diff --git a/arch/tile/include/asm/ftrace.h b/arch/tile/include/asm/ftrace.h
index 13a9bb81a8ab4754ff93da4c8f17da8a134ce4ec..738d239b792f0aab7a974f75600bd7adad676b4f 100644
--- a/arch/tile/include/asm/ftrace.h
+++ b/arch/tile/include/asm/ftrace.h
@@ -23,6 +23,8 @@
 #ifndef __ASSEMBLY__
 extern void __mcount(void);
 
+#define ARCH_SUPPORTS_FTRACE_OPS 1
+
 #ifdef CONFIG_DYNAMIC_FTRACE
 static inline unsigned long ftrace_call_adjust(unsigned long addr)
 {
diff --git a/arch/tile/kernel/ftrace.c b/arch/tile/kernel/ftrace.c
index 8d52d83cc51617b8d8e055dba3672cdcbdd46f3a..0c0996175b1ed613cf6fe4d3e89cec85ea5c6a6f 100644
--- a/arch/tile/kernel/ftrace.c
+++ b/arch/tile/kernel/ftrace.c
@@ -74,7 +74,11 @@ static unsigned long ftrace_gen_branch(unsigned long pc, unsigned long addr,
 			create_JumpOff_X1(pcrel_by_instr);
 	}
 
-	if (addr == FTRACE_ADDR) {
+	/*
+	 * Also put { move r10, lr; jal ftrace_stub } in a bundle, which
+	 * is used to replace the instruction in address ftrace_call.
+	 */
+	if (addr == FTRACE_ADDR || addr == (unsigned long)ftrace_stub) {
 		/* opcode: or r10, lr, zero */
 		opcode_x0 =
 			create_Dest_X0(10) |
diff --git a/arch/tile/kernel/mcount_64.S b/arch/tile/kernel/mcount_64.S
index 3c2b8d5e1d1a02feb744da440a3bab224b125d6d..6c6702451962360cefa822f403c18628a31bb2e7 100644
--- a/arch/tile/kernel/mcount_64.S
+++ b/arch/tile/kernel/mcount_64.S
@@ -81,7 +81,12 @@ STD_ENTRY(ftrace_caller)
 
 	/* arg1: self return address */
 	/* arg2: parent's return address */
-	{ move	r0, lr; move	r1, r10 }
+	/* arg3: ftrace_ops */
+	/* arg4: regs (but make it NULL) */
+	{ move	r0, lr;  moveli        r2, hw2_last(function_trace_op) }
+	{ move	r1, r10; shl16insli    r2, r2, hw1(function_trace_op) }
+	{ movei	r3, 0;   shl16insli    r2, r2, hw0(function_trace_op) }
+	ld	r2,r2
 
 	.global	ftrace_call
 ftrace_call: