This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Rework -fno-omit-frame-pointer support on i386
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: Adhemerval Zanella <adhemerval dot zanella at linaro dot org>
- Cc: GNU C Library <libc-alpha at sourceware dot org>
- Date: Fri, 17 Feb 2017 07:44:54 -0800
- Subject: Re: [PATCH] Rework -fno-omit-frame-pointer support on i386
- Authentication-results: sourceware.org; auth=none
- References: <1483966550-21038-1-git-send-email-adhemerval.zanella@linaro.org> <CAMe9rOpsn6_y3MY6X=j-L7FncmWdkoQkQazvxRQUz=W_msRCeQ@mail.gmail.com> <0e80c379-4c5b-9d00-7331-6c4035fac5e2@linaro.org>
On Fri, Feb 17, 2017 at 6:47 AM, Adhemerval Zanella
<adhemerval.zanella@linaro.org> wrote:
>
>
> On 16/02/2017 17:24, H.J. Lu wrote:
>> On Mon, Jan 9, 2017 at 4:55 AM, Adhemerval Zanella
>>> +# Check if CFLAGS sets -fno-omit-frame-pointer
>>> +AC_CACHE_CHECK(for if compiler flags sets -fno-omit-frame-pointer,
>>> + libc_cv_no_omit_frame_pointer,
>>> + [cat > conftest.c << EOF
>>> + void foo (void) { }
>>> +EOF
>>> + libc_cv_no_omit_frame_pointer=no
>>> + if AC_TRY_COMMAND(${CC-cc} $CFLAGS -S conftest.c -o conftest.s 1>&AS_MESSAGE_LOG_FD); then
>>> + if grep 'push.*%ebp' conftest.s >/dev/null &&
>>> + grep 'pop.*%ebp' conftest.s >/dev/null; then
>>> + libc_cv_no_omit_frame_pointer=yes
>>> + fi
>>> + fi
>>> + rm -f conftest.c conftest.s
>>> + ])
>>
>> I prefer something like
>>
>> [hjl@gnu-fer something like
>>
>> [hjl@gnu-6 tmp]$ cat ebp.c
>> void
>> foo (int i)
>> {
>> register int reg asm ("ebp") = i;
>> asm ("# %0" : : "r" (reg));
>> }
>> [hjl@gnu-6 tmp]$ gcc -S -m32 -Os ebp.c
>> ebp.c: In function ‘foo’:
>> ebp.c:6:1: error: bp cannot be used in asm here
>> }
>> ^
>> [hjl@gnu-6 tmp]$ gcc -S -m32 -O2 ebp.c
>> [hjl@gnu-6 tmp]$
>>
>>> +if test $libc_cv_no_omit_frame_pointer = yes; then
>>> + AC_DEFINE(NO_OMIT_FRAME_POINTER)
>>> +fi
>>
>> How about something like CAN_USE_REGISTER_ASM_EBP?
>
> I implemented both your suggestions in this version:
>
> [BZ #21029]
> * config.h.in [CAN_USE_REGISTER_ASM_EBP]: New define.
> * sysdeps/unix/sysv/linux/i386/Makefile
> [$(subdir) = elf] (sysdep-dl-routines): Add libc-do-syscall.
> (uses-6-syscall-arguments): Remove.
> [$(subdir) = misc] (CFLAGS-epoll_pwait.o): Likewise.
> [$(subdir) = misc] (CFLAGS-epoll_pwait.os): Likewise.
> [$(subdir) = misc] (CFLAGS-mmap.o): Likewise.
> [$(subdir) = misc] (CFLAGS-mmap.os): Likewise.
> [$(subdir) = misc] (CFLAGS-mmap64.o): Likewise.
> [$(subdir) = misc] (CFLAGS-mmap64.os): Likewise.
> [$(subdir) = misc] (CFLAGS-pselect.o): Likewise.
> [$(subdir) = misc] (cflags-pselect.o): Likewise.
> [$(subdir) = misc] (cflags-pselect.os): Likewise.
> [$(subdir) = misc] (cflags-rtld-mmap.os): Likewise.
> [$(subdir) = sysvipc] (cflags-semtimedop.o): Likewise.
