[2.27] x86-64: Use IFUNC memcpy and mempcpy in libc.a

  On Sat, Jul 08, 2017 at 04:59:04PM -0700, H.J. Lu wrote:
> On Sat, Jul 8, 2017 at 4:12 PM, Tulio Magno Quites Machado Filho
> <tuliom@linux.vnet.ibm.com> wrote:
> > "H.J. Lu" <hjl.tools@gmail.com> writes:
> >
> >> On Sat, Jul 8, 2017 at 11:31 AM, Tulio Magno Quites Machado Filho
> >> <tuliom@linux.vnet.ibm.com> wrote:
> >>> Changes since version 1:
> >>>
> >>>  - Added a testcase.  This is now validating both statically and
> >>>    dynamically linked executables.
> >>>  - Fixed an issue in the $(foreach ..) in sysdeps/powerpc/powerpc64le/Makefile.
> >>>  - Added a comment to csu/libc-start.c
> >>>  - Added a comment to csu/libc-tls.c
> >>>
> >>> -- 8< --
> >>>
> >>> The patch proposed by Peter Bergner [1] to libgc in order to fix
> >>> [BZ #21707] adds a dependency on a symbol provided by the loader,
> >>> forcing the loader to be linked to tests after libgcc was linked.
> >>>
> >>> It also requires to read the thread pointer during IRELA relocations.
> >>>
> >>> Tested on powerpc, powerpc64, powerpc64le, s390x and x86_64.
> >>>
> >>> [1] https://sourceware.org/ml/libc-alpha/2017-06/msg01383.html
> >>>
> >>> 2017-07-08  Tulio Magno Quites Machado Filho  <tuliom@linux.vnet.ibm.com>
> >>>
> >>>         [BZ #21707]
> >>>         * csu/libc-start.c (LIBC_START_MAIN): Perform IREL{,A}
> >>>         relocations after initializing the TCB on statically linked
> >>>         executables..
> >>>         * csu/libc-tls.c (__libc_setup_tls): Add a comment about
> >>>         IREL{,A} relocations.
> >>>         * elf/Makefile (tests-static-normal): Add tst-tlsifunc-static.
> >>>         (tests): Add tst-tlsifunc.
> >>>         * elf/tst-tlsifunc.c: New file.
> >>>         * elf/tst-tlsifunc-static.c: Likewise.
> >>>         * sysdeps/powerpc/powerpc64le/Makefile (f128-loader-link): New
> >>>         variable.
> >>>         [$(subdir) = math] (test-float128% test-ifloat128%): Force
> >>>         linking to the loader after linking to libgcc.
> >>>         [$(subdir) = wcsmbs stdlib] (bug-strtod bug-strtod2 bug-strtod2)
> >>>         (tst-strtod-round tst-wcstod-round tst-strtod6 tst-strrom)
> >>>         (tst-strfrom-locale strfrom-skeleton): Likewise.
> >>> ---
> >>>  csu/libc-start.c                     | 11 +++---
> >>>  csu/libc-tls.c                       |  2 ++
> >>>  elf/Makefile                         |  5 +--
> >>>  elf/tst-tlsifunc-static.c            | 19 +++++++++++
> >>>  elf/tst-tlsifunc.c                   | 66 ++++++++++++++++++++++++++++++++++++
> >>>  sysdeps/powerpc/powerpc64le/Makefile | 10 ++++++
> >>>  6 files changed, 107 insertions(+), 6 deletions(-)
> >>>  create mode 100644 elf/tst-tlsifunc-static.c
> >>>  create mode 100644 elf/tst-tlsifunc.c
> >>>
> >>> diff --git a/csu/libc-start.c b/csu/libc-start.c
> >>> index c2dd159..84b7f99 100644
> >>> --- a/csu/libc-start.c
> >>> +++ b/csu/libc-start.c
> >>> @@ -188,12 +188,15 @@ LIBC_START_MAIN (int (*main) (int, char **, char ** MAIN_AUXVEC_DECL),
> >>>
> >>>    ARCH_INIT_CPU_FEATURES ();
> >>>
> >>> -  /* Perform IREL{,A} relocations.  */
> >>> -  apply_irel ();
> >>> -
> >>>    /* The stack guard goes into the TCB, so initialize it early.  */
> >>>    __libc_setup_tls ();
> >>>
> >>> +  /* Perform IREL{,A} relocations.
> >>> +     Note: the relocations must happen after TLS initialization so that
> >>> +     IFUNC resolvers can benefit from thread-local storage, e.g. powerpc's
> >>> +     hwcap and platform fields available in the TCB.  */
> >>> +  apply_irel ();
> >>> +
> >>>    /* Set up the stack checker's canary.  */
> >>>    uintptr_t stack_chk_guard = _dl_setup_stack_chk_guard (_dl_random);
> >>>  # ifdef THREAD_SET_STACK_GUARD
> >>> @@ -224,7 +227,7 @@ LIBC_START_MAIN (int (*main) (int, char **, char ** MAIN_AUXVEC_DECL),
> >>>    __pointer_chk_guard_local = pointer_chk_guard;
> >>>  # endif
> >>>
> >>> -#endif
> >>> +#endif /* !SHARED  */
> >>>
> >>
> >> apply_irel should be called as early as possible.
> >
> > Why?  Could you elaborate, please?
> >
> 
> To use IFUNC in static executables, apply_irel should be called before
> any functions with IFUNC implementation is called.  At the moment,
> a few functions are used before apply_irel is called.  To address it,
> we can move apply_irel forward.  Call it later makes it worse.
> 

FYI, this is the patch I am submitting for glibc 2.27.

H.J.
---
Since apply_irel is called before memcpy and mempcpy are called, we
an use IFUNC memcpy and mempcpy in libc.a.

	* sysdeps/x86_64/memmove.S (MEMCPY_SYMBOL): Don't check SHARED.
	(MEMPCPY_SYMBOL): Likewise.
	* sysdeps/x86_64/multiarch/memcpy-ssse3-back.S: Also include
	in libc.a.
	* sysdeps/x86_64/multiarch/memcpy-ssse3.S: Likewise.
	* sysdeps/x86_64/multiarch/memmove-avx512-no-vzeroupper.S:
	Likewise.
	* sysdeps/x86_64/multiarch/memcpy.c: Also include in libc.a.
	(__hidden_ver1): Don't use in libc.a.
	* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S
	(__mempcpy): Don't create a weak alias in libc.a.
	* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S: Don't
	check SHARED.
	* sysdeps/x86_64/multiarch/mempcpy.c: Also include in libc.a.
	(__hidden_ver1): Don't use in libc.a.
---
 sysdeps/x86_64/memmove.S                           |  4 ++--
 sysdeps/x86_64/multiarch/memcpy-ssse3-back.S       |  5 +----
 sysdeps/x86_64/multiarch/memcpy-ssse3.S            |  5 +----
 sysdeps/x86_64/multiarch/memcpy.c                  |  8 ++++----
 .../multiarch/memmove-avx512-no-vzeroupper.S       |  6 ------
 .../x86_64/multiarch/memmove-sse2-unaligned-erms.S |  4 +---
 .../x86_64/multiarch/memmove-vec-unaligned-erms.S  | 22 +++++-----------------
 sysdeps/x86_64/multiarch/mempcpy.c                 |  8 ++++----
 8 files changed, 18 insertions(+), 44 deletions(-)

[2.27] x86-64: Use IFUNC memcpy and mempcpy in libc.a

Commit Message

Comments

Patch