CVE-2024-56545

In the Linux kernel, the following vulnerability has been resolved:

HID: hyperv: streamline driver probe to avoid devres issues

It was found that unloading 'hid_hyperv' module results in a devres
complaint:

 ...
 hv_vmbus: unregistering driver hid_hyperv
 ------------[ cut here ]------------
 WARNING: CPU: 2 PID: 3983 at drivers/base/devres.c:691 devres_release_group+0x1f2/0x2c0
 ...
 Call Trace:
  <TASK>
  ? devres_release_group+0x1f2/0x2c0
  ? __warn+0xd1/0x1c0
  ? devres_release_group+0x1f2/0x2c0
  ? report_bug+0x32a/0x3c0
  ? handle_bug+0x53/0xa0
  ? exc_invalid_op+0x18/0x50
  ? asm_exc_invalid_op+0x1a/0x20
  ? devres_release_group+0x1f2/0x2c0
  ? devres_release_group+0x90/0x2c0
  ? rcu_is_watching+0x15/0xb0
  ? __pfx_devres_release_group+0x10/0x10
  hid_device_remove+0xf5/0x220
  device_release_driver_internal+0x371/0x540
  ? klist_put+0xf3/0x170
  bus_remove_device+0x1f1/0x3f0
  device_del+0x33f/0x8c0
  ? __pfx_device_del+0x10/0x10
  ? cleanup_srcu_struct+0x337/0x500
  hid_destroy_device+0xc8/0x130
  mousevsc_remove+0xd2/0x1d0 [hid_hyperv]
  device_release_driver_internal+0x371/0x540
  driver_detach+0xc5/0x180
  bus_remove_driver+0x11e/0x2a0
  ? __mutex_unlock_slowpath+0x160/0x5e0
  vmbus_driver_unregister+0x62/0x2b0 [hv_vmbus]
  ...

And the issue seems to be that the corresponding devres group is not
allocated. Normally, devres_open_group() is called from
__hid_device_probe() but Hyper-V HID driver overrides 'hid_dev->driver'
with 'mousevsc_hid_driver' stub and basically re-implements
__hid_device_probe() by calling hid_parse() and hid_hw_start() but not
devres_open_group(). hid_device_probe() does not call __hid_device_probe()
for it. Later, when the driver is removed, hid_device_remove() calls
devres_release_group() as it doesn't check whether hdev->driver was
initially overridden or not.

The issue seems to be related to the commit 62c68e7cee33 ("HID: ensure
timely release of driver-allocated resources") but the commit itself seems
to be correct.

Fix the issue by dropping the 'hid_dev->driver' override and using
hid_register_driver()/hid_unregister_driver() instead. Alternatively, it
would have been possible to rely on the default handling but
HID_CONNECT_DEFAULT implies HID_CONNECT_HIDRAW and it doesn't seem to work
for mousevsc as-is.