v4.14.9 on ARMv5: OK with gcc 4.8/5.4, NOK with gcc 7.2

v4.14.9 on ARMv5: OK with gcc 4.8/5.4, NOK with gcc 7.2

[Thomas Petazzoni]

Hello,

I'm pretty sure it has already been reported, but just in case: I'm
testing v4.14.9 on an ARMv5 platform (Kirkwood-based OpenBlocks A7).
mvebu_v5_defconfig boots fine when built with gcc 4.8 (good old
Sourcery toolchain) or gcc 5.4 (some old Linaro toolchain), but doesn't
boot at all when built with gcc 7.2 (Fedora toolchain). Even when
configured and booted with earlyprintk, the kernel doesn't say anything.

The toolchain being used is the default ARM toolchain provided by
Fedora 27. It says:

gcc version 7.2.1 20170915 (Red Hat Cross 7.2.1-1) (GCC)
GNU ld version 2.29.1-1.fc27

So it's quite bleeding edge.

Let me know if this has already been fixed, or if I should test a
patch. I have not tested 4.15-rc or next at this point, assuming that
if a fix had been made, it would have been backported to 4.14-stable.

Best regards,

Thomas Petazzoni
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

v4.14.9 on ARMv5: OK with gcc 4.8/5.4, NOK with gcc 7.2

[Russell King - ARM Linux]

On Thu, Jan 04, 2018 at 06:02:38PM +0100, Thomas Petazzoni wrote:
> Hello,
> 
> I'm pretty sure it has already been reported, but just in case: I'm
> testing v4.14.9 on an ARMv5 platform (Kirkwood-based OpenBlocks A7).
> mvebu_v5_defconfig boots fine when built with gcc 4.8 (good old
> Sourcery toolchain) or gcc 5.4 (some old Linaro toolchain), but doesn't
> boot at all when built with gcc 7.2 (Fedora toolchain). Even when
> configured and booted with earlyprintk, the kernel doesn't say anything.
> 
> The toolchain being used is the default ARM toolchain provided by
> Fedora 27. It says:
> 
> gcc version 7.2.1 20170915 (Red Hat Cross 7.2.1-1) (GCC)
> GNU ld version 2.29.1-1.fc27
> 
> So it's quite bleeding edge.
> 
> Let me know if this has already been fixed, or if I should test a
> patch. I have not tested 4.15-rc or next at this point, assuming that
> if a fix had been made, it would have been backported to 4.14-stable.

It's a known issue -

http://archive.armlinux.org.uk/lurker/message/20171018.095713.a56a717b.en.html

but so far no patches have been forthcoming for the kernel, because
(I guess) it's deemed to be a gcc bug.

It's one of those discussions that just seemed to die without any
real conclusion.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up

v4.14.9 on ARMv5: OK with gcc 4.8/5.4, NOK with gcc 7.2

[Andrea Adami]

On Thu, Jan 4, 2018 at 6:16 PM, Russell King - ARM Linux
<linux@armlinux.org.uk> wrote:
> On Thu, Jan 04, 2018 at 06:02:38PM +0100, Thomas Petazzoni wrote:
>> Hello,
>>
>> I'm pretty sure it has already been reported, but just in case: I'm
>> testing v4.14.9 on an ARMv5 platform (Kirkwood-based OpenBlocks A7).
>> mvebu_v5_defconfig boots fine when built with gcc 4.8 (good old
>> Sourcery toolchain) or gcc 5.4 (some old Linaro toolchain), but doesn't
>> boot at all when built with gcc 7.2 (Fedora toolchain). Even when
>> configured and booted with earlyprintk, the kernel doesn't say anything.
>>
>> The toolchain being used is the default ARM toolchain provided by
>> Fedora 27. It says:
>>
>> gcc version 7.2.1 20170915 (Red Hat Cross 7.2.1-1) (GCC)
>> GNU ld version 2.29.1-1.fc27
>>
>> So it's quite bleeding edge.
>>
>> Let me know if this has already been fixed, or if I should test a
>> patch. I have not tested 4.15-rc or next at this point, assuming that
>> if a fix had been made, it would have been backported to 4.14-stable.
>
> It's a known issue -
>
> http://archive.armlinux.org.uk/lurker/message/20171018.095713.a56a717b.en.html
>
> but so far no patches have been forthcoming for the kernel, because
> (I guess) it's deemed to be a gcc bug.
>
> It's one of those discussions that just seemed to die without any
> real conclusion.
>
> --
> RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
> FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
> According to speedtest.net: 8.21Mbps down 510kbps up
>
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel


I am happy to report the gcc bug has been fixed in trunk.
I am as well dubious where they say this should also fix gcc6 where we
have no known issues with gcc6.