> [$(subdir) = sysvipc] (cflags-semtimedop.os): Likewise.
> [$(subdir) = io] (CFLAGS-posix_fadvise64.o): Likewise.
> [$(subdir) = io] (CFLAGS-posix_fadvise64.os): Likewise.
> [$(subdir) = io] (CFLAGS-posix_fallocate.o): Likewise.
> [$(subdir) = io] (CFLAGS-posix_fallocate.os): Likewise.
> [$(subdir) = io] (CFLAGS-posix_fallocate64.o): Likewise.
> [$(subdir) = io] (CFLAGS-posix_fallocate64.os): Likewise.
> [$(subdir) = io] (CFLAGS-sync_file_range.o): Likewise.
> [$(subdir) = io] (CFLAGS-sync_file_range.os): Likewise.
> [$(subdir) = io] (CFLAGS-fallocate.o): Likewise.
> [$(subdir) = io] (CFLAGS-fallocate.os): Likewise.
> [$(subdir) = io] (CFLAGS-fallocate64.o): Likewise.
> [$(subdir) = io] (CFLAGS-fallocate64.os): Likewise.
> [$(subdir) = nptl] (CFLAGS-pthread_rwlock_timedrdlock.o):
> Likewise.
> [$(subdir) = nptl] (CFLAGS-pthread_rwlock_timedrdlock.os):
> Likewise.
> [$(subdir) = nptl] (CFLAGS-pthread_rwlock_timedrwlock.o):
> Likewise.
> [$(subdir) = nptl] (CFLAGS-pthread_rwlock_timedrwlock.os):
> Likewise.
> [$(subdir) = nptl] (CFLAGS-sem_wait.o): Likewise.
> [$(subdir) = nptl] (CFLAGS-sem_wait.os): Likewise.
> [$(subdir) = nptl] (CFLAGS-sem_timedwait.o): Likewise.
> [$(subdir) = nptl] (CFLAGS-sem_timedwait.os): Likewise.
> * sysdeps/unix/sysv/linux/i386/configure.ac: Add check if compiler allows
> ebp on inline assembly.
> * sysdeps/unix/sysv/linux/i386/configure: Regenerate.
> * sysdeps/unix/sysv/linux/i386/sysdep.h (OPTIMIZE_FOR_GCC_5):
> Set if CAN_USE_REGISTER_ASM_EBP is set.
> (check_consistency): Likewise.
> ---
>
> diff --git a/config.h.in b/config.h.in
> index 7bfe923..fb2cc51 100644
> --- a/config.h.in
> +++ b/config.h.in
> @@ -259,4 +259,8 @@
> /* Build glibc with tunables support. */
> #define HAVE_TUNABLES 0
>
> +/* Some compiler options may now allow to use ebp in __asm__ (used mainly
> + in i386 6 argument syscall issue). */
> +#define CAN_USE_REGISTER_ASM_EBP 0
> +
> #endif
> diff --git a/sysdeps/unix/sysv/linux/i386/Makefile b/sysdeps/unix/sysv/linux/i386/Makefile
> index 9609752..6aac0df 100644
> --- a/sysdeps/unix/sysv/linux/i386/Makefile
> +++ b/sysdeps/unix/sysv/linux/i386/Makefile
> @@ -1,47 +1,18 @@
> # The default ABI is 32.
> default-abi := 32
>
> -# %ebp is used to pass the 6th argument to system calls, so these
> -# system calls are incompatible with a frame pointer.
