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 <a...@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.
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.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.
* lib/target-supports.exp (check_effective_target_fentry): New.
OK for master?
Thanks.
--
H.J.
From 99d64e9355dfbb8e07582c62589530e6ad0257db Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hjl.to...@gmail.com>
Date: Wed, 2 Jul 2025 08:58:23 +0800
Subject: [PATCH] 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 <a...@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.
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.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.
* lib/target-supports.exp (check_effective_target_fentry): New.
Signed-off-by: H.J. Lu <hjl.to...@gmail.com>
---
gcc/config.in | 6 +++
gcc/config/i386/i386-options.cc | 11 ++++-
gcc/configure | 46 +++++++++++++++++++-
gcc/configure.ac | 35 +++++++++++++++
gcc/doc/install.texi | 11 +++++
gcc/testsuite/gcc.target/i386/pr120881-1a.c | 4 ++
gcc/testsuite/gcc.target/i386/pr120881-1b.c | 4 ++
gcc/testsuite/gcc.target/i386/pr120881-1c.c | 3 ++
gcc/testsuite/gcc.target/i386/pr120881-1d.c | 3 ++
gcc/testsuite/gcc.target/i386/pr120881-2a.c | 21 +++++++++
gcc/testsuite/gcc.target/i386/pr120881-2b.c | 6 +++
gcc/testsuite/lib/target-supports.exp | 48 +++++++++++++++++++++
12 files changed, 195 insertions(+), 3 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr120881-1a.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr120881-1b.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr120881-1c.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr120881-1d.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr120881-2a.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr120881-2b.c
diff --git a/gcc/config.in b/gcc/config.in
index ab62c1566cb..353d1bc9407 100644
--- a/gcc/config.in
+++ b/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
diff --git a/gcc/config/i386/i386-options.cc b/gcc/config/i386/i386-options.cc
index 09cb1337f94..53658496efd 100644
--- a/gcc/config/i386/i386-options.cc
+++ b/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");
diff --git a/gcc/configure b/gcc/configure
index f056cfe9677..5dce9dfc60b 100755
--- a/gcc/configure
+++ b/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 "${target}" in
+ i?86-*-linux* | x86_64-*-linux*)
+ # Enable -mfentry by default on Linux.
+ enable_x86_64_mfentry=yes
+ ;;
+ esac
+fi
+
+gif=`if test x$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
diff --git a/gcc/configure.ac b/gcc/configure.ac
index 58bf63f8be9..e1fb1d33682 100644
--- a/gcc/configure.ac
+++ b/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 "${target}" in
+ i?86-*-linux* | x86_64-*-linux*)
+ # Enable -mfentry by default on Linux.
+ enable_x86_64_mfentry=yes
+ ;;
+ esac
+fi
+
+gif=`if test x$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)
diff --git a/gcc/doc/install.texi b/gcc/doc/install.texi
index 80ee2cd6eba..09ea87aa812 100644
--- a/gcc/doc/install.texi
+++ b/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.
diff --git a/gcc/testsuite/gcc.target/i386/pr120881-1a.c b/gcc/testsuite/gcc.target/i386/pr120881-1a.c
new file mode 100644
index 00000000000..3d9ac0e9e86
--- /dev/null
+++ b/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 } */
diff --git a/gcc/testsuite/gcc.target/i386/pr120881-1b.c b/gcc/testsuite/gcc.target/i386/pr120881-1b.c
new file mode 100644
index 00000000000..082640726b1
--- /dev/null
+++ b/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 } */
diff --git a/gcc/testsuite/gcc.target/i386/pr120881-1c.c b/gcc/testsuite/gcc.target/i386/pr120881-1c.c
new file mode 100644
index 00000000000..c21979f8eb1
--- /dev/null
+++ b/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" } */
diff --git a/gcc/testsuite/gcc.target/i386/pr120881-1d.c b/gcc/testsuite/gcc.target/i386/pr120881-1d.c
new file mode 100644
index 00000000000..f74af23ff5c
--- /dev/null
+++ b/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" } */
diff --git a/gcc/testsuite/gcc.target/i386/pr120881-2a.c b/gcc/testsuite/gcc.target/i386/pr120881-2a.c
new file mode 100644
index 00000000000..c7be72ace57
--- /dev/null
+++ b/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 "*-*-*" } {^(?:1|\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 ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr120881-2b.c b/gcc/testsuite/gcc.target/i386/pr120881-2b.c
new file mode 100644
index 00000000000..43a12f00774
--- /dev/null
+++ b/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"
+
diff --git a/gcc/testsuite/lib/target-supports.exp b/gcc/testsuite/lib/target-supports.exp
index 25ceeea95cb..c0474f519b7 100644
--- a/gcc/testsuite/lib/target-supports.exp
+++ b/gcc/testsuite/lib/target-supports.exp
@@ -14509,3 +14509,51 @@ 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 isn't 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
+ }
+ }]
+}
--
2.50.0
