x86-64: Add --enable-x86-64-mfentry

When profiling is enabled with shrink wrapping, the mcount call may not
be placed at the function entry after

	pushq %rbp
	movq %rsp,%rbp

As the result, the profile data may be skewed which makes PGO less
effective.

Add --enable-x86-64-mfentry to enable -mfentry by default to use
__fentry__, added to glibc in 2010 by:

commit d22e4cc9397ed41534c9422d0b0ffef8c77bfa53
Author: Andi Kleen <ak@linux.intel.com>
Date:   Sat Aug 7 21:24:05 2010 -0700

    x86: Add support for frame pointer less mcount

instead of mcount, which is placed before the prologue so that -pg can
be used with -fshrink-wrap-separate enabled at -O1.  This option is
64-bit only because __fentry__ doesn't support PIC in 32-bit mode.  The
default it to enable -mfentry when targeting glibc.

Also warn -pg without -mfentry with shrink wrapping enabled.  The warning
is disable for PIC in 32-bit mode.

gcc/

	PR target/120881
	* config.in: Regenerated.
	* configure: Likewise.
	* configure.ac: Add --enable-x86-64-mfentry.
	* config/i386/i386-options.cc (ix86_option_override_internal):
	Enable __fentry__ in 64-bit mode if ENABLE_X86_64_MFENTRY is set
	to 1.  Warn -pg without -mfentry with shrink wrapping enabled.
	* doc/install.texi: Document --enable-x86-64-mfentry.

gcc/testsuite/

	PR target/120881
	* gcc.dg/20021014-1.c: Add additional -mfentry -fno-pic options
	for x86.
	* gcc.dg/aru-2.c: Likewise.
	* gcc.dg/nest.c: Likewise.
	* gcc.dg/pr32450.c: Likewise.
	* gcc.dg/pr43643.c: Likewise.
	* gcc.target/i386/pr104447.c: Likewise.
	* gcc.target/i386/pr113122-3.c: Likewise.
	* gcc.target/i386/pr119386-1.c: Add additional -mfentry if not
	ia32.
	* gcc.target/i386/pr119386-2.c: Likewise.
	* gcc.target/i386/pr120881-1a.c: New test.
	* gcc.target/i386/pr120881-1b.c: Likewise.
	* gcc.target/i386/pr120881-1c.c: Likewise.
	* gcc.target/i386/pr120881-1d.c: Likewise.
	* gcc.target/i386/pr120881-2a.c: Likewise.
	* gcc.target/i386/pr120881-2b.c: Likewise.
	* gcc.target/i386/pr82699-1.c: Add additional -mfentry.
	* lib/target-supports.exp (check_effective_target_fentry): New.

Signed-off-by: H.J. Lu <hjl.tools@gmail.com>

gcc/config.in

@@ -318,6 +318,12 @@
 #endif
 
 
+/* Define to enable -mfentry by default on x86-64. */
+#ifndef USED_FOR_TARGET
+#undef ENABLE_X86_64_MFENTRY
+#endif
+
+
 /* Define to the name of a file containing a list of extra machine modes for
    this architecture. */
 #ifndef USED_FOR_TARGET

gcc/config/i386/i386-options.cc

@@ -2839,7 +2839,9 @@ ix86_option_override_internal (bool main_args_p,
 
   /* Set the default value for -mfentry.  */
   if (!opts_set->x_flag_fentry)
-    opts->x_flag_fentry = TARGET_SEH;
+    opts->x_flag_fentry = (TARGET_SEH
+			   || (TARGET_64BIT_P (opts->x_ix86_isa_flags)
+			       && ENABLE_X86_64_MFENTRY));
   else
     {
       if (!TARGET_64BIT_P (opts->x_ix86_isa_flags) && opts->x_flag_pic
@@ -2850,6 +2852,13 @@ ix86_option_override_internal (bool main_args_p,
 	sorry ("%<-mno-fentry%> isn%'t compatible with SEH");
     }
 