> -uses-6-syscall-arguments = -fomit-frame-pointer
> -
> ifeq ($(subdir),misc)
> sysdep_routines += ioperm iopl vm86
> -CFLAGS-epoll_pwait.o += $(uses-6-syscall-arguments)
> -CFLAGS-epoll_pwait.os += $(uses-6-syscall-arguments)
> -CFLAGS-mmap.o += $(uses-6-syscall-arguments)
> -CFLAGS-mmap.os += $(uses-6-syscall-arguments)
> -CFLAGS-mmap64.o += $(uses-6-syscall-arguments)
> -CFLAGS-mmap64.os += $(uses-6-syscall-arguments)
> -CFLAGS-pselect.o += $(uses-6-syscall-arguments)
> -CFLAGS-pselect.os += $(uses-6-syscall-arguments)
> -CFLAGS-rtld-mmap.os += $(uses-6-syscall-arguments)
> -endif
> -
> -ifeq ($(subdir),sysvipc)
> -CFLAGS-semtimedop.o += $(uses-6-syscall-arguments)
> -CFLAGS-semtimedop.os += $(uses-6-syscall-arguments)
> endif
>
> ifeq ($(subdir),elf)
> +sysdep-dl-routines += libc-do-syscall
> sysdep-others += lddlibc4
> install-bin += lddlibc4
> endif
>
> ifeq ($(subdir),io)
> sysdep_routines += libc-do-syscall
> -CFLAGS-posix_fadvise64.o += $(uses-6-syscall-arguments)
> -CFLAGS-posix_fadvise64.os += $(uses-6-syscall-arguments)
> -CFLAGS-posix_fallocate.o += $(uses-6-syscall-arguments)
> -CFLAGS-posix_fallocate.os += $(uses-6-syscall-arguments)
> -CFLAGS-posix_fallocate64.o += $(uses-6-syscall-arguments)
> -CFLAGS-posix_fallocate64.os += $(uses-6-syscall-arguments)
> -CFLAGS-sync_file_range.o += $(uses-6-syscall-arguments)
> -CFLAGS-sync_file_range.os += $(uses-6-syscall-arguments)
> -CFLAGS-fallocate.o += $(uses-6-syscall-arguments)
> -CFLAGS-fallocate.os += $(uses-6-syscall-arguments)
> -CFLAGS-fallocate64.o += $(uses-6-syscall-arguments)
> -CFLAGS-fallocate64.os += $(uses-6-syscall-arguments)
> endif
>
> ifeq ($(subdir),nptl)
> @@ -61,14 +32,6 @@ ifeq ($(subdir),nptl)
> # pull in __syscall_error routine
> libpthread-routines += sysdep
> libpthread-shared-only-routines += sysdep
> -CFLAGS-pthread_rwlock_timedrdlock.o += $(uses-6-syscall-arguments)
> -CFLAGS-pthread_rwlock_timedrdlock.os += $(uses-6-syscall-arguments)
> -CFLAGS-pthread_rwlock_timedwrlock.o += $(uses-6-syscall-arguments)
> -CFLAGS-pthread_rwlock_timedwrlock.os += $(uses-6-syscall-arguments)
> -CFLAGS-sem_wait.o += $(uses-6-syscall-arguments)
> -CFLAGS-sem_wait.os += $(uses-6-syscall-arguments)
> -CFLAGS-sem_timedwait.o += $(uses-6-syscall-arguments)
> -CFLAGS-sem_timedwait.os += $(uses-6-syscall-arguments)
> endif
>
> ifeq ($(subdir),rt)
> diff --git a/sysdeps/unix/sysv/linux/i386/configure b/sysdeps/unix/sysv/linux/i386/configure
> index eb72659..ae2c356 100644
> --- a/sysdeps/unix/sysv/linux/i386/configure
> +++ b/sysdeps/unix/sysv/linux/i386/configure
> @@ -3,5 +3,44 @@
>
> arch_minimum_kernel=2.6.32
>
> +# Check if CFLAGS allows compiler to use ebp register in inline assembly.
