[Rpm-maint] Followup on Multiple ABI architectures

Panu Matilainen pmatilai at laiskiainen.org
Thu Nov 3 14:29:30 UTC 2011


On 10/19/2011 08:17 PM, Jon Masters wrote:
> Folks,
>
> I would like to come to some kind of resolution that works for us in the
> Fedora ARM project, and continue the longer term planning for better
> multiple ABI support, which it is clear won't be happening overnight.
>
> It is my proposal that we temporarily carry the patch posted previously:
>
> http://lists.rpm.org/pipermail/rpm-maint/2011-September/003078.html
>
> Until such time as RPM has generic support for multiple ABIs. This is
> very similar to that used in SPARC and so on, and we can conditionally
> compile this away for builds that will target ARMv5, for example. I am
> told it isn't worth reworking this to use HW_CAP (the existing stuff for
> other architectures doesn't either) and that we might aswell leave the
> current approach since it will eventually all get replaced anyway.
>
> It's not my intention to shove this in now and forget about it, but we
> need something from upstream for Fedora ARM, and there was previous
> buy-in to carry this for the time being...can we do that please? :)

Oh, I meant carrying the patch in Fedora for now, just like the 
sparc-niagara patch is, conditionally applied if you need that. I dont 
see a huge value add putting it temporarily in upstream just to be able 
to say "its upstream" especially when upstream said its ok to carry it 
downstream for the time being :)

OTOH in the meantime I've played with the hwcaps approach a bit, if 
somebody can test whether the attached patch (diffed against 4.9.x since 
that's what you probably are working with, but should apply to HEAD too) 
actually works for ARM (and Sparc Niagara for that matter) that'd be 
swell. I dont have any meaningful way to test either arch.

HWCAP is not the holy grail either, this being a library we cannot rely 
on the auxiliary vector being available at the time we get to it, so it 
needs to read /proc ... which in turn might not be there. Of course in 
normal system it is, and we cache the value so if rpm is initialized 
before going into chroot (as typically is the case) then it can be 
reinitialized in the chroot without losing anything, but if 
initialization is attempted in an environment where /proc isn't mounted 
then its going to just fail :-/

	- Panu -
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: rpm-4.9.x-hwcap-0.patch
URL: <http://lists.rpm.org/pipermail/rpm-maint/attachments/20111103/86e2f734/attachment.ksh>


More information about the Rpm-maint mailing list