Re: [PATCH 1/4] kvm: Destroy & free KVM devices on release

[Gleb Natapov <gleb@redhat.com>]

On Wed, Oct 30, 2013 at 08:30:22AM -0600, Alex Williamson wrote:
> On Wed, 2013-10-30 at 12:40 +0200, Gleb Natapov wrote:
> > On Tue, Oct 29, 2013 at 10:13:22AM -0600, Alex Williamson wrote:
> > > The KVM device interface allocates a struct kvm_device and calls
> > > kvm_device_ops.create on it from KVM VM ioctl KVM_CREATE_DEVICE.
> > > This returns a file descriptor to the user for them to set/get/check
> > > further attributes.  On closing the file descriptor, one would assume
> > > that kvm_device_ops.destroy is called and all traces of the device
> > > would go away.  One would be wrong, it actually does nothing more
> > > than release the struct kvm reference, waiting until the VM is
> > > destroyed before doing more.  This leaves devices that only want a
> > > single instance of themselves per VM in a tough spot.
> > > 
> > This is by design. Otherwise locking will be needed on each device access
> > and for interrupt controllers this is unnecessary serialization and
> > overhead. Device API is not designed for devices that can go away while
> > machine is running anyway, so after creation device is only destroyed
> > during VM destruction.
> 
> Hmm, ok.  In that case I can drop this patch and I think the rest just
> boils down to userspace use of the device.  I had been close()'ing the
> kvm device fd when all QEMU vfio devices are detached, but I can just as
> easily leave it open in case a new device is added later.  I'll send out
> a new series after doing some more review and testing.  Do you have any
> comments on the rest of the series?  Thanks,
> 
If I understand 4/4 correctly if there is VFIO device connected we
assume non coherent domain. How hard it would be to do proper checking
in this path series?

--
			Gleb.