop\n.endr\n"
18 #define nops(n)		__asm__ __volatile__ (__nops(n))
19 
20 
21 /* These barriers need to enforce ordering on both devices or memory. */
22 #define __mb()		RISCV_FENCE(iorw, iorw)
23 #define __rmb()		RISCV_FENCE(ir, ir)
24 #define __wmb()		RISCV_FENCE(ow, ow)
25 
26 /* These barriers do not need to enforce ordering on devices, just memory. */
27 #define __smp_mb()	RISCV_FENCE(rw, rw)
28 #define __smp_rmb()	RISCV_FENCE(r, r)
29 #define __smp_wmb()	RISCV_FENCE(w, w)
30 
31 #define __smp_store_release(p, v)					\
32 do {									\
33 	compiletime_assert_atomic_type(*p);				\
34 	RISCV_FENCE(rw, w);						\
35 	WRITE_ONCE(*p, v);						\
36 } while (0)
37 
38 #define __smp_load_acquire(p)						\
39 ({									\
40 	typeof(*p) ___p1 = READ_ONCE(*p);				\
41 	compiletime_assert_atomic_type(*p);				\
42 	RISCV_FENCE(r, rw);						\
43 	___p1;								\
44 })
45 
46 /*
47  * This is a very specific barrier: it's currently only used in two places in
48  * the kernel, both in the scheduler.  See include/linux/spinlock.h for the two
49  * orderings it guarantees, but the "critical section is RCsc" guarantee
50  * mandates a barrier on RISC-V.  The sequence looks like:
51  *
52  *    lr.aq lock
53  *    sc    lock <= LOCKED
54  *    smp_mb__after_spinlock()
55  *    // critical section
56  *    lr    lock
57  *    sc.rl lock <= UNLOCKED
58  *
59  * The AQ/RL pair provides a RCpc critical section, but there's not really any
60  * way we can take advantage of that here because the ordering is only enforced
61  * on that one lock.  Thus, we're just doing a full fence.
62  *
63  * Since we allow writeX to be called from preemptive regions we need at least
64  * an "o" in the predecessor set to ensure device writes are visible before the
65  * task is marked as available for scheduling on a new hart.  While I don't see
66  * any concrete reason we need a full IO fence, it seems safer to just upgrade
67  * this in order to avoid any IO crossing a scheduling boundary.  In both
68  * instances the scheduler pairs this with an mb(), so nothing is necessary on
69  * the new hart.
70  */
71 #define smp_mb__after_spinlock()	RISCV_FENCE(iorw, iorw)
72 
73 #include <asm-generic/barrier.h>
74 
75 #endif /* __ASSEMBLY__ */
76 
77 #endif /* _ASM_RISCV_BARRIER_H */
78