From f091d5a426447cc427680bdd3adc7773aa2867df Mon Sep 17 00:00:00 2001
From: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Date: Fri, 8 Nov 2019 19:20:22 +0300
Subject: ARC: ARCv2: jump label: implement jump label patching

Implement jump label patching for ARC. Jump labels provide
an interface to generate dynamic branches using
self-modifying code.

This allows us to implement conditional branches where
changing branch direction is expensive but branch selection
is basically 'free'

This implementation uses 32-bit NOP and BRANCH instructions
which forced to be aligned by 4 to guarantee that they don't
cross L1 cache line boundary and can be update atomically.

Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
---
 arch/arc/include/asm/cache.h      |  2 ++
 arch/arc/include/asm/jump_label.h | 72 +++++++++++++++++++++++++++++++++++++++
 2 files changed, 74 insertions(+)
 create mode 100644 arch/arc/include/asm/jump_label.h

(limited to 'arch/arc/include')

diff --git a/arch/arc/include/asm/cache.h b/arch/arc/include/asm/cache.h
index 918804c7c1a4..d8ece4292388 100644
--- a/arch/arc/include/asm/cache.h
+++ b/arch/arc/include/asm/cache.h
@@ -25,6 +25,8 @@
 
 #ifndef __ASSEMBLY__
 
+#include <linux/build_bug.h>
+
 /* Uncached access macros */
 #define arc_read_uncached_32(ptr)	\
 ({					\
diff --git a/arch/arc/include/asm/jump_label.h b/arch/arc/include/asm/jump_label.h
new file mode 100644
index 000000000000..9d9618079739
--- /dev/null
+++ b/arch/arc/include/asm/jump_label.h
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_ARC_JUMP_LABEL_H
+#define _ASM_ARC_JUMP_LABEL_H
+
+#ifndef __ASSEMBLY__
+
+#include <linux/stringify.h>
+#include <linux/types.h>
+
+#define JUMP_LABEL_NOP_SIZE 4
+
+/*
+ * NOTE about '.balign 4':
+ *
+ * To make atomic update of patched instruction available we need to guarantee
+ * that this instruction doesn't cross L1 cache line boundary.
+ *
+ * As of today we simply align instruction which can be patched by 4 byte using
+ * ".balign 4" directive. In that case patched instruction is aligned with one
+ * 16-bit NOP_S if this is required.
+ * However 'align by 4' directive is much stricter than it actually required.
+ * It's enough that our 32-bit instruction don't cross L1 cache line boundary /
+ * L1 I$ fetch block boundary which can be achieved by using
+ * ".bundle_align_mode" assembler directive. That will save us from adding
+ * useless NOP_S padding in most of the cases.
+ *
+ * TODO: switch to ".bundle_align_mode" directive using whin it will be
+ * supported by ARC toolchain.
+ */
+
+static __always_inline bool arch_static_branch(struct static_key *key,
+					       bool branch)
+{
+	asm_volatile_goto(".balign "__stringify(JUMP_LABEL_NOP_SIZE)"	\n"
+		 "1:							\n"
+		 "nop							\n"
+		 ".pushsection __jump_table, \"aw\"			\n"
+		 ".word 1b, %l[l_yes], %c0				\n"
+		 ".popsection						\n"
+		 : : "i" (&((char *)key)[branch]) : : l_yes);
+
+	return false;
+l_yes:
+	return true;
+}
+
+static __always_inline bool arch_static_branch_jump(struct static_key *key,
+						    bool branch)
+{
+	asm_volatile_goto(".balign "__stringify(JUMP_LABEL_NOP_SIZE)"	\n"
+		 "1:							\n"
+		 "b %l[l_yes]						\n"
+		 ".pushsection __jump_table, \"aw\"			\n"
+		 ".word 1b, %l[l_yes], %c0				\n"
+		 ".popsection						\n"
+		 : : "i" (&((char *)key)[branch]) : : l_yes);
+
+	return false;
+l_yes:
+	return true;
+}
+
+typedef u32 jump_label_t;
+
+struct jump_entry {
+	jump_label_t code;
+	jump_label_t target;
+	jump_label_t key;
+};
+
+#endif  /* __ASSEMBLY__ */
+#endif
-- 
cgit v1.2.3