From: "H.J. Lu" <hjl.tools@gmail.com>
To: Fangrui Song <maskray@google.com>
Cc: GNU C Library <libc-alpha@sourceware.org>,
Florian Weimer <fweimer@redhat.com>
Subject: Re: [PATCH v2] Set the retain attribute on _elf_set_element if CC supports [BZ #27492]
Date: Mon, 29 Mar 2021 16:11:38 -0700 [thread overview]
Message-ID: <CAMe9rOqYiQPrYqt-wLA9qyDP_wH13T54fAChvzwsJ7zLG19J3Q@mail.gmail.com> (raw)
In-Reply-To: <20210323041457.p7ds4qp7qchxobk5@google.com>
On Mon, Mar 22, 2021 at 9:15 PM Fangrui Song <maskray@google.com> wrote:
>
>
> On 2021-03-22, H.J. Lu wrote:
> >On Sun, Mar 21, 2021 at 9:40 PM Fangrui Song <maskray@google.com> wrote:
> >>
> >> On 2021-03-21, H.J. Lu wrote:
> >> >On Sun, Mar 21, 2021 at 9:01 PM Fangrui Song via Libc-alpha
> >> ><libc-alpha@sourceware.org> wrote:
> >> >>
> >> >> So that text_set_element/data_set_element/bss_set_element defined
> >> >> variables will be retained by the linker.
> >> >>
> >> >> Note: 'used' and 'retain' are orthogonal: 'used' makes sure the variable
> >> >> will not be optimized out; 'retain' prevents section garbage collection
> >> >> if the linker support SHF_GNU_RETAIN.
> >> >>
> >> >> GNU ld 2.37 and LLD 13 will support -z start-stop-gc which allow C
> >> >> identifier name sections to be GCed even if there are live
> >> >> __start_/__stop_ references.
> >> >>
> >> >> Without the change, there are some static linking problems, e.g.
> >> >> _IO_cleanup (libio/genops.c) may be discarded by ld --gc-sections, so
> >> >> stdout is not flushed on exit.
> >> >>
> >> >> Note: GCC may warning ‘retain’ attribute ignored while __has_attribute(retain) is 1
> >> >> (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99587).
> >> >> ---
> >> >> config.h.in | 3 +++
> >> >> configure | 23 +++++++++++++++++++++++
> >> >> configure.ac | 16 ++++++++++++++++
> >> >> include/libc-symbols.h | 30 +++++++++++++++++++++---------
> >> >> 4 files changed, 63 insertions(+), 9 deletions(-)
> >> >>
> >> >> diff --git a/config.h.in b/config.h.in
> >> >> index f21bf04e47..cc875e90f9 100644
> >> >> --- a/config.h.in
> >> >> +++ b/config.h.in
> >> >> @@ -187,6 +187,9 @@
> >> >> /* Define if gcc supports attribute ifunc. */
> >> >> #undef HAVE_GCC_IFUNC
> >> >>
> >> >> +/* Define if gcc supports attribute retain. */
> >> >> +#undef HAVE_GCC_RETAIN
> >> >> +
> >> >> /* Define if the linker defines __ehdr_start. */
> >> >> #undef HAVE_EHDR_START
> >> >>
> >> >> diff --git a/configure b/configure
> >> >> index 37cef37413..28494de748 100755
> >> >> --- a/configure
> >> >> +++ b/configure
> >> >> @@ -4105,6 +4105,29 @@ fi
> >> >> $as_echo "$libc_cv_textrel_ifunc" >&6; }
> >> >>
> >> >>
> >> >> +# Check if gcc supports attribute ifunc as it is used in libc_ifunc macro.
> >> >> +{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for gcc attribute retain support" >&5
> >> >> +$as_echo_n "checking for gcc attribute retain support... " >&6; }
> >> >> +if ${libc_cv_gcc_retain+:} false; then :
> >> >> + $as_echo_n "(cached) " >&6
> >> >> +else
> >> >> + cat > conftest.c <<EOF
> >> >> +static int var __attribute__ ((used, retain, section ("__libc_atexit")));
> >> >> +EOF
> >> >> +libc_cv_gcc_retain=no
> >> >> +if ${CC-cc} -Werror -c conftest.c -o /dev/null 1>&5 \
> >> >> + 2>&5 ; then
> >> >> + libc_cv_gcc_retain=yes
> >> >> +fi
> >> >> +rm -f conftest*
> >> >> +fi
> >> >> +{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libc_cv_gcc_retain" >&5
> >> >> +$as_echo "$libc_cv_gcc_retain" >&6; }
> >> >> +if test $libc_cv_gcc_retain = yes; then
> >> >> + $as_echo "#define HAVE_GCC_RETAIN 1" >>confdefs.h
> >> >> +
> >> >> +fi
> >> >> +
> >> >> # Check if gcc warns about alias for function with incompatible types.