+  if (!opts->x_flag_fentry
+      && (TARGET_64BIT_P (opts->x_ix86_isa_flags) || !opts->x_flag_pic)
+      && opts->x_flag_shrink_wrap
+      && opts->x_profile_flag)
+    warning (0, "%<-pg%> without %<-mfentry%> may be unreliable with "
+	     "shrink wrapping");
+
   if (TARGET_SEH && TARGET_CALL_MS2SYSV_XLOGUES)
     sorry ("%<-mcall-ms2sysv-xlogues%> isn%'t currently supported with SEH");
 

gcc/configure

@@ -1064,6 +1064,7 @@ enable_versioned_jit
 enable_default_pie
 enable_cet
 enable_s390_excess_float_precision
+enable_x86_64_mfentry
 '
       ac_precious_vars='build_alias
 host_alias
@@ -1842,6 +1843,7 @@ Optional Features:
   --enable-s390-excess-float-precision
                           on s390 targets, evaluate float with double
                           precision when in standards-conforming mode
+  --enable-x86-64-mfentry enable -mfentry by default on x86-64 targets
 
 Optional Packages:
   --with-PACKAGE[=ARG]    use PACKAGE [ARG=yes]
@@ -21520,7 +21522,7 @@ else
   lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
   lt_status=$lt_dlunknown
   cat > conftest.$ac_ext <<_LT_EOF
-#line 21523 "configure"
+#line 21525 "configure"
 #include "confdefs.h"
 
 #if HAVE_DLFCN_H
@@ -21626,7 +21628,7 @@ else
   lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
   lt_status=$lt_dlunknown
   cat > conftest.$ac_ext <<_LT_EOF
-#line 21629 "configure"
+#line 21631 "configure"
 #include "confdefs.h"
 
 #if HAVE_DLFCN_H
@@ -35022,6 +35024,46 @@ $as_echo "#define ENABLE_S390_EXCESS_FLOAT_PRECISION 1" >>confdefs.h
   ;;
 esac
 
+# On x86-64, when profiling is enabled with shrink wrapping, the mcount
+# call may not be placed at the function entry after
+#	pushq %rbp
+#	movq %rsp,%rbp
+# As the result, the profile data may be skewed which makes PGO less
+# effective:
+# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120881
+# Enable -mfentry by default on x86-64 to put the profiling counter call
+# before the prologue.
+# Check whether --enable-x86-64-mfentry was given.
+if test "${enable_x86_64_mfentry+set}" = set; then :
+  enableval=$enable_x86_64_mfentry; case "${enableval}" in
+   yes | no | auto)
+     enable_x86_64_mfentry=$enableval
+     ;;
+   *)
+     as_fn_error $? "'$enable_x86_64_mfentry' is an invalid value for --enable-x86-64-mfentry.  Valid choices are 'yes', 'no' and 'auto'." "$LINENO" 5
+     ;;
+   esac
+else
+  enable_x86_64_mfentry=auto
+fi
+
+
+if test x"$enable_x86_64_mfentry" = xauto; then
+  case "${target}" in
+  i?86-*-*gnu* | x86_64-*-*gnu*)
+    # Enable -mfentry by default with glibc on x86.
+    enable_x86_64_mfentry=yes
+    ;;
+  esac
+fi
+
+gif=`if test x$enable_x86_64_mfentry = xyes; then echo 1; else echo 0; fi`
+
+cat >>confdefs.h <<_ACEOF
+#define ENABLE_X86_64_MFENTRY $gif
+_ACEOF
+
+
 # Check if the linker supports '-z now'
 ld_now_support=no
 { $as_echo "$as_me:${as_lineno-$LINENO}: checking linker -z now option" >&5

