On Mittwoch, 11. Dezember 2019, 08:23:19 CET Guillaume Gardet wrote:
> 
> > -----Original Message-----
> > From: Adrian Schröter <[email protected]>
> > Sent: 11 December 2019 08:09
> > To: Andreas Färber <[email protected]>
> > Cc: Guillaume Gardet <[email protected]>; openSUSE ARM ML
> > <[email protected]>; Andrew Wafaa <[email protected]>;
> > Andreas Schwab <[email protected]>; Dirk Müller <[email protected]>
> > Subject: Re: [opensuse-arm] arm 32bit vs arm 32bit
> >
> > On Dienstag, 10. Dezember 2019, 18:06:04 CET Andreas Färber wrote:
> > > Hi,
> > >
> > > Am 10.12.19 um 16:21 schrieb Guillaume Gardet:
> > > >> -----Original Message-----
> > > >> From: Adrian Schröter <[email protected]>
> > > >> Sent: 07 December 2019 12:46
> > > >> To: openSUSE ARM ML ([email protected]) <opensuse-
> > > >> [email protected]>
> > > >> Subject: [opensuse-arm] arm 32bit vs arm 32bit
> > > >>
> > > >> Hi,
> > > >>
> > > >> I was going to add 32bit package configs to repackage armv7hl libs
> > > >> for aarch64 installations (for personal need, samsung binary only
> > > >> printer driver)
> > > >>
> > > >> However, I noticed that armv7hl 32bit userland would conflict with
> > > >> aarch64_ilp32 definitions.
> > > >>
> > > >> Was there already any discussion how a mixed arch installation should 
> > > >> look
> > alike?
> > > >>
> > > >> My proposal would be (for a aarch64 installation):
> > > >>
> > > >>  armv[567]* libs should be installed in /lib (and /usr/lib) via
> > > >> *-32bit*.aarch64.rpm /lib to stay compatible with armv[567] 
> > > >> installations.
> > > >>
> > > >>  aarch64_ilp32 libs should be installed in /lib-ilp32 via
> > > >> *-ilp32*.aarch64.rpm current config seems to put these in -32bit
> > > >> packages, what seems to be wrong to me.
> > > >>
> > > >>  Do we also need to take care about aarch32?
> > > >>
> > > >> any opinion about this?
> > > >
> > > > AFAIK, you cannot use armv7 libs/bins as is on arm64 systems.
> > >
> > > It depends on the CPU. ThunderX and Kunpeng 920 don't support AArch32
> > > mode, but most Cortex cores still do.
> > >
> > > I concur that the outlined ilp32-as-32bit setup sounds wrong, in
> > > particular since that is not even mainline-supported still.
> >
> > okay, so I can drop that part of the code again moving ilp32 into /lib for 
> > compat
> > packages ....
> >
> > > There were discussions or even recommendations on cross-distro or so
> > > some years back on naming/placement - Andy might remember.
> > >
> > > As for AArch32, I assume you mean armv8l? I don't think we currently
> > > build any packages for it, at least we have no separate scheduler.
> >
> > yes, question is if it may appear at some point of time and where to put 
> > libraries
> > then.
> >
> > but this is maybe to far ahead, if it is /lib as well we would anyway have 
> > the
> > conflict with armv7l. So I think, I just go forward and adapt baselibs 
> > config to
> > generate armv7l (and armv6l?) libs into /lib.
> 
> I think zypper does not allow armv7 rpms to be installed on aarch64. Which is 
> right for aarch64 SoC which do not support AArch32.

that won't be a problem for the repackaged compat lib packages, since the would 
appear as aarch64.
Similar to what we do on x86_64, eg:

  libstdc++-devel-32bit-9-1.6.x86_64.rpm

so it is a x86_64.rpm which provides the 32bit libs.

But you are right, we may want to support also to install armv7hl rpms 
afterwards for
applications.

> Not sure how we could handle this difference between SoC properly.

me neither, the bins would just not run. But that is a hardware limitation and 
an OS limitation from my POV.

-- 

Adrian Schroeter <[email protected]>
Build Infrastructure Project Manager

SUSE Software Solutions Germany GmbH,  Maxfeldstr. 5, 90409 Nuernberg, Germany
(HRB 247165, AG München), Geschäftsführer: Felix Imendörffer




--
To unsubscribe, e-mail: [email protected]
To contact the owner, e-mail: [email protected]

Reply via email to