s390/ftrace: hotpatch support for function tracing

Make use of gcc's hotpatch support to generate better code for ftrace
function tracing.
The generated code now contains only a six byte nop in each function
prologue instead of a 24 byte code block which will be runtime patched to
support function tracing.
With the new code generation the runtime overhead for supporting function
tracing is close to zero, while the original code did show a significant
performance impact.

Acked-by: Steven Rostedt <[e-mail]>
Signed-off-by: Heiko Carstens <[e-mail]>
Signed-off-by: Martin Schwidefsky <[e-mail]>

arch/s390/Kconfig

···
 	select HAVE_BPF_JIT if 64BIT && PACK_STACK
 	select HAVE_CMPXCHG_DOUBLE
 	select HAVE_CMPXCHG_LOCAL
-	select HAVE_C_RECORDMCOUNT
 	select HAVE_DEBUG_KMEMLEAK
 	select HAVE_DYNAMIC_FTRACE if 64BIT
 	select HAVE_DYNAMIC_FTRACE_WITH_REGS if 64BIT

arch/s390/Makefile

···
 cflags-$(CONFIG_WARN_DYNAMIC_STACK) += -mwarn-dynamicstack
 endif
 
+ifdef CONFIG_FUNCTION_TRACER
+# make use of hotpatch feature if the compiler supports it
+cc_hotpatch	:= -mhotpatch=0,3
+ifeq ($(call cc-option-yn,$(cc_hotpatch)),y)
+CC_FLAGS_FTRACE := $(cc_hotpatch)
+KBUILD_AFLAGS	+= -DCC_USING_HOTPATCH
+KBUILD_CFLAGS	+= -DCC_USING_HOTPATCH
+endif
+endif
+
 KBUILD_CFLAGS	+= -mbackchain -msoft-float $(cflags-y)
 KBUILD_CFLAGS	+= -pipe -fno-strength-reduce -Wno-sign-compare
 KBUILD_AFLAGS	+= $(aflags-y)

arch/s390/include/asm/ftrace.h

···
 
 #define ARCH_SUPPORTS_FTRACE_OPS 1
 
+#ifdef CC_USING_HOTPATCH
+#define MCOUNT_INSN_SIZE	6
+#else
 #define MCOUNT_INSN_SIZE	24
 #define MCOUNT_RETURN_FIXUP	18
+#endif
 
 #ifndef __ASSEMBLY__
 
···
 static inline void ftrace_generate_nop_insn(struct ftrace_insn *insn)
 {
 #ifdef CONFIG_FUNCTION_TRACER
+#ifdef CC_USING_HOTPATCH
+	/* brcl 0,0 */
+	insn->opc = 0xc004;
+	insn->disp = 0;
+#else
 	/* jg .+24 */
 	insn->opc = 0xc0f4;
 	insn->disp = MCOUNT_INSN_SIZE / 2;
+#endif
 #endif
 }
 
 static inline int is_ftrace_nop(struct ftrace_insn *insn)
 {
 #ifdef CONFIG_FUNCTION_TRACER
+#ifdef CC_USING_HOTPATCH
+	if (insn->disp == 0)
+		return 1;
+#else
 	if (insn->disp == MCOUNT_INSN_SIZE / 2)
 		return 1;
+#endif
 #endif
 	return 0;
 }

arch/s390/kernel/Makefile

···
 
 ifdef CONFIG_FUNCTION_TRACER
 # Don't trace early setup code and tracing code
-CFLAGS_REMOVE_early.o = -pg
-CFLAGS_REMOVE_ftrace.o = -pg
+CFLAGS_REMOVE_early.o = $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_ftrace.o = $(CC_FLAGS_FTRACE)
 endif
 
 #

arch/s390/kernel/ftrace.c

···
  *	lg	%r14,8(%r15)		# offset 18
  * The jg instruction branches to offset 24 to skip as many instructions
  * as possible.
+ * In case we use gcc's hotpatch feature the original and also the disabled
+ * function prologue contains only a single six byte instruction and looks
+ * like this:
+ * >	brcl	0,0			# offset 0
+ * To enable ftrace the code gets patched like above and afterwards looks
+ * like this:
+ * >	brasl	%r0,ftrace_caller	# offset 0
  */
 
 unsigned long ftrace_plt;
···
 	if (probe_kernel_read(&old, (void *) rec->ip, sizeof(old)))
 		return -EFAULT;
 	if (addr == MCOUNT_ADDR) {
-		/* Initial code replacement; we expect to see stg r14,8(r15) */
+		/* Initial code replacement */
+#ifdef CC_USING_HOTPATCH
+		/* We expect to see brcl 0,0 */
+		ftrace_generate_nop_insn(&orig);
+#else
+		/* We expect to see stg r14,8(r15) */
 		orig.opc = 0xe3e0;
 		orig.disp = 0xf0080024;
+#endif
 		ftrace_generate_nop_insn(&new);
 	} else if (old.opc == BREAKPOINT_INSTRUCTION) {
 		/*

arch/s390/kernel/kprobes.c

···
 		/*
 		 * If kprobes patches the instruction that is morphed by
 		 * ftrace make sure that kprobes always sees the branch
-		 * "jg .+24" that skips the mcount block
+		 * "jg .+24" that skips the mcount block or the "brcl 0,0"
+		 * in case of hotpatch.
 		 */
 		ftrace_generate_nop_insn((struct ftrace_insn *)p->ainsn.insn);
 		p->ainsn.is_ftrace_insn = 1;

arch/s390/kernel/mcount.S

···
 	.globl	ftrace_regs_caller
 	.set	ftrace_regs_caller,ftrace_caller
 	lgr	%r1,%r15
+#ifndef CC_USING_HOTPATCH
 	aghi	%r0,MCOUNT_RETURN_FIXUP
+#endif
 	aghi	%r15,-STACK_FRAME_SIZE
 	stg	%r1,__SF_BACKCHAIN(%r15)
 	stg	%r1,(STACK_PTREGS_GPRS+15*8)(%r15)

scripts/recordmcount.pl

···
     $cc .= " -m32";
 
 } elsif ($arch eq "s390" && $bits == 64) {
-    $mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*R_390_(PC|PLT)32DBL\\s+_mcount\\+0x2\$";
-    $mcount_adjust = -14;
+    if ($cc =~ /-DCC_USING_HOTPATCH/) {
+	$mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*c0 04 00 00 00 00\\s*brcl\\s*0,[0-9a-f]+ <([^\+]*)>\$";
+	$mcount_adjust = 0;
+    } else {
+	$mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*R_390_(PC|PLT)32DBL\\s+_mcount\\+0x2\$";
+	$mcount_adjust = -14;
+    }
     $alignment = 8;
     $type = ".quad";
     $ld .= " -m elf64_s390";