gcc/configure.ac

@@ -7972,6 +7972,41 @@ standards-compatible mode on s390 targets.])
   ;;
 esac
 
+# On x86-64, when profiling is enabled with shrink wrapping, the mcount
+# call may not be placed at the function entry after
+#	pushq %rbp
+#	movq %rsp,%rbp
+# As the result, the profile data may be skewed which makes PGO less
+# effective:
+# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120881
+# Enable -mfentry by default on x86-64 to put the profiling counter call
+# before the prologue.
+AC_ARG_ENABLE(x86-64-mfentry,
+  [AS_HELP_STRING([--enable-x86-64-mfentry],
+		  [enable -mfentry by default on x86-64 targets])],
+  [case "${enableval}" in
+   yes | no | auto)
+     enable_x86_64_mfentry=$enableval
+     ;;
+   *)
+     AC_MSG_ERROR(['$enable_x86_64_mfentry' is an invalid value for --enable-x86-64-mfentry.  Valid choices are 'yes', 'no' and 'auto'.])
+     ;;
+   esac],
+  [enable_x86_64_mfentry=auto])
+
+if test x"$enable_x86_64_mfentry" = xauto; then
+  case "${target}" in
+  i?86-*-*gnu* | x86_64-*-*gnu*)
+    # Enable -mfentry by default with glibc on x86.
+    enable_x86_64_mfentry=yes
+    ;;
+  esac
+fi
+
+gif=`if test x$enable_x86_64_mfentry = xyes; then echo 1; else echo 0; fi`
+AC_DEFINE_UNQUOTED(ENABLE_X86_64_MFENTRY, $gif,
+[Define to enable -mfentry by default on x86-64.])
+
 # Check if the linker supports '-z now'
 ld_now_support=no
 AC_MSG_CHECKING(linker -z now option)

gcc/doc/install.texi

@@ -2667,6 +2667,17 @@ target binutils supports @code{Intel CET} instructions and disabled
 otherwise.  In this case, the target libraries are configured to get
 additional @option{-fcf-protection} option.
 
+@item --enable-x86-64-mfentry
+@itemx --disable-x86-64-mfentry
+Enable @option {-mfentry} by default on x86-64 to put the profiling
+counter call, @code{__fentry__}, before the prologue so that @option{-pg}
+can be used with @option{-fshrink-wrap} which is enabled at @option{-O1}.
+This configure option is 64-bit only because @code{__fentry__} doesn't
+support PIC in 32-bit mode.
+
+@option{--enable-x86-64-mfentry=auto} is default.  @option{-mfentry} is
+enabled on Linux/x86-64 by default.
+
 @item --with-riscv-attribute=@samp{yes}, @samp{no} or @samp{default}
 Generate RISC-V attribute by default, in order to record extra build
 information in object.

gcc/testsuite/gcc.dg/20021014-1.c

@@ -2,6 +2,7 @@
 /* { dg-require-profiling "-p" } */
 /* { dg-options "-O2 -p" } */
 /* { dg-options "-O2 -p -static" { target hppa*-*-hpux* } } */
+/* { dg-additional-options "-mfentry -fno-pic" { target i?86-*-* x86_64-*-* } } */
 /* { dg-error "profiler" "No profiler support" { target xstormy16-*-* } 0 } */
 /* { dg-message "" "consider using `-pg' instead of `-p' with gprof(1)" { target *-*-freebsd* } 0 } */
 

gcc/testsuite/gcc.dg/aru-2.c

@@ -1,6 +1,7 @@
 /* { dg-do run } */
 /* { dg-require-profiling "-pg" } */
 /* { dg-options "-O2 -pg" } */
+/* { dg-additional-options "-mfentry -fno-pic" { target i?86-*-* x86_64-*-* } } */
 
 static int __attribute__((noinline))
 bar (int x)

gcc/testsuite/gcc.dg/nest.c

@@ -3,6 +3,7 @@
 /* { dg-require-profiling "-pg" } */
 /* { dg-options "-O2 -pg" } */
 /* { dg-options "-O2 -pg -static" { target hppa*-*-hpux* } } */
+/* { dg-additional-options "-mfentry -fno-pic" { target i?86-*-* x86_64-*-* } } */
 /* { dg-error "profiler" "No profiler support" { target xstormy16-*-* } 0 } */
 
 extern void abort (void);

