Re: Compiling the s390-ccw bios with clang (was: Re: s390-ccw: warning: writing 1 byte into a region of size 0)

[Daniel P . Berrangé]

On Fri, Apr 23, 2021 at 10:52:35AM +0200, Thomas Huth wrote:
> On 23/04/2021 10.47, Daniel P. Berrangé wrote:
> > On Fri, Apr 23, 2021 at 10:22:28AM +0200, Thomas Huth wrote:
> > > On 23/04/2021 10.07, Cornelia Huck wrote:
> > > > On Fri, 23 Apr 2021 09:57:08 +0200
> > > > Thomas Huth <thuth@redhat.com> wrote:
> > > > 
> > > > > On 23/04/2021 08.52, Christian Borntraeger wrote:
> > > > > [...]
> > > > > > We can of course discuss if we compile the BIOS for z10 instead of 
> > > > > > z900. TCG
> > > > > > in the mean time can handle up to z13 and z10 is now also 13 years 
> > > > > > old.
> > > > > 
> > > > > I'd really like to see us supporting Clang in the s390-ccw bios, too, 
> > > > > since
> > > > > it provides additional useful compiler warnings ... but switching the 
> > > > > -mz900
> > > > > to -mz10 here also means that we could not boot VMs anymore that use 
> > > > > a CPU
> > > > > that is older than the z10...
> > > > 
> > > > We could still boot a kernel/initrd directly, couldn't we?
> > > 
> > > Yes, but that will certainly require some documentation effort to make it
> > > clear to the users that they need to use "-kernel" in case they want to 
> > > run
> > > an older guest...
> > > 
> > > > > Is anybody still using such old CPUs? Should we maybe deprecate all 
> > > > > CPUs
> > > > > that are older than the z10 in QEMU? Alternatively, we could try to 
> > > > > detect
> > > > > Clang in the Makefile, and only use -mz10 in that case and continue 
> > > > > to use
> > > > > -mz900 in the other case...?
> > > > 
> > > > So, the issue with clang is that it compiles to at least a z10, right?
> > > 
> > > Right, Clang does not support anything that is older than a z10.
> > 
> > IIUC, according to wikipedia
> > 
> >     - z10 series was introduced in 2008
> >     - z900 series was introduced in 2000
> > 
> > Even the z10 is well older than the oldest OS platform we support.
> > 
> > Though I presume people keep mainframes deployed for longer than commodity
> > x86 hardware, it doesn't seem too unreasonable to say z10 is the oldest
> > we'll support.
>
> Well, we're talking about *guest* support here. So with that argumentation,
> we could also remove support for old CPUs like "486" or "SandyBridge" from
> qemu-system-x86_64 ... should we maybe also start the deprecation there?

Oh true, I forgot it would have implications for usability of the -cpu
flag choices, I was only thinking about the guest OS, as you can sitll
use a guest OS compiled for z900 even if the BIOS is targetting z10.

So need to compile BIOS with z900 is effectively equivalent to the need
to compile x86 BIOS for oldest generation CPU.

Downstream distros have the choice to limit what CPUs they wish to
support, but upstream if QEMU targets legacy emulation use cases, then
we want to support all CPUs.

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|