s390/ftrace: provide separate ftrace_caller/ftrace_regs_caller implementations

ftrace_regs_caller is an alias to ftrace_caller - making ftrace_caller
quite heavyweight. Split the function and provide an ftrace_caller
implementation which comes with fewer instructions. Especially getting
rid of 'stosm' on each function entry should help here, e.g. to
have less performance impact on live patched functions.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

arch/s390/include/asm/ftrace.h

@@ -17,7 +17,6 @@
 
 void ftrace_caller(void);
 
-extern char ftrace_graph_caller_end;
 extern void *ftrace_func;
 
 struct dyn_arch_ftrace { };

arch/s390/kernel/mcount.S

@@ -33,13 +33,15 @@ ENDPROC(ftrace_stub)
 #define TRACED_FUNC_FRAME_SIZE	STACK_FRAME_OVERHEAD
 #endif
 
-ENTRY(ftrace_caller)
-	.globl	ftrace_regs_caller
-	.set	ftrace_regs_caller,ftrace_caller
+	.macro	ftrace_regs_entry, allregs=0
 	stg	%r14,(__SF_GPRS+8*8)(%r15)	# save traced function caller
+
+	.if \allregs == 1
 	lghi	%r14,0				# save condition code
 	ipm	%r14				# don't put any instructions
 	sllg	%r14,%r14,16			# clobbering CC before this point
+	.endif
+
 	lgr	%r1,%r15
 	# allocate stack frame for ftrace_caller to contain traced function
 	aghi	%r15,-TRACED_FUNC_FRAME_SIZE
@@ -49,13 +51,30 @@ ENTRY(ftrace_caller)
 	# allocate pt_regs and stack frame for ftrace_trace_function
 	aghi	%r15,-STACK_FRAME_SIZE
 	stg	%r1,(STACK_PTREGS_GPRS+15*8)(%r15)
+
+	.if \allregs == 1
 	stg	%r14,(STACK_PTREGS_PSW)(%r15)
-	lg	%r14,(__SF_GPRS+8*8)(%r1)	# restore original return address
 	stosm	(STACK_PTREGS_PSW)(%r15),0
+	.endif
+
+	lg	%r14,(__SF_GPRS+8*8)(%r1)	# restore original return address
 	aghi	%r1,-TRACED_FUNC_FRAME_SIZE
 	stg	%r1,__SF_BACKCHAIN(%r15)
 	stg	%r0,(STACK_PTREGS_PSW+8)(%r15)
 	stmg	%r2,%r14,(STACK_PTREGS_GPRS+2*8)(%r15)
+	.endm
+
+SYM_CODE_START(ftrace_regs_caller)
+	ftrace_regs_entry	1
+	j	ftrace_common
+SYM_CODE_END(ftrace_regs_caller)
+
+SYM_CODE_START(ftrace_caller)
+	ftrace_regs_entry	0
+	j	ftrace_common
+SYM_CODE_END(ftrace_caller)
+
+SYM_CODE_START(ftrace_common)
 #ifdef CONFIG_HAVE_MARCH_Z196_FEATURES
 	aghik	%r2,%r0,-MCOUNT_INSN_SIZE
 	lgrl	%r4,function_trace_op
@@ -74,24 +93,22 @@ ENTRY(ftrace_caller)
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 # The j instruction gets runtime patched to a nop instruction.
 # See ftrace_enable_ftrace_graph_caller.
-	.globl ftrace_graph_caller
-ftrace_graph_caller:
-	j	ftrace_graph_caller_end
+SYM_INNER_LABEL(ftrace_graph_caller, SYM_L_GLOBAL)
+	j	.Lftrace_graph_caller_end
 	lmg	%r2,%r3,(STACK_PTREGS_GPRS+14*8)(%r15)
 	lg	%r4,(STACK_PTREGS_PSW+8)(%r15)
 	brasl	%r14,prepare_ftrace_return
 	stg	%r2,(STACK_PTREGS_GPRS+14*8)(%r15)
-ftrace_graph_caller_end:
-	.globl	ftrace_graph_caller_end
+.Lftrace_graph_caller_end:
 #endif
 	lg	%r1,(STACK_PTREGS_PSW+8)(%r15)
 	lmg	%r2,%r15,(STACK_PTREGS_GPRS+2*8)(%r15)
 	BR_EX	%r1
-ENDPROC(ftrace_caller)
+SYM_CODE_END(ftrace_common)
 
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 
-ENTRY(return_to_handler)
+SYM_FUNC_START(return_to_handler)
 	stmg	%r2,%r5,32(%r15)
 	lgr	%r1,%r15
 	aghi	%r15,-STACK_FRAME_OVERHEAD
@@ -101,6 +118,6 @@ ENTRY(return_to_handler)
 	lgr	%r14,%r2
 	lmg	%r2,%r5,32(%r15)
 	BR_EX	%r14
-ENDPROC(return_to_handler)
+SYM_FUNC_END(return_to_handler)
 
 #endif