gcc/testsuite/gcc.dg/pr32450.c

@@ -3,7 +3,7 @@
 /* { dg-do run } */
 /* { dg-require-profiling "-pg" } */
 /* { dg-options "-O2 -pg" } */
-/* { dg-options "-O2 -pg -mtune=core2" { target { i?86-*-* x86_64-*-* } } } */
+/* { dg-options "-O2 -pg -mtune=core2 -mfentry -fno-pic" { target { i?86-*-* x86_64-*-* } } } */
 /* { dg-options "-O2 -pg -static" { target hppa*-*-hpux* } } */
 
 extern void abort (void);

gcc/testsuite/gcc.dg/pr43643.c

@@ -4,6 +4,7 @@
 /* { dg-require-profiling "-pg" } */
 /* { dg-options "-O2 -pg" } */
 /* { dg-options "-O2 -pg -static" { target hppa*-*-hpux* } } */
+/* { dg-additional-options "-mfentry -fno-pic" { target i?86-*-* x86_64-*-* } } */
 
 extern char *strdup (const char *);
 

gcc/testsuite/gcc.target/i386/pr104447.c

@@ -1,6 +1,6 @@
 /* { dg-do compile } */
 /* { dg-require-profiling "-pg" } */
-/* { dg-options "-O2 -pg" } */
+/* { dg-options "-O2 -pg -mfentry -fno-pic" } */
 
 int
 bar (int x)

gcc/testsuite/gcc.target/i386/pr113122-3.c

@@ -1,7 +1,7 @@
 /* PR target/113122 */
 /* { dg-do assemble { target *-*-linux* } } */
 /* { dg-require-effective-target masm_intel } */
-/* { dg-options "-fprofile -O2 -masm=intel" } */
+/* { dg-options "-fprofile -mfentry -fno-pic -O2 -masm=intel" } */
 
 void
 func (void)

gcc/testsuite/gcc.target/i386/pr119386-1.c

@@ -1,7 +1,9 @@
 /* PR target/119386 */
 /* { dg-do compile { target *-*-linux* } } */
 /* { dg-options "-O2 -fpic -pg" } */
-/* { dg-final { scan-assembler "call\[ \t\]+mcount@PLT" } } */
+/* { dg-additional-options "-mfentry" { target { ! ia32 }  } } */
+/* { dg-final { scan-assembler "call\[ \t\]+mcount@PLT" { target ia32 } } } */
+/* { dg-final { scan-assembler "call\[ \t\]+__fentry__@PLT" { target { ! ia32 } } } } */
 
 int
 main ()

gcc/testsuite/gcc.target/i386/pr119386-2.c

@@ -1,7 +1,8 @@
 /* PR target/119386 */
 /* { dg-do compile { target *-*-linux* } } */
 /* { dg-options "-O2 -fpic -fno-plt -pg" } */
-/* { dg-final { scan-assembler "call\[ \t\]+\\*mcount@GOTPCREL\\(" { target { ! ia32 } } } } */
+/* { dg-additional-options "-mfentry" { target { ! ia32 }  } } */
+/* { dg-final { scan-assembler "call\[ \t\]+\\*__fentry__@GOTPCREL" { target { ! ia32 } } } } */
 /* { dg-final { scan-assembler "call\[ \t\]+\\*mcount@GOT\\(" { target ia32 } } } */
 
 

gcc/testsuite/gcc.target/i386/pr120881-1a.c

@@ -0,0 +1,4 @@
+/* { dg-do compile { target fpic } } */
+/* { dg-require-profiling "-pg" } */
+/* { dg-options "-O2 -pg -mno-fentry -fno-pic" } */
+/* { dg-message "'-pg' without '-mfentry' may be unreliable with shrink wrapping" "" { target *-*-* } 0 } */

gcc/testsuite/gcc.target/i386/pr120881-1b.c

@@ -0,0 +1,4 @@
+/* { dg-do compile { target { fpic && { ! ia32 } } } } */
+/* { dg-require-profiling "-pg" } */
+/* { dg-options "-O2 -pg -mno-fentry -fpic" } */
+/* { dg-message "'-pg' without '-mfentry' may be unreliable with shrink wrapping" "" { target *-*-* } 0 } */

