Re: [PATCH v9 10/21] vfio/pci: introduce CONFIG_VFIO_PCI_ZDEV_KVM

[Matthew Rosato]

On 6/8/22 2:19 AM, Thomas Huth wrote:
> On 06/06/2022 22.33, Matthew Rosato wrote:
> > The current contents of vfio-pci-zdev are today only useful in a KVM
> > environment; let's tie everything currently under vfio-pci-zdev to
> > this Kconfig statement and require KVM in this case, reducing complexity
> > (e.g. symbol lookups).
> >
> > Signed-off-by: Matthew Rosato <mjrosato@xxxxxxxxxxxxx>
> > ---
> >   drivers/vfio/pci/Kconfig      | 11 +++++++++++
> >   drivers/vfio/pci/Makefile     |  2 +-
> >   include/linux/vfio_pci_core.h |  2 +-
> >   3 files changed, 13 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/vfio/pci/Kconfig b/drivers/vfio/pci/Kconfig
> > index 4da1914425e1..f9d0c908e738 100644
> > --- a/drivers/vfio/pci/Kconfig
> > +++ b/drivers/vfio/pci/Kconfig
> > @@ -44,6 +44,17 @@ config VFIO_PCI_IGD
> >         To enable Intel IGD assignment through vfio-pci, say Y.
> >   endif
> > +config VFIO_PCI_ZDEV_KVM
> > +    bool "VFIO PCI extensions for s390x KVM passthrough"
> > +    depends on S390 && KVM
> > +    default y
> > +    help
> > +      Support s390x-specific extensions to enable support for 
> > enhancements
> > +      to KVM passthrough capabilities, such as interpretive execution of
> > +      zPCI instructions.
> > +
> > +      To enable s390x KVM vfio-pci extensions, say Y.
>
> Is it still possible to disable CONFIG_VFIO_PCI_ZDEV_KVM ? Looking at 
> the later patches (e.g. 20/21 where you call kvm_s390_pci_zpci_op() from 
> kvm-s390.c), it rather seems to me that it currently cannot be disabled 
> independently (as long as KVM is enabled).

Yes, you can build with, for example, CONFIG_VFIO_PCI_ZDEV_KVM=n and 
CONFIG_KVM=m -- I tested it again just now.  The result is kvm and 
vfio-pci are built and vfio-pci works, but none of the vfio-pci-zdev 
extensions are available (including zPCI interpretation).

This is accomplished via the placement of some IS_ENABLED checks.  Some 
calls (e.g. AEN init) are fenced by 
IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM).  There are also some areas that 
are fenced off via a call to kvm_s390_pci_interp_allowed() which also 
includes an IS_ENABLED check along with checks for facility and cpu id.

Using patch 20 as an example, KVM_CAP_S390_ZPCI_OP will always be 
reported as unavailable to userspace if CONFIG_VFIO_PCI_ZDEV_KVM=n due 
to the call to kvm_s390_pci_interp_allowed().  If userspace sends us the 
ioctl anyway, we will return -EINVAL because there is again a 
IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM) check before we read the ioctl args 
from userspace.

> So if you want to make this selectable by the user, I think you have to 
> put some more #ifdefs in the following patches.
> But if this was not meant to be selectable by the user, I think it 
> should not get a help text and rather be selected by the KVM switch in 
> arch/s390/kvm/Kconfig instead of having a "default y".
>
>   Thomas