From mboxrd@z Thu Jan  1 00:00:00 1970
From: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Date: Sat, 28 Apr 2018 15:39:00 +0200
Subject: [Buildroot] [PATCH 1/1] zeromq: add libatomic linkage when
 available
In-Reply-To: <20180426195309.26042-1-asafka7@gmail.com>
References: <20180426195309.26042-1-asafka7@gmail.com>
Message-ID: <20180428153900.0f809228@windsurf>
List-Id: <buildroot.busybox.net>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: buildroot@busybox.net

Hello Asaf,

On Thu, 26 Apr 2018 22:53:09 +0300, Asaf Kahlon wrote:
> Fixes:
> http://autobuild.buildroot.net/results/ff5f7d3bcd713db9c9d92a746d6ad7134748fe16
> http://autobuild.buildroot.net/results/9a743da1c3be673288c3482a9e7ddea12750206f
> http://autobuild.buildroot.net/results/818b12387e6117569c4fab310b98c11fac80c047
> 
> If libatomic is available, it should be added to LIBS in CONF_ENV, since without
> it we can get into a state in which symbols like __atomic_fetch_sub_4,
> __atomic_compare_exchange_4 and __atomic_fetch_add_4 can't be found.
> 
> Signed-off-by: Asaf Kahlon <asafka7@gmail.com>
> ---
>  package/zeromq/zeromq.mk | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/package/zeromq/zeromq.mk b/package/zeromq/zeromq.mk
> index 8273cad763..214bd9e35c 100644
> --- a/package/zeromq/zeromq.mk
> +++ b/package/zeromq/zeromq.mk
> @@ -58,4 +58,8 @@ else
>  ZEROMQ_CONF_OPTS += --disable-libunwind
>  endif
>  
> +ifeq ($(BR2_TOOLCHAIN_HAS_LIBATOMIC),y)
> +ZEROMQ_CONF_OPTS += LIBS=-latomic
> +endif

While this does fix the problem, I don't think it's the most ideal fix.
Indeed, the solution should be upstreamed. Currently, zeromq
configure.ac uses LIBZMQ_CHECK_ATOMIC_INTRINSICS, which is implemented
in acinclude.m4, to detect the availability of atomic intrinsics.

It goes like this:

AC_DEFUN([LIBZMQ_CHECK_ATOMIC_INTRINSICS], [{
    AC_MSG_CHECKING(whether compiler supports __atomic_Xxx intrinsics)
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
/* atomic intrinsics test */
int v = 0;
int main (int, char **)
{
    int t = __atomic_add_fetch (&v, 1, __ATOMIC_ACQ_REL);
    return t;
}
    ])],
    [AC_MSG_RESULT(yes) ; libzmq_cv_has_atomic_instrisics="yes" ; $1],
    [AC_MSG_RESULT(no)  ; libzmq_cv_has_atomic_instrisics="no"  ; $2]
    )
}])

It does only an AC_COMPILE_IFELSE(), so it succeeds on Sparc, as the
need for -latomic only appears at link time.

So I believe it should be changed to test with AC_LINK_IFELSE, and if
it fails, test with AC_LINK_IFELSE with -latomic, to determine

 (1) If atomic intrinsics are available

 (2) If -latomic is needed to use atomic intrinsics

Do you think you could have a look at implementing something like this ?

That being said, I agree that a number of other packages in Buildroot
are fixed by just passing LIBS=-latomic.

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
https://bootlin.com