gcc/testsuite/gcc.target/i386/pr120881-1c.c

@@ -0,0 +1,3 @@
+/* { dg-do compile { target { fpic && ia32 } } } */
+/* { dg-require-profiling "-pg" } */
+/* { dg-options "-O2 -pg -mno-fentry -fpic" } */

gcc/testsuite/gcc.target/i386/pr120881-1d.c

@@ -0,0 +1,3 @@
+/* { dg-do compile { target { fpic && ia32 } } } */
+/* { dg-require-profiling "-pg" } */
+/* { dg-options "-O2 -pg -mno-fentry -fno-shrink-wrap -fno-pic" } */

gcc/testsuite/gcc.target/i386/pr120881-2a.c

@@ -0,0 +1,21 @@
+/* { dg-do compile { target fentry } } */
+/* { dg-options "-O2 -pg" } */
+/* Keep labels and directives ('.cfi_startproc', '.cfi_endproc').  */
+/* { dg-final { check-function-bodies "**" "" "" { target "*-*-*" } {^\t?\.} } } */
+
+/*
+**f2:
+**.LFB[0-9]+:
+**	.cfi_startproc
+**	call	__fentry__
+**...
+*/
+
+extern void f1 (void);
+
+void
+f2 (int count)
+{
+  for (int i = 0; i < count; ++i)
+    f1 ();
+}

gcc/testsuite/gcc.target/i386/pr120881-2b.c

@@ -0,0 +1,6 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-rtl-pro_and_epilogue -march=x86-64" } */
+/* { dg-final { scan-rtl-dump "Now spread 1 times" "pro_and_epilogue" } } */
+
+#include "pr120881-2a.c"
+

gcc/testsuite/gcc.target/i386/pr82699-1.c

@@ -1,5 +1,5 @@
 /* { dg-do compile { target *-*-linux* } } */
-/* { dg-options "-O2 -fno-pic -fcf-protection -pg -fasynchronous-unwind-tables" } */
+/* { dg-options "-O2 -mfentry -fno-pic -fcf-protection -pg -fasynchronous-unwind-tables" } */
 /* { dg-final { scan-assembler-times {\t\.cfi_startproc\n\tendbr} 1 } } */
 
 extern int bar (int);

gcc/testsuite/lib/target-supports.exp

@@ -14528,3 +14528,51 @@ proc check_effective_target_foldable_pi_based_trigonometry { } {
 	}
     }]
 }
+#
+# Return 1 if the x86-64 target enables -mfentry by default, 0
+# otherwise.  Cache the result.
+
+proc check_effective_target_fentry { } {
+    global tool
+    global GCC_UNDER_TEST
+
+    if { ![check_effective_target_x86] } {
+	return 0
+    }
+
+    # Need auto-host.h to check linker support.
+    if { ![file exists ../../auto-host.h ] } {
+	return 0
+    }
+
+    return [check_cached_effective_target fentry {
+	# Set up and compile to see if ENABLE_X86_64_MFENTRY is
+	# non-zero.  Include the current process ID in the file
+	# names to prevent conflicts with invocations for multiple
+	# testsuites.
+
+	set src pie[pid].c
+	set obj pie[pid].o
+
+	set f [open $src "w"]
+	puts $f "#include \"../../auto-host.h\""
+	puts $f "#if ENABLE_X86_64_MFENTRY == 0 || !defined __x86_64__"
+	puts $f "# error -mfentry is not enabled by default."
+	puts $f "#endif"
+	close $f
+
+	verbose "check_effective_target_fentry compiling testfile $src" 2
+	set lines [${tool}_target_compile $src $obj object ""]
+
+	file delete $src
+	file delete $obj
+
+	if [string match "" $lines] then {
+	    verbose "check_effective_target_fentry testfile compilation passed" 2
+	    return 1
+	} else {
+	    verbose "check_effective_target_fentry testfile compilation failed" 2
+	    return 0
+	}
+    }]
+}