> >> >> { $as_echo "$as_me:${as_lineno-$LINENO}: checking if compiler warns about alias for function with incompatible types" >&5
> >> >> $as_echo_n "checking if compiler warns about alias for function with incompatible types... " >&6; }
> >> >> diff --git a/configure.ac b/configure.ac
> >> >> index 16b15b6f90..94ad713820 100644
> >> >> --- a/configure.ac
> >> >> +++ b/configure.ac
> >> >> @@ -707,6 +707,22 @@ fi
> >> >> rm -f conftest*])
> >> >> AC_SUBST(libc_cv_textrel_ifunc)
> >> >>
> >> >> +# Check if gcc supports attribute ifunc as it is used in libc_ifunc macro.
> >> >> +AC_CACHE_CHECK([for gcc attribute retain support],
> >> >> + libc_cv_gcc_retain, [dnl
> >> >> +cat > conftest.c <<EOF
> >> >> +static int var __attribute__ ((used, retain, section ("__libc_atexit")));
> >> >> +EOF
> >> >> +libc_cv_gcc_retain=no
> >> >> +if ${CC-cc} -Werror -c conftest.c -o /dev/null 1>&AS_MESSAGE_LOG_FD \
> >> >> + 2>&AS_MESSAGE_LOG_FD ; then
> >> >> + libc_cv_gcc_retain=yes
> >> >> +fi
> >> >> +rm -f conftest*])
> >> >> +if test $libc_cv_gcc_retain = yes; then
> >> >> + AC_DEFINE(HAVE_GCC_RETAIN)
> >> >> +fi
> >> >> +
> >> >> # Check if gcc warns about alias for function with incompatible types.
> >> >> AC_CACHE_CHECK([if compiler warns about alias for function with incompatible types],
> >> >> libc_cv_gcc_incompatible_alias, [dnl
> >> >> diff --git a/include/libc-symbols.h b/include/libc-symbols.h
> >> >> index c83e550b03..f8eafcf1f3 100644
> >> >> --- a/include/libc-symbols.h
> >> >> +++ b/include/libc-symbols.h
> >> >> @@ -373,17 +373,29 @@ for linking")
> >> >>
> >> >> /* These are all done the same way in ELF.
> >> >> There is a new section created for each set. */
> >> >> -#ifdef SHARED
> >> >> +#ifdef HAVE_GCC_RETAIN
> >> >> +# ifdef SHARED
> >> >> /* When building a shared library, make the set section writable,
> >> >> - because it will need to be relocated at run time anyway. */
> >> >> -# define _elf_set_element(set, symbol) \
> >> >> - static const void *__elf_set_##set##_element_##symbol##__ \
> >> >> - __attribute__ ((used, section (#set))) = &(symbol)
> >> >
> >> >You should define __attribute_used_and_retain__ instead.
> >> >
> >> >> + because it will need to be relocated at run time anyway. */
> >> >> +# define _elf_set_element(set, symbol) \
> >> >> + static const void *__elf_set_##set##_element_##symbol##__ \
> >> >> + __attribute__ ((used, retain, section (#set))) = &(symbol)
> >> >> +# else
> >> >> +# define _elf_set_element(set, symbol) \
> >> >> + static const void *const __elf_set_##set##_element_##symbol##__ \
> >> >> + __attribute__ ((used, retain, section (#set))) = &(symbol)
> >> >> +# endif
> >> >> #else
> >> >> -# define _elf_set_element(set, symbol) \
> >> >> - static const void *const __elf_set_##set##_element_##symbol##__ \
> >> >> - __attribute__ ((used, section (#set))) = &(symbol)
> >> >> -#endif
> >> >> +# ifdef SHARED
> >> >> +# define _elf_set_element(set, symbol) \
> >> >> + static const void *__elf_set_##set##_element_##symbol##__ \
> >> >> + __attribute__ ((used, section (#set))) = &(symbol)
> >> >> +# else
> >> >> +# define _elf_set_element(set, symbol) \
> >> >> + static const void *const __elf_set_##set##_element_##symbol##__ \
> >> >> + __attribute__ ((used, section (#set))) = &(symbol)
> >> >> +# endif
> >> >> +#endif /* HAVE_GCC_RETAIN */
> >> >>
> >> >> /* Define SET as a symbol set. This may be required (it is in a.out) to
> >> >> be able to use the set's contents. */
> >> >> --
> >> >> 2.31.0.rc2.261.g7f71774620-goog
> >> >>
> >> >
> >> >libc.a is an installed file. Does this work with older linkers?
> >> >
> >> >--
> >> >H.J.
> >>
> >> How about
> >>
> >> /* These are all done the same way in ELF.
> >> There is a new section created for each set. */
> >> #ifdef HAVE_GCC_RETAIN
> >> # define attribute_used_retain_section(sec) \
> >> __attribute__ ((used, retain, section (sec)))
> >> #else
> >
> >__attribute_used_and_retain__ is more flex and can be used
> >without section.