> +
> +{ $as_echo "$as_me:${as_lineno-$LINENO}: checking if compiler flags allows ebp in inline assembly" >&5
> +$as_echo_n "checking if compiler flags allows ebp in inline assembly... " >&6; }
> +if ${libc_cv_can_use_register_asm_ebp+:} false; then :
> + $as_echo_n "(cached) " >&6
> +else
> +
> +cat confdefs.h - <<_ACEOF >conftest.$ac_ext
> +/* end confdefs.h. */
> +
> + void foo (int i)
> + {
> + register int reg asm ("ebp") = i;
> + asm ("# %0" : : "r" (reg));
> + }
> +int
> +main ()
> +{
> +
> + ;
> + return 0;
> +}
> +_ACEOF
> +if ac_fn_c_try_compile "$LINENO"; then :
> + libc_cv_can_use_register_asm_ebp=yes
> +else
> + libc_cv_can_use_register_asm_ebp=no
> +fi
> +rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
> +
> +fi
> +{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libc_cv_can_use_register_asm_ebp" >&5
> +$as_echo "$libc_cv_can_use_register_asm_ebp" >&6; }
> +if test $libc_cv_can_use_register_asm_ebp = yes; then
> + $as_echo "#define CAN_USE_REGISTER_ASM_EBP 1" >>confdefs.h
> +
> +fi
> +
> libc_cv_gcc_unwind_find_fde=yes
> ldd_rewrite_script=sysdeps/unix/sysv/linux/ldd-rewrite.sed
> diff --git a/sysdeps/unix/sysv/linux/i386/configure.ac b/sysdeps/unix/sysv/linux/i386/configure.ac
> index 1a11da6..1cd632e 100644
> --- a/sysdeps/unix/sysv/linux/i386/configure.ac
> +++ b/sysdeps/unix/sysv/linux/i386/configure.ac
> @@ -3,5 +3,22 @@ GLIBC_PROVIDES dnl See aclocal.m4 in the top level source directory.
>
> arch_minimum_kernel=2.6.32
>
> +# Check if CFLAGS allows compiler to use ebp register in inline assembly.
> +AC_CACHE_CHECK([if compiler flags allows ebp in inline assembly],
> + libc_cv_can_use_register_asm_ebp, [
> +AC_COMPILE_IFELSE(
> + [AC_LANG_PROGRAM([
> + void foo (int i)
> + {
> + register int reg asm ("ebp") = i;
> + asm ("# %0" : : "r" (reg));
> + }])],
> + [libc_cv_can_use_register_asm_ebp=yes],
> + [libc_cv_can_use_register_asm_ebp=no])
> +])
> +if test $libc_cv_can_use_register_asm_ebp = yes; then
> + AC_DEFINE(CAN_USE_REGISTER_ASM_EBP)
> +fi
> +
> libc_cv_gcc_unwind_find_fde=yes
> ldd_rewrite_script=sysdeps/unix/sysv/linux/ldd-rewrite.sed
> diff --git a/sysdeps/unix/sysv/linux/i386/sysdep.h b/sysdeps/unix/sysv/linux/i386/sysdep.h
> index baf4642..c40db59 100644
> --- a/sysdeps/unix/sysv/linux/i386/sysdep.h
> +++ b/sysdeps/unix/sysv/linux/i386/sysdep.h
> @@ -44,9 +44,9 @@
> /* Since GCC 5 and above can properly spill %ebx with PIC when needed,
> we can inline syscalls with 6 arguments if GCC 5 or above is used
> to compile glibc. Disable GCC 5 optimization when compiling for
> - profiling since asm ("ebp") can't be used to put the 6th argument
> - in %ebp for syscall. */
> -#if __GNUC_PREREQ (5,0) && !defined PROF
> + profiling or when -fno-omit-frame-pointer is used since asm ("ebp")
> + can't be used to put the 6th argument in %ebp for syscall. */
> +#if __GNUC_PREREQ (5,0) && !defined PROF && CAN_USE_REGISTER_ASM_EBP
Should it be "defined CAN_USE_REGISTER_ASM_EBP"
since CAN_USE_REGISTER_ASM_EBP may not be defined.
> # define OPTIMIZE_FOR_GCC_5
> #endif
>
> @@ -611,7 +611,8 @@ struct libc_do_syscall_args
> #endif
>
> /* Consistency check for position-independent code. */
> -#if defined __PIC__ && !defined OPTIMIZE_FOR_GCC_5
> +#if defined __PIC__ && !defined OPTIMIZE_FOR_GCC_5 \
> + && !CAN_USE_REGISTER_ASM_EBP
No need for this. If CAN_USE_REGISTER_ASM_EBP
isn't defined, OPTIMIZE_FOR_GCC_5 won't be defined.
> # define check_consistency() \
> ({ int __res; \
> __asm__ __volatile__ \
> --
> 2.7.4
>
--
H.J.