Gustavo Luiz Duarte <gustavold(a)linux.vnet.ibm.com> wrote on 01/20/2016
08:51:06 AM:
From: Gustavo Luiz Duarte <gustavold(a)linux.vnet.ibm.com>
To: Florian Weimer <fweimer(a)redhat.com>
Cc: ppc(a)lists.fedoraproject.org, Dan Horák <dan(a)danny.cz>,
carlos(a)redhat.com
Date: 01/20/2016 08:57 AM
Subject: Re: revisiting the architecture support on Power
On 01/20/2016 10:13 AM, Florian Weimer wrote:
> On 01/20/2016 12:16 PM, Dan Horák wrote:
>
> > The Fedora infrastructure is currently using Power8 HW for the main
Koji
> > build system [1] to produce the binary rpms for the
distrobution and
we
> > have planned to reuse the previous Power7+ builders for
COPR [2] and
> > Fedora Cloud.
> >
> > Carlos, Florian, could you please comment on the situation?
>
> Dan,
>
> I think there is a fundamental misunderstanding here. As long as their
> is active upstream support, the glibc team will provide whatever
> architectures you request, and make a best-effort attempt at isolating
> issues. (Personally, I lack ppc64 assembly knowledge, so it's rare
that
> I can fix architecture-specific bugs.)
>
> We declined to fix bug 1295797 because the Fedora POWER wiki page
>
> <
https://fedoraproject.org/wiki/Architectures/
PowerPC#Supported_Architectures>
>
> clearly says that on ppc64le, the minimum architecture requirement is
> POWER8. If this is not what you want, please update the wiki page,
make
> sure that both gcc, redhat-rpm-config, and binutils have been
configured
> in the appropriate way, and rebuild glibc to see if the
problematic
> instructions are still there. If you have trouble getting the
toolchain
> side sorted out, please let me know. Bug 1295797 appears to be
a
> toolchain configuration issue because the offending file is not
compiled
> with -mcpu=power8 or anything like that (assuming that the
debugging
> information can be trusted, which says that it's not the POWER8 memcpy
> which triggers the crash, but the C implementation).
>
> But the key task for you is that you need to decide what you want and
> document that. On the glibc side, we do not have a strong preference
> for POWER8. POWER7 should just work fine for ppc64le with a
> properly-configured toolchain. (Earlier architecture versions may be
> more difficult, I have not checked that.)
I understand that there are benefits to moving up to POWER8, and
that includes performance gain and we will be safer in a completely
supported setup (supported by IBM and upstream communities).
On the other hand, the fact that POWER7 is able to run little endian
has helped Fedora a lot (we built the whole ppc64le distro solely
based on POWER7). Moving to POWER8 as a requirement now would have a
great impact on those people who are in the same situation we were
not long ago, relying on POWER7 hardware to run little endian.
I vote for updating the wiki and keep the POWER7 support on ppc64le
for now. We can always re-evaluate this decision in the future if we
like.
The wiki page is correct. IBM will not be supporting LE on the P7+
architecture.
There are no customers using that setup, it requires a special version of
FW in
order to run as well. So from an IBM point of view LE is support on Power
8 and
newer systems only.
As far as I know Power7 LE is only in the labs for build machines etc. I
will work
with the Fedora release and infrastructure folks to see what P8 only will
mean to them.
thanks
Mike Wolf
[]'s
Gustavo
>
> Florian
> _______________________________________________
> ppc mailing list
> ppc(a)lists.fedoraproject.org
>
http://lists.fedoraproject.org/admin/lists/ppc@lists.fedoraproject.org
>
_______________________________________________
ppc mailing list
ppc(a)lists.fedoraproject.org
http://lists.fedoraproject.org/admin/lists/ppc@lists.fedoraproject.org