Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <cover.1587401492.git.christophe.leroy@c-s.fr> <CAK8P3a11t=piuDq+TuGMo0xDiN06OLMxsym0PGiWNv5qYSUXLQ@mail.gmail.com>
 <268ff4cf-2f30-fead-fef5-e8e5f8f17aba@c-s.fr>
In-Reply-To: <268ff4cf-2f30-fead-fef5-e8e5f8f17aba@c-s.fr>
From:   Arnd Bergmann <arnd@arndb.de>
Date:   Tue, 21 Apr 2020 09:35:05 +0200
Message-ID: <CAK8P3a3iHf3TAZtiBH4L23LqyfVUmTpKk339Wtv+d1FcuYRD6A@mail.gmail.com>
Subject: Re: [PATCH v7 0/7] powerpc: switch VDSO to C implementation
To:     Christophe Leroy <christophe.leroy@c-s.fr>
Cc:     Benjamin Herrenschmidt <benh@kernel.crashing.org>,
        Paul Mackerras <paulus@samba.org>,
        Michael Ellerman <mpe@ellerman.id.au>,
        Nathan Lynch <nathanl@linux.ibm.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        linuxppc-dev <linuxppc-dev@lists.ozlabs.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Vincenzo Frascino <vincenzo.frascino@arm.com>,
        Andy Lutomirski <luto@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 8BIT
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Apr 21, 2020 at 8:56 AM Christophe Leroy
<christophe.leroy@c-s.fr> wrote:
> Le 20/04/2020 à 21:57, Arnd Bergmann a écrit :
>
> Yes that was one of the objective, be able to add that as a second step.
> First objective was adding the missing clocks and increasing
> maintainability.
>
> I can add a patch for that now, it looks easy, but how do I test it ?
> vdsotest from Nathan doesn't seem to test that.

I wasn't aware of vdsotest, but I suppose it should be fixed in there
as well. It's only one extra call that is supposed to be available on
all 32-bit architectures.

The other thing to try is musl libc 1.2, or any distro based on it.
https://distfiles.adelielinux.org/adelie/1.0/iso/rc1/ has a ppc32
image file.

I think this is all you should need to do to enable vdso in musl:

--- a/arch/powerpc/syscall_arch.h
+++ b/arch/powerpc/syscall_arch.h
@@ -92,3 +92,9 @@ static inline long __syscall6(long n, long a, long
b, long c, long d, long e, lo

 #define SO_RCVTIMEO_OLD  18
 #define SO_SNDTIMEO_OLD  19
+
+#define VDSO_USEFUL
+#define VDSO_CGT32_SYM "__vdso_clock_gettime"
+#define VDSO_CGT32_VER "LINUX_2.6"
+#define VDSO_CGT_SYM "__vdso_clock_gettime64"
+#define VDSO_CGT_VER "LINUX_2.6"

At the moment, the vdso is just disabled for powerpc because there is no
__vdso_clock_gettime64 definition in the kernel.

> > Without it, any 32-bit user space has to go through the system call
> > for time()/getttimeofday()/clock_gettime() when built with a
> > modern libc.
> >
>
> How modern ? I have glibc 2.28, the only symbols it seems to know are
> (extracted with 'strings') :
>
> __vdso_clock_gettime
> __vdso_time
> __vdso_get_tbfreq
> __vdso_getcpu
> __vdso_clock_getres
> __vdso_sigtramp32
> __vdso_sigtramp_rt32
> __vdso_gettimeofday

Mainline glibc still doesn't have it, I don't know when it will,
presumably either
the coming 2.32 release, or 2.33. There is a tree at
https://github.com/lmajewski/y2038_glibc/commits/y2038_edge that has the
necessary changes, and it is slowly getting merged upstream.

       Arnd