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

[Halil Pasic]

On Wed, 20 Feb 2019 10:27:31 +0100
Cornelia Huck <cohuck@xxxxxxxxxx> wrote:

> On Tue, 19 Feb 2019 22:31:17 +0100
> Pierre Morel <pmorel@xxxxxxxxxxxxx> 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@xxxxxxxxxxxxx>
> > >> Reported-by: Christian Borntraeger <borntraeger@xxxxxxxxxx>
> > >> Signed-off-by: Pierre Morel <pmorel@xxxxxxxxxxxxx>
> > >> ---
> > >> Â drivers/s390/crypto/vfio_ap_drv.cÂÂÂÂ | 48 
> > >> +++++++++++++++++++++++++++++------
> > >> Â drivers/s390/crypto/vfio_ap_ops.cÂÂÂÂ |Â 4 +--
> > >> Â drivers/s390/crypto/vfio_ap_private.h |Â 1 +
> > >> Â 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");
> > >> Â static struct ap_driver vfio_ap_drv;
> > >> -static struct device_type vfio_ap_dev_type = {
> > >> -ÂÂÂ .name = VFIO_AP_DEV_TYPE_NAME,
> > >> -};
> > >> -
> > >> Â struct ap_matrix_dev *matrix_dev;
> > >> Â /* Only type 10 adapters (CEX4 and later) are supported
> > >> @@ -62,6 +58,27 @@ static void vfio_ap_matrix_dev_release(struct 
> > >> device *dev)
> > >> ÂÂÂÂÂ kfree(matrix_dev);
> > >> Â }
> > >> +static int matrix_bus_match(struct device *dev, struct device_driver 
> > >> *drv)
> > >> +{
> > >> +ÂÂÂ 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.
> 


I agree, no need to complicate this any further.

> > 
> > >   
> > >> +}
> > >> +
> > >> +static struct bus_type matrix_bus = {
> > >> +ÂÂÂ .name = "vfio_ap",
> > >> +ÂÂÂ .match = &matrix_bus_match,
> > >> +};
> > >> +
> > >> +static int matrix_probe(struct device *dev)
> > >> +{
> > >> +ÂÂÂ return 0;
> > >> +}
> > >> +
> > >> +static struct device_driver matrix_driver = {
> > >> +ÂÂÂ .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.
> 

Nod.

> > 
> > 
> > >   
> > >> +ÂÂÂ .bus = &matrix_bus,
> > >> +ÂÂÂ .probe = matrix_probe,  
> > > 
> > > I would add:
> > >  ÂÂÂÂ.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.
> 

I see this as a reason to suppress_bind_attrs. It is much easier than to
think about what should happen when one unbinds the matrix device from
the vfio_ap driver on the vfio_ap bus. With the code as is it seems to
just keep working as if nothing happened.
And /sys/devices/vfio_ap/matrix/mdev_supported_types/ referencing the
name of the driver that is already gone sounds a bit weird.

Regards,
Halil