virtio_pci: defer kfree until release callback

commit 63bd62a0 upstream.

A struct device which has just been unregistered can live on past the
point at which a driver decides to drop it's initial reference to the
kobject gained on allocation.

This implies that when releasing a virtio device, we can't free a struct
virtio_device until the underlying struct device has been released,
which might not happen immediately on device_unregister().

Unfortunately, this is exactly what virtio pci does:
it has an empty release callback, and frees memory immediately
after unregistering the device.

This causes an easy to reproduce crash if CONFIG_DEBUG_KOBJECT_RELEASE
it enabled.

To fix, free the memory only once we know the device is gone in the release
callback.
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
[ luis: backported to 3.16:
  - file rename: virtio_pci_legacy.c -> virtio_pci.c ]
Signed-off-by: Luis Henriques <luis.henriques@canonical.com>

drivers/virtio/virtio_pci.c

@@ -668,11 +668,10 @@ static const struct virtio_config_ops virtio_pci_config_ops = {
 
 static void virtio_pci_release_dev(struct device *_d)
 {
-	/*
-	 * No need for a release method as we allocate/free
-	 * all devices together with the pci devices.
-	 * Provide an empty one to avoid getting a warning from core.
-	 */
+	struct virtio_device *vdev = dev_to_virtio(_d);
+	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
+
+	kfree(vp_dev);
 }
 
 /* the PCI probing function */
@@ -760,7 +759,6 @@ static void virtio_pci_remove(struct pci_dev *pci_dev)
 	pci_iounmap(pci_dev, vp_dev->ioaddr);
 	pci_release_regions(pci_dev);
 	pci_disable_device(pci_dev);
-	kfree(vp_dev);
 }
 
 #ifdef CONFIG_PM_SLEEP