> >
> >> # define attribute_used_retain_section(sec) \
> >> __attribute__ ((used, section (sec)))
> >> #endif
> >> #ifdef SHARED
> >> /* When building a shared library, make the set section writable,
> >> because it will need to be relocated at run time anyway. */
> >> # define _elf_set_element(set, symbol) \
> >> static const void *__elf_set_##set##_element_##symbol##__ \
> >> attribute_used_retain_section(#set) = &(symbol)
> >> #else
> >> # define _elf_set_element(set, symbol) \
> >> static const void *const __elf_set_##set##_element_##symbol##__ \
> >> attribute_used_retain_section(#set) = &(symbol)
> >> #endif
> >> ? Yes, older linkers just ignore unknown section flags. That is the ELF spirit.
> >
> >
> >
> >--
> >H.J.
>
> +#ifdef HAVE_GCC_RETAIN
> +# define attribute_used_retain __attribute__ ((__used__, __retain__))
> +#else
> +# define attribute_used_retain __attribute__ ((__used__))
> +#endif
> +
> /* Symbol set support macros. */
>
> /* Make SYMBOL, which is in the text segment, an element of SET. */
> @@ -377,12 +383,12 @@ for linking")
> /* When building a shared library, make the set section writable,
> because it will need to be relocated at run time anyway. */
> # define _elf_set_element(set, symbol) \
> - static const void *__elf_set_##set##_element_##symbol##__ \
> - __attribute__ ((used, section (#set))) = &(symbol)
> + static const void *__elf_set_##set##_element_##symbol##__ \
> + attribute_used_retain __attribute__ ((section (#set))) = &(symbol)
> #else
> # define _elf_set_element(set, symbol) \
> - static const void *const __elf_set_##set##_element_##symbol##__ \
> - __attribute__ ((used, section (#set))) = &(symbol)
> + static const void *const __elf_set_##set##_element_##symbol##__ \
> + attribute_used_retain __attribute__ ((section (#set))) = &(symbol)
> #endif
>
Please submit a v2 patch.
Thanks.
--
H.J.
next prev parent reply other threads:[~2021-03-29 23:12 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-03-15 1:10 [PATCH] " Fangrui Song
2021-03-15 7:28 ` Florian Weimer
2021-03-15 7:45 ` Fāng-ruì Sòng
2021-03-15 8:13 ` Florian Weimer
2021-03-22 4:00 ` [PATCH v2] " Fangrui Song
2021-03-22 4:24 ` H.J. Lu
2021-03-22 4:40 ` Fangrui Song
2021-03-22 20:28 ` H.J. Lu
2021-03-23 4:14 ` Fangrui Song
2021-03-29 23:11 ` H.J. Lu [this message]
2021-03-29 23:24 ` [PATCH v3] " Fangrui Song
2021-03-30 18:17 ` H.J. Lu
2021-03-30 19:17 ` Fāng-ruì Sòng
2021-03-30 19:30 ` H.J. Lu
2021-03-31 4:34 ` Fāng-ruì Sòng
2021-03-31 16:17 ` H.J. Lu
2021-03-31 18:09 ` Fāng-ruì Sòng
2021-03-31 18:57 ` H.J. Lu
2021-03-31 19:39 ` [PATCH v4] " Fangrui Song
2021-03-31 23:38 ` H.J. Lu
2021-04-01 0:02 ` Fangrui Song
2021-04-01 1:00 ` H.J. Lu
2021-04-01 1:06 ` [PATCH v5] " Fangrui Song
2021-04-01 12:52 ` H.J. Lu
2021-04-02 3:23 ` Fangrui Song
2021-04-02 14:14 ` H.J. Lu
2021-04-02 17:09 ` Fangrui Song
2021-04-02 17:33 ` H.J. Lu
2021-04-03 18:02 ` Fangrui Song
2021-04-03 20:47 ` H.J. Lu
2021-04-03 21:57 ` Fangrui Song
2021-04-05 1:55 ` H.J. Lu
2021-04-05 18:17 ` Fangrui Song
2021-04-05 20:35 ` H.J. Lu
2021-04-05 21:03 ` Fangrui Song
2021-04-05 21:58 ` H.J. Lu
2021-04-06 21:46 ` Fangrui Song
2021-04-09 22:36 ` Fāng-ruì Sòng
2021-04-16 3:51 ` Fāng-ruì Sòng
2021-04-16 13:00 ` H.J. Lu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAMe9rOqYiQPrYqt-wLA9qyDP_wH13T54fAChvzwsJ7zLG19J3Q@mail.gmail.com \
--to=hjl.tools@gmail.com \
--cc=fweimer@redhat.com \
--cc=libc-alpha@sourceware.org \
--cc=maskray@google.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).