RISC-V: Make __NR_riscv_flush_icache visible to userspace

We were hoping to avoid making this visible to userspace, but it looks
like we're going to have to because QEMU's user-mode emulation doesn't
want to emulate a vDSO.  Having vDSO-only system calls was a bit
unothodox anyway, so I think in this case it's OK to just make the
actual system call number public.

This patch simply moves the definition of __NR_riscv_flush_icache
availiable to userspace, which results in the deletion of the now empty
vdso-syscalls.h.

Changes since v1:

* I've moved the definition into uapi/asm/syscalls.h rathen than
  uapi/asm/unistd.h.  This allows me to keep asm/unistd.h, so we can
  keep the syscall table macros sane.
* As a side effect of the above, this no longer disables all system
  calls on RISC-V.  Whoops!

Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

arch/riscv/include/uapi/asm/syscalls.h

@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2017 SiFive
+ */
+
+#ifndef _ASM__UAPI__SYSCALLS_H
+#define _ASM__UAPI__SYSCALLS_H
+
+/*
+ * Allows the instruction cache to be flushed from userspace.  Despite RISC-V
+ * having a direct 'fence.i' instruction available to userspace (which we
+ * can't trap!), that's not actually viable when running on Linux because the
+ * kernel might schedule a process on another hart.  There is no way for
+ * userspace to handle this without invoking the kernel (as it doesn't know the
+ * thread->hart mappings), so we've defined a RISC-V specific system call to
+ * flush the instruction cache.
+ *
+ * __NR_riscv_flush_icache is defined to flush the instruction cache over an
+ * address range, with the flush applying to either all threads or just the
+ * caller.  We don't currently do anything with the address range, that's just
+ * in there for forwards compatibility.
+ */
+#define __NR_riscv_flush_icache (__NR_arch_specific_syscall + 15)
+__SYSCALL(__NR_riscv_flush_icache, sys_riscv_flush_icache)
+
+#endif