Re: [PATCH v5 8/8] hw/vfio: all vfio files can now be common files

On Mar 18 26, Pierrick Bouvier wrote:
> -specific_ss.add_all(when: 'CONFIG_VFIO', if_true: vfio_ss)
> +system_ss.add_all(when: 'CONFIG_VFIO', if_true: vfio_ss)

Apologies for digging out this old thread, but parts of the vfio code
need linux/kvm.h, which includes asm/kvm.h. Consequently it can only be
fully architecture agnostic based on the assumption that KVM will only
ever be able to support one guest architecture which must be the same as
the host architecture.

On s390x, we are currently working on a KVM change that breaks this
assumption:
https://lore.kernel.org/all/[email protected]/

We are in the process of working out what changes this would require on
the qemu side, and will send a patch series soon for discussion, but
we've noticed that this patch series in particular was a breaking change
for us when moving to v11.1.

Fortunately, this series also extracted the architecture dependent code
in VFIO into the kvm-helpers.c file, so for us the most obvious solution
would be to keep kvm-helpers.c in specific_ss, while the rest of vfio
can go to system_ss. Would that be acceptable?

Thanks,
Andreas