Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Fri, 15 Mar 2019 16:29:35 -0700
From:   Matthias Kaehlcke <mka@chromium.org>
To:     Nick Desaulniers <ndesaulniers@google.com>
Cc:     Andy Lutomirski <luto@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        "H . Peter Anvin" <hpa@zytor.com>, x86@kernel.org,
        LKML <linux-kernel@vger.kernel.org>,
        Manoj Gupta <manojgupta@chromium.org>,
        Tiancong Wang <tcwang@chromium.org>,
        Stephen Hines <srhines@google.com>,
        clang-built-linux@googlegroups.com,
        Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>
Subject: Re: [PATCH] lib: Add shared copy of __lshrti3 from libgcc
Message-ID: <20190315232935.GD112750@google.com>
References: <CAG5bF+S6OvBnsaR6UpMCqjDR9_hMo6qRMHGiW+iCaRYQW4C3YA@mail.gmail.com>
 <CAKwvOdkfb_DT2p2vvxK2ye6JpGYB4qqCc8yECU_JEkrc+6HXDA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <CAKwvOdkfb_DT2p2vvxK2ye6JpGYB4qqCc8yECU_JEkrc+6HXDA@mail.gmail.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Mar 15, 2019 at 03:06:37PM -0700, 'Nick Desaulniers' via Clang Built Linux wrote:
> On Fri, Mar 15, 2019 at 1:54 PM Matthias Kaehlcke <mka@chromium.org> wrote:
> >
> > The compiler may emit calls to __lshrti3 from the compiler runtime
> > library, which results in undefined references:
> >
> > arch/x86/kvm/x86.o: In function `mul_u64_u64_shr':
> >   include/linux/math64.h:186: undefined reference to `__lshrti3'
> 
> Looks like Clang will emit this at -Oz (but not -O2):
> https://godbolt.org/z/w1_2YC
> 
> >
> > Add a copy of the __lshrti3 libgcc routine (from gcc v4.9.2).
> 
> Has it changed since? If so why not a newer version of libgcc_s?

Our compiler folks who maintain this gcc version dug this up for
me. In the gcc sources there is no direct implementation of __lshrti3,
I was told it is somehow derived from __lshrdi3.

> > Include the function for x86 builds with clang, which is the
> > environment where the above error was observed.
> >
> > Signed-off-by: Matthias Kaehlcke <mka@chromium.org>
> > ---
> >  arch/x86/Kconfig       |  1 +
> >  include/linux/libgcc.h | 16 ++++++++++++++++
> >  lib/Kconfig            |  3 +++
> >  lib/Makefile           |  1 +
> >  lib/lshrti3.c          | 31 +++++++++++++++++++++++++++++++
> >  5 files changed, 52 insertions(+)
> >  create mode 100644 lib/lshrti3.c
> >
> > diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> > index c1f9b3cf437c..a5e0d923845d 100644
> > --- a/arch/x86/Kconfig
> > +++ b/arch/x86/Kconfig
> > @@ -105,6 +105,7 @@ config X86
> >         select GENERIC_IRQ_PROBE
> >         select GENERIC_IRQ_RESERVATION_MODE
> >         select GENERIC_IRQ_SHOW
> > +       select GENERIC_LIB_LSHRTI3              if CC_IS_CLANG
> >         select GENERIC_PENDING_IRQ              if SMP
> >         select GENERIC_SMP_IDLE_THREAD
> >         select GENERIC_STRNCPY_FROM_USER
> > diff --git a/include/linux/libgcc.h b/include/linux/libgcc.h
> > index 32e1e0f4b2d0..a71036471838 100644
> > --- a/include/linux/libgcc.h
> > +++ b/include/linux/libgcc.h
> > @@ -22,15 +22,26 @@
> >  #include <asm/byteorder.h>
> >
> >  typedef int word_type __attribute__ ((mode (__word__)));
> > +typedef int TItype __attribute__ ((mode (TI)));
> 
> Well that's an interesting new compiler attribute.
> https://stackoverflow.com/questions/4559025/what-does-gcc-attribute-modexx-actually-do
> Please use `__mode(TI)` from include/linux/compiler_attributes.h ex.
> typedef int TItype __mode(TI);

