MIPS: jump_label: Use compact branches for >= r6

MIPSr6 introduced compact branches which have no delay slots. Make use
of them for jump labels in order to avoid the need for a nop to fill the
branch or jump delay slot, saving 4 bytes of code for each static branch.

Signed-off-by: Paul Burton <paul.burton@mips.com>
Cc: linux-mips@vger.kernel.org

arch/mips/include/asm/jump_label.h

@@ -11,6 +11,7 @@
 #ifndef __ASSEMBLY__
 
 #include <linux/types.h>
+#include <asm/isa-rev.h>
 
 #define JUMP_LABEL_NOP_SIZE 4
 
@@ -21,9 +22,14 @@
 #endif
 
 #ifdef CONFIG_CPU_MICROMIPS
-#define B_INSN "b32"
+# define B_INSN "b32"
+# define J_INSN "j32"
+#elif MIPS_ISA_REV >= 6
+# define B_INSN "bc"
+# define J_INSN "bc"
 #else
-#define B_INSN "b"
+# define B_INSN "b"
+# define J_INSN "j"
 #endif
 
 static __always_inline bool arch_static_branch(struct static_key *key, bool branch)
@@ -42,7 +48,7 @@ static __always_inline bool arch_static_branch(struct static_key *key, bool bran
 
 static __always_inline bool arch_static_branch_jump(struct static_key *key, bool branch)
 {
-	asm_volatile_goto("1:\tj %l[l_yes]\n\t"
+	asm_volatile_goto("1:\t" J_INSN " %l[l_yes]\n\t"
 		".pushsection __jump_table,  \"aw\"\n\t"
 		WORD_INSN " 1b, %l[l_yes], %0\n\t"
 		".popsection\n\t"

arch/mips/kernel/jump_label.c

@@ -40,18 +40,38 @@ void arch_jump_label_transform(struct jump_entry *e,
 {
 	union mips_instruction *insn_p;
 	union mips_instruction insn;
+	long offset;
 
 	insn_p = (union mips_instruction *)msk_isa16_mode(e->code);
 
-	/* Jump only works within an aligned region its delay slot is in. */
-	BUG_ON((e->target & ~J_RANGE_MASK) != ((e->code + 4) & ~J_RANGE_MASK));
-
 	/* Target must have the right alignment and ISA must be preserved. */
 	BUG_ON((e->target & J_ALIGN_MASK) != J_ISA_BIT);
 
 	if (type == JUMP_LABEL_JMP) {
-		insn.j_format.opcode = J_ISA_BIT ? mm_j32_op : j_op;
-		insn.j_format.target = e->target >> J_RANGE_SHIFT;
+		if (!IS_ENABLED(CONFIG_CPU_MICROMIPS) && MIPS_ISA_REV >= 6) {
+			offset = e->target - ((unsigned long)insn_p + 4);
+			offset >>= 2;
+
+			/*
+			 * The branch offset must fit in the instruction's 26
+			 * bit field.
+			 */
+			WARN_ON((offset >= BIT(25)) ||
+				(offset < -(long)BIT(25)));
+
+			insn.j_format.opcode = bc6_op;
+			insn.j_format.target = offset;
+		} else {
+			/*
+			 * Jump only works within an aligned region its delay
+			 * slot is in.
+			 */
+			WARN_ON((e->target & ~J_RANGE_MASK) !=
+				((e->code + 4) & ~J_RANGE_MASK));
+
+			insn.j_format.opcode = J_ISA_BIT ? mm_j32_op : j_op;
+			insn.j_format.target = e->target >> J_RANGE_SHIFT;
+		}
 	} else {
 		insn.word = 0; /* nop */
 	}