On Sat, Jul 19, 2025 at 08:01:04PM +0200, Hans de Goede wrote:
Before commit df6d7277e552 ("i2c: core: Do not dereference fwnode in struct device"), i2c_unregister_device() only called fwnode_handle_put() on of_node-s in the form of calling of_node_put(client->dev.of_node).
But after this commit the i2c_client's fwnode now unconditionally gets fwnode_handle_put() on it.
When the i2c_client has no primary (ACPI / OF) fwnode but it does have a software fwnode, the software-node will be the primary node and fwnode_handle_put() will put() it.
But for the software fwnode device_remove_software_node() will also put() it leading to a double free:
[ 82.665598] ------------[ cut here ]------------ [ 82.665609] refcount_t: underflow; use-after-free. [ 82.665808] WARNING: CPU: 3 PID: 1502 at lib/refcount.c:28 refcount_warn_saturate+0xba/0x11 ... [ 82.666830] RIP: 0010:refcount_warn_saturate+0xba/0x110 ... [ 82.666962] <TASK> [ 82.666971] i2c_unregister_device+0x60/0x90
Fix this by not calling fwnode_handle_put() when the primary fwnode is a software-node.
Thanks for the fix! It doesn't look elegant, but I think it's all due to the (current) design of the fwnode linked list. Reviewed-by: Andy Shevchenko andriy.shevchenko@linux.intel.com