s390/ftrace: optimize mcount code

Reduce the number of executed instructions within the mcount block if
function tracing is enabled. We achieve that by using a non-standard
C function call ABI. Since the called function is also written in
assembler this is not a problem.
This also allows to replace the unconditional store at the beginning
of the mcount block with a larl instruction, which doesn't touch
memory.

In theory we could also patch the first instruction of the mcount block
to enable and disable function tracing. However this would break kprobes.
This could be fixed with implementing the "kprobes_on_ftrace" feature;
however keeping the odd jprobes working seems not to be possible without
a lot of code churn. Therefore keep the code easy and simply accept one
wasted 1-cycle "larl" instruction per function prologue.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/ftrace.h

@@ -19,7 +19,7 @@ static inline unsigned long ftrace_call_adjust(unsigned long addr)
 #endif /* __ASSEMBLY__ */
 
 #ifdef CONFIG_64BIT
-#define MCOUNT_INSN_SIZE  12
+#define MCOUNT_INSN_SIZE  18
 #else
 #define MCOUNT_INSN_SIZE  22
 #endif

arch/s390/kernel/ftrace.c

 /*
  * Dynamic function tracer architecture backend.
  *
- * Copyright IBM Corp. 2009
+ * Copyright IBM Corp. 2009,2014
  *
  *   Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>,
  *		Martin Schwidefsky <schwidefsky@de.ibm.com>
@@ -17,6 +17,7 @@
 #include <asm/asm-offsets.h>
 #include "entry.h"
 
+void mcount_replace_code(void);
 void ftrace_disable_code(void);
 void ftrace_enable_insn(void);
 
@@ -24,38 +25,50 @@ void ftrace_enable_insn(void);
 /*
  * The 64-bit mcount code looks like this:
  *	stg	%r14,8(%r15)		# offset 0
- * >	larl	%r1,<&counter>		# offset 6
- * >	brasl	%r14,_mcount		# offset 12
+ *	larl	%r1,<&counter>		# offset 6
+ *	brasl	%r14,_mcount		# offset 12
  *	lg	%r14,8(%r15)		# offset 18
- * Total length is 24 bytes. The middle two instructions of the mcount
- * block get overwritten by ftrace_make_nop / ftrace_make_call.
+ * Total length is 24 bytes. The complete mcount block initially gets replaced
+ * by ftrace_make_nop. Subsequent calls to ftrace_make_call / ftrace_make_nop
+ * only patch the jg/lg instruction within the block.
+ * Note: we do not patch the first instruction to an unconditional branch,
+ * since that would break kprobes/jprobes. It is easier to leave the larl
+ * instruction in and only modify the second instruction.
  * The 64-bit enabled ftrace code block looks like this:
- *	stg	%r14,8(%r15)		# offset 0
+ *	larl	%r0,.+24		# offset 0
  * >	lg	%r1,__LC_FTRACE_FUNC	# offset 6
- * >	lgr	%r0,%r0			# offset 12
- * >	basr	%r14,%r1		# offset 16
- *	lg	%r14,8(%15)		# offset 18
- * The return points of the mcount/ftrace function have the same offset 18.
- * The 64-bit disable ftrace code block looks like this:
- *	stg	%r14,8(%r15)		# offset 0
+ *	br	%r1			# offset 12
+ *	brcl	0,0			# offset 14
+ *	brc	0,0			# offset 20
+ * The ftrace function gets called with a non-standard C function call ABI
+ * where r0 contains the return address. It is also expected that the called
+ * function only clobbers r0 and r1, but restores r2-r15.
+ * The return point of the ftrace function has offset 24, so execution
+ * continues behind the mcount block.
+ *	larl	%r0,.+24		# offset 0
  * >	jg	.+18			# offset 6
- * >	lgr	%r0,%r0			# offset 12
- * >	basr	%r14,%r1		# offset 16
- *	lg	%r14,8(%15)		# offset 18
+ *	br	%r1			# offset 12
+ *	brcl	0,0			# offset 14
+ *	brc	0,0			# offset 20
  * The jg instruction branches to offset 24 to skip as many instructions
  * as possible.
  */
 asm(
 	"	.align	4\n"
+	"mcount_replace_code:\n"
+	"	larl	%r0,0f\n"
 	"ftrace_disable_code:\n"
 	"	jg	0f\n"
-	"	lgr	%r0,%r0\n"
-	"	basr	%r14,%r1\n"
+	"	br	%r1\n"
+	"	brcl	0,0\n"
+	"	brc	0,0\n"
 	"0:\n"
 	"	.align	4\n"
 	"ftrace_enable_insn:\n"
 	"	lg	%r1,"__stringify(__LC_FTRACE_FUNC)"\n");
 