"
This should also fix
incorrect generation of ldrd/strd with unaligned accesses that could
previously have occurred on ARMv5e where all such operations must be
64-bit aligned."

https://github.com/gcc-mirror/gcc/commit/f59996b56aaa1c1d62a16cbb4010775b624cbde0

In OpenEmbedded we backported the patch a while ago, see the
discussion about qemu booting where real hw doesn't:
http://lists.openembedded.org/pipermail/openembedded-core/2017-November/144045.html


Regards
Andrea

v4.14.9 on ARMv5: OK with gcc 4.8/5.4, NOK with gcc 7.2

[Russell King - ARM Linux]

On Thu, Jan 04, 2018 at 07:19:45PM +0100, Andrea Adami wrote:
> On Thu, Jan 4, 2018 at 6:16 PM, Russell King - ARM Linux
> <linux@armlinux.org.uk> wrote:
> > On Thu, Jan 04, 2018 at 06:02:38PM +0100, Thomas Petazzoni wrote:
> >> Hello,
> >>
> >> I'm pretty sure it has already been reported, but just in case: I'm
> >> testing v4.14.9 on an ARMv5 platform (Kirkwood-based OpenBlocks A7).
> >> mvebu_v5_defconfig boots fine when built with gcc 4.8 (good old
> >> Sourcery toolchain) or gcc 5.4 (some old Linaro toolchain), but doesn't
> >> boot at all when built with gcc 7.2 (Fedora toolchain). Even when
> >> configured and booted with earlyprintk, the kernel doesn't say anything.
> >>
> >> The toolchain being used is the default ARM toolchain provided by
> >> Fedora 27. It says:
> >>
> >> gcc version 7.2.1 20170915 (Red Hat Cross 7.2.1-1) (GCC)
> >> GNU ld version 2.29.1-1.fc27
> >>
> >> So it's quite bleeding edge.
> >>
> >> Let me know if this has already been fixed, or if I should test a
> >> patch. I have not tested 4.15-rc or next at this point, assuming that
> >> if a fix had been made, it would have been backported to 4.14-stable.
> >
> > It's a known issue -
> >
> > http://archive.armlinux.org.uk/lurker/message/20171018.095713.a56a717b.en.html
> >
> > but so far no patches have been forthcoming for the kernel, because
> > (I guess) it's deemed to be a gcc bug.
> >
> > It's one of those discussions that just seemed to die without any
> > real conclusion.
> >
> > --
> > RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
> > FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
> > According to speedtest.net: 8.21Mbps down 510kbps up
> >
> > _______________________________________________
> > linux-arm-kernel mailing list
> > linux-arm-kernel at lists.infradead.org
> > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
> 
> 
> I am happy to report the gcc bug has been fixed in trunk.
> I am as well dubious where they say this should also fix gcc6 where we
> have no known issues with gcc6.
> 
> "
> This should also fix
> incorrect generation of ldrd/strd with unaligned accesses that could
> previously have occurred on ARMv5e where all such operations must be
> 64-bit aligned."
> 
> https://github.com/gcc-mirror/gcc/commit/f59996b56aaa1c1d62a16cbb4010775b624cbde0
> 
> In OpenEmbedded we backported the patch a while ago, see the
> discussion about qemu booting where real hw doesn't:
> http://lists.openembedded.org/pipermail/openembedded-core/2017-November/144045.html

Thanks for reporting the update Andrea!

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up

v4.14.9 on ARMv5: OK with gcc 4.8/5.4, NOK with gcc 7.2

[Thomas Petazzoni]

Hello,

On Thu, 4 Jan 2018 22:47:41 +0000, Russell King - ARM Linux wrote:

> > I am happy to report the gcc bug has been fixed in trunk.
> > I am as well dubious where they say this should also fix gcc6 where we
> > have no known issues with gcc6.
> > 
> > "
> > This should also fix
> > incorrect generation of ldrd/strd with unaligned accesses that could
> > previously have occurred on ARMv5e where all such operations must be
> > 64-bit aligned."
> > 
> > https://github.com/gcc-mirror/gcc/commit/f59996b56aaa1c1d62a16cbb4010775b624cbde0
> > 
> > In OpenEmbedded we backported the patch a while ago, see the
> > discussion about qemu booting where real hw doesn't:
> > http://lists.openembedded.org/pipermail/openembedded-core/2017-November/144045.html  
> 
> Thanks for reporting the update Andrea!

Thanks Russell and Andrea for the feedback. Good to know that the
problem was already known and even fixed. I'll wait for the gcc 7.x fix
to land in my distro toolchain, and will keep using an older toolchain
until then.

Thanks!

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com