riscv: make use of the barrier functions from Linux

Replace the barrier functions in arch/riscv/include/asm/io.h with those
defined in barrier.h, which is imported from Linux. This version is
modified to remove the include statement of asm-generic/barrier.h, which
is not available in U-Boot or required.

Signed-off-by: Lukas Auer <lukas.auer@aisec.fraunhofer.de>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Rick Chen <rick@andestech.com>

arch/riscv/include/asm/barrier.h

@@ -0,0 +1,67 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2012 ARM Ltd.
+ * Copyright (C) 2013 Regents of the University of California
+ * Copyright (C) 2017 SiFive
+ *
+ * Taken from Linux arch/riscv/include/asm/barrier.h, which is based on
+ * arch/arm/include/asm/barrier.h
+ */
+
+#ifndef _ASM_RISCV_BARRIER_H
+#define _ASM_RISCV_BARRIER_H
+
+#ifndef __ASSEMBLY__
+
+#define nop()		__asm__ __volatile__ ("nop")
+
+#define RISCV_FENCE(p, s) \
+	__asm__ __volatile__ ("fence " #p "," #s : : : "memory")
+
+/* These barriers need to enforce ordering on both devices or memory. */
+#define mb()		RISCV_FENCE(iorw,iorw)
+#define rmb()		RISCV_FENCE(ir,ir)
+#define wmb()		RISCV_FENCE(ow,ow)
+
+/* These barriers do not need to enforce ordering on devices, just memory. */
+#define __smp_mb()	RISCV_FENCE(rw,rw)
+#define __smp_rmb()	RISCV_FENCE(r,r)
+#define __smp_wmb()	RISCV_FENCE(w,w)
+
+#define __smp_store_release(p, v)					\
+do {									\
+	compiletime_assert_atomic_type(*p);				\
+	RISCV_FENCE(rw,w);						\
+	WRITE_ONCE(*p, v);						\
+} while (0)
+
+#define __smp_load_acquire(p)						\
+({									\
+	typeof(*p) ___p1 = READ_ONCE(*p);				\
+	compiletime_assert_atomic_type(*p);				\
+	RISCV_FENCE(r,rw);						\
+	___p1;								\
+})
+
+/*
+ * This is a very specific barrier: it's currently only used in two places in
+ * the kernel, both in the scheduler.  See include/linux/spinlock.h for the two
+ * orderings it guarantees, but the "critical section is RCsc" guarantee
+ * mandates a barrier on RISC-V.  The sequence looks like:
+ *
+ *    lr.aq lock
+ *    sc    lock <= LOCKED
+ *    smp_mb__after_spinlock()
+ *    // critical section
+ *    lr    lock
+ *    sc.rl lock <= UNLOCKED
+ *
+ * The AQ/RL pair provides a RCpc critical section, but there's not really any
+ * way we can take advantage of that here because the ordering is only enforced
+ * on that one lock.  Thus, we're just doing a full fence.
+ */
+#define smp_mb__after_spinlock()	RISCV_FENCE(rw,rw)
+
+#endif /* __ASSEMBLY__ */
+
+#endif /* _ASM_RISCV_BARRIER_H */

arch/riscv/include/asm/io.h

@@ -10,6 +10,7 @@
 #ifdef __KERNEL__
 
 #include <linux/types.h>
+#include <asm/barrier.h>
 #include <asm/byteorder.h>
 
 static inline void sync(void)
@@ -91,13 +92,9 @@ static inline phys_addr_t virt_to_phys(void *vaddr)
 #define __raw_readl(a)			__arch_getl(a)
 #define __raw_readq(a)			__arch_getq(a)
 
-/*
+#define dmb()		mb()
- * TODO: The kernel offers some more advanced versions of barriers, it might
+#define __iormb()	rmb()
- * have some advantages to use them instead of the simple one here.
+#define __iowmb()	wmb()
- */
-#define dmb()		__asm__ __volatile__ ("" : : : "memory")
-#define __iormb()	dmb()
-#define __iowmb()	dmb()
 
 static inline void writeb(u8 val, volatile void __iomem *addr)
 {