Re: [PATCH v2 0/3] driver core: fix unbalanced reference counting for software nodes

[D . Manresa]

Tested your v2 on a Surface Pro 7+, where ipu-bridge attaches the IPU's
software node with set_secondary_fwnode() after device_add(), which is
the case the series is about. Reading the node's kobject reference count
directly across a PCI remove of 0000:00:05.0:

  6.19.8, unpatched:  4 bound, 3 while removed, stuck at 3
  7.3 + this series:  4 bound, 4 while removed

The 4 is one registration reference plus three child port nodes. No
refcount_t, KASAN or other splats either way.

How much that covers: the rescan does not re-probe, because mainline
ipu-bridge only unregisters its IPU HID node on the error path of
ipu_bridge_init(), so the next software_node_register() hits -EEXIST.
Separate leak, unrelated to your patch, but it means only the first
cycle really exercises the fixed path.

Tested-by: D. Manresa <dmanresa@xxxxxxxxx>
Assisted-by: LLM