+#define MCOUNT_BLOCK_SIZE	24
+#define MCOUNT_INSN_OFFSET	6
 #define FTRACE_INSN_SIZE	6
 
 #else /* CONFIG_64BIT */
@@ -116,6 +129,16 @@ int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
 int ftrace_make_nop(struct module *mod, struct dyn_ftrace *rec,
 		    unsigned long addr)
 {
+#ifdef CONFIG_64BIT
+	/* Initial replacement of the whole mcount block */
+	if (addr == MCOUNT_ADDR) {
+		if (probe_kernel_write((void *) rec->ip - MCOUNT_INSN_OFFSET,
+				       mcount_replace_code,
+				       MCOUNT_BLOCK_SIZE))
+			return -EPERM;
+		return 0;
+	}
+#endif
 	if (probe_kernel_write((void *) rec->ip, ftrace_disable_code,
 			       MCOUNT_INSN_SIZE))
 		return -EPERM;

arch/s390/kernel/mcount64.S

@@ -16,7 +16,6 @@ ENTRY(ftrace_stub)
 	br	%r14
 
 #define STACK_FRAME_SIZE  (STACK_FRAME_OVERHEAD + __PT_SIZE)
-#define STACK_PARENT_IP   (STACK_FRAME_SIZE + 8)
 #define STACK_PTREGS	  (STACK_FRAME_OVERHEAD)
 #define STACK_PTREGS_GPRS (STACK_PTREGS + __PT_GPRS)
 #define STACK_PTREGS_PSW  (STACK_PTREGS + __PT_PSW)
@@ -31,40 +30,39 @@ ENTRY(ftrace_caller)
 	aghi	%r15,-STACK_FRAME_SIZE
 	stg	%r1,__SF_BACKCHAIN(%r15)
 	stg	%r1,(STACK_PTREGS_GPRS+15*8)(%r15)
-	stmg	%r0,%r13,STACK_PTREGS_GPRS(%r15)
-	stg	%r14,(STACK_PTREGS_PSW+8)(%r15)
+	stg	%r0,(STACK_PTREGS_PSW+8)(%r15)
+	stmg	%r2,%r14,(STACK_PTREGS_GPRS+2*8)(%r15)
 #ifdef CONFIG_HAVE_MARCH_Z196_FEATURES
-	aghik	%r2,%r14,-MCOUNT_INSN_SIZE
+	aghik	%r2,%r0,-MCOUNT_INSN_SIZE
 	lgrl	%r4,function_trace_op
-	lgrl	%r14,ftrace_trace_function
+	lgrl	%r1,ftrace_trace_function
 #else
-	lgr	%r2,%r14
+	lgr	%r2,%r0
 	aghi	%r2,-MCOUNT_INSN_SIZE
 	larl	%r4,function_trace_op
 	lg	%r4,0(%r4)
-	larl	%r14,ftrace_trace_function
-	lg	%r14,0(%r14)
+	larl	%r1,ftrace_trace_function
+	lg	%r1,0(%r1)
 #endif
-	lg	%r3,STACK_PARENT_IP(%r15)
+	lgr	%r3,%r14
 	la	%r5,STACK_PTREGS(%r15)
-	basr	%r14,%r14
+	basr	%r14,%r1
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 # The j instruction gets runtime patched to a nop instruction.
 # See ftrace_enable_ftrace_graph_caller. The patched instruction is:
 #	j	.+4
 ENTRY(ftrace_graph_caller)
 	j	ftrace_graph_caller_end
-	lg	%r2,STACK_PARENT_IP(%r15)
+	lg	%r2,(STACK_PTREGS_GPRS+14*8)(%r15)
 	lg	%r3,(STACK_PTREGS_PSW+8)(%r15)
 	brasl	%r14,prepare_ftrace_return
-	stg	%r2,STACK_PARENT_IP(%r15)
+	stg	%r2,(STACK_PTREGS_GPRS+14*8)(%r15)
 ftrace_graph_caller_end:
 	.globl	ftrace_graph_caller_end
 #endif
-	lmg	%r0,%r13,STACK_PTREGS_GPRS(%r15)
-	lg	%r14,(STACK_PTREGS_PSW+8)(%r15)
-	aghi	%r15,STACK_FRAME_SIZE
-	br	%r14
+	lg	%r1,(STACK_PTREGS_PSW+8)(%r15)
+	lmg	%r2,%r15,(STACK_PTREGS_GPRS+2*8)(%r15)
+	br	%r1
 
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER