Re: [PATCH v2 1/1] s390: vfio_ap: link the vfio_ap devices to the vfio_ap bus subsystem

[Cornelia Huck <cohuck@redhat.com>]

On Tue, 19 Feb 2019 22:31:17 +0100
Pierre Morel <pmorel@linux.ibm.com> wrote:

> On 19/02/2019 19:52, Tony Krowiak wrote:
> > On 2/18/19 1:08 PM, Pierre Morel wrote:  
> >> Libudev relies on having a subsystem link for non-root devices. To
> >> avoid libudev (and potentially other userspace tools) choking on the
> >> matrix device let us introduce a vfio_ap bus and with that the vfio_ap
> >> bus subsytem, and make the matrix device reside within it.
> >>
> >> Doing this we need to suppress the forced link from the matrix device to
> >> the vfio_ap driver and we suppress the device_type we do not need
> >> anymore.
> >>
> >> Since the associated matrix driver is not the vfio_ap driver any more,
> >> we have to change the search for the devices on the vfio_ap driver in
> >> the function vfio_ap_verify_queue_reserved.
> >>
> >> Reported-by: Marc Hartmayer <mhartmay@linux.ibm.com>
> >> Reported-by: Christian Borntraeger <borntraeger@de.ibm.com>
> >> Signed-off-by: Pierre Morel <pmorel@linux.ibm.com>
> >> ---
> >>   drivers/s390/crypto/vfio_ap_drv.c     | 48 
> >> +++++++++++++++++++++++++++++------
> >>   drivers/s390/crypto/vfio_ap_ops.c     |  4 +--
> >>   drivers/s390/crypto/vfio_ap_private.h |  1 +
> >>   3 files changed, 43 insertions(+), 10 deletions(-)
> >>
> >> diff --git a/drivers/s390/crypto/vfio_ap_drv.c 
> >> b/drivers/s390/crypto/vfio_ap_drv.c
> >> index 31c6c84..8e45559 100644
> >> --- a/drivers/s390/crypto/vfio_ap_drv.c
> >> +++ b/drivers/s390/crypto/vfio_ap_drv.c
> >> @@ -24,10 +24,6 @@ MODULE_LICENSE("GPL v2");
> >>   static struct ap_driver vfio_ap_drv;
> >> -static struct device_type vfio_ap_dev_type = {
> >> -    .name = VFIO_AP_DEV_TYPE_NAME,
> >> -};
> >> -
> >>   struct ap_matrix_dev *matrix_dev;
> >>   /* Only type 10 adapters (CEX4 and later) are supported
> >> @@ -62,6 +58,27 @@ static void vfio_ap_matrix_dev_release(struct 
> >> device *dev)
> >>       kfree(matrix_dev);
> >>   }
> >> +static int matrix_bus_match(struct device *dev, struct device_driver 
> >> *drv)
> >> +{
> >> +    return 1;  
> > 
> > I think we should verify the following:
> > 
> > * dev == matrix_dev->device
> > * drv == &matrix_driver
> > 
> > The model employed is for the matrix device to be a singleton, so I
> > think we should verify that the matrix device and driver defined herein
> > ought to be the only possible choices for a match. Of course, doing so
> > will require some restructuring of this patch.  
> 
> I think Conny already answered this question.

Not quite :), but I don't think we need any magic in there, as there's
only one device and only one driver on that bus. No need to make this
more complicated.

> 
> >   
> >> +}
> >> +
> >> +static struct bus_type matrix_bus = {
> >> +    .name = "vfio_ap",
> >> +    .match = &matrix_bus_match,
> >> +};
> >> +
> >> +static int matrix_probe(struct device *dev)
> >> +{
> >> +    return 0;
> >> +}
> >> +
> >> +static struct device_driver matrix_driver = {
> >> +    .name = "vfio_ap",  
> > 
> > This is the same name used for the original device driver. I think
> > this driver ought to have a different name to avoid confusion.
> > How about vfio_ap_matrix or some other name to differentiate the
> > two.  
> 
> I would like too, but changing this will change the path to the mediated 
> device supported type.

Yes, we don't want to change that.

> 
> 
> >   
> >> +    .bus = &matrix_bus,
> >> +    .probe = matrix_probe,  
> > 
> > I would add:
> >      .suppress_bind_attrs = true;
> > 
> > This will remove the sysfs bind/unbind interfaces. Since there is only
> > one matrix device and it's lifecycle is controlled herein, there is no
> > sense in allowing a root user to bind/unbind it.
> >   
> 
> OTOH bind/unbind has no impact.
> If no one else ask for this I will not change what has already been 
> reviewed by Conny and Christian.

As we only have one driver, it does not really make sense anyway.