ok

> >  #ifdef __BIG_ENDIAN
> >  struct DWstruct {
> >         int high, low;
> >  };
> > +
> > +struct DWstruct128 {
> > +       long long high, low;
> > +};
> > +
> >  #elif defined(__LITTLE_ENDIAN)
> >  struct DWstruct {
> >         int low, high;
> >  };
> > +
> > +struct DWstruct128 {
> > +       long long low, high;
> > +};
> > +
> >  #else
> >  #error I feel sick.
> >  #endif
> > @@ -40,4 +51,9 @@ typedef union {
> >         long long ll;
> >  } DWunion;
> >
> > +typedef union {
> > +       struct DWstruct128 s;
> > +       TItype ll;
> > +} DWunion128;
> > +
> >  #endif /* __ASM_LIBGCC_H */
> > diff --git a/lib/Kconfig b/lib/Kconfig
> > index a9e56539bd11..369e10259ea6 100644
> > --- a/lib/Kconfig
> > +++ b/lib/Kconfig
> > @@ -624,6 +624,9 @@ config GENERIC_LIB_ASHRDI3
> >  config GENERIC_LIB_LSHRDI3
> >         bool
> >
> > +config GENERIC_LIB_LSHRTI3
> > +       bool
> > +
> >  config GENERIC_LIB_MULDI3
> >         bool
> >
> > diff --git a/lib/Makefile b/lib/Makefile
> > index 4e066120a0d6..42648411f451 100644
> > --- a/lib/Makefile
> > +++ b/lib/Makefile
> > @@ -277,6 +277,7 @@ obj-$(CONFIG_PARMAN) += parman.o
> >  obj-$(CONFIG_GENERIC_LIB_ASHLDI3) += ashldi3.o
> >  obj-$(CONFIG_GENERIC_LIB_ASHRDI3) += ashrdi3.o
> >  obj-$(CONFIG_GENERIC_LIB_LSHRDI3) += lshrdi3.o
> > +obj-$(CONFIG_GENERIC_LIB_LSHRTI3) += lshrti3.o
> >  obj-$(CONFIG_GENERIC_LIB_MULDI3) += muldi3.o
> >  obj-$(CONFIG_GENERIC_LIB_CMPDI2) += cmpdi2.o
> >  obj-$(CONFIG_GENERIC_LIB_UCMPDI2) += ucmpdi2.o
> > diff --git a/lib/lshrti3.c b/lib/lshrti3.c
> > new file mode 100644
> > index 000000000000..2d2123bb3030
> > --- /dev/null
> > +++ b/lib/lshrti3.c
> > @@ -0,0 +1,31 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +
> > +#include <linux/export.h>
> > +#include <linux/libgcc.h>
> > +
> > +long long __lshrti3(long long u, word_type b)

This signature matches with the gcc documentation
(https://gcc.gnu.org/onlinedocs/gccint/Integer-library-routines.html),
however the gcc implementation of the function has 128-bit values as
input/output, so I guess to meet gcc's expectations the 'long long's
need to be changed to TItype.

> > +{
> > +       DWunion128 uu, w;
> > +       word_type bm;
> > +
> > +       if (b == 0)
> > +               return u;
> > +
> > +       uu.ll = u;
> > +       bm = 64 - b;
> > +
> > +       if (bm <= 0) {
> > +               w.s.high = 0;
> > +               w.s.low = (unsigned long long) uu.s.high >> -bm;
> > +       } else {
> > +               const unsigned long long carries =
> > +                       (unsigned long long) uu.s.high << bm;
> > +               w.s.high = (unsigned long long) uu.s.high >> b;
> > +               w.s.low = ((unsigned long long) uu.s.low >> b) | carries;
> > +       }
> > +
> > +       return w.ll;
> > +}
> > +#ifndef BUILD_VDSO
> > +EXPORT_SYMBOL(__lshrti3);
> > +#endif
> 
> I don't think you want this.  Maybe that was carried over from
> https://lkml.org/lkml/2019/3/15/669
> by accident?  The above linkage error mentions arch/x86/kvm/x86.o
> which I wouldn't expect to be linked into the VDSO image?

I put it preemptively, since I know from
https://lkml.org/lkml/2019/3/15/669 that the vDSO build is unhappy
about EXPORT_SYMBOL. Obviously we can also wait until a vDSO needs to
include this file.

Thanks

Matthias