Re: [PATCH v12 2/3] hwmon: ltc4283: Add support for the LTC4283 Swap Controller

On 4/30/26 09:50, Nuno Sá wrote:
> On Thu, Apr 30, 2026 at 07:52:55AM -0700, Guenter Roeck wrote:
>> On 4/30/26 03:03, [email protected] wrote:
>> ...
>>>> +	id = (client->adapter->nr << 10) | client->addr;
>>>> +	adev = __devm_auxiliary_device_create(dev, KBUILD_MODNAME, "gpio",
>>>> +					      &st->gpio_mask, id);
>>>> +	if (!adev)
>>>> +		return dev_err_probe(dev, -ENODEV, "Failed to add GPIO device\n");
>>>
>>> Can this error path lead to a use-after-free?
>>>
>>> According to the I2C subsystem rules, debugfs entries attached to the
>>> client debugfs directory are only cleaned up during the client device
>>> removal function on driver unbind.
>>>
>>> If __devm_auxiliary_device_create() fails, the probe function aborts and
>>> the devres subsystem frees the st pointer. However, the debugfs files
>>> persist and retain operations that operate on the freed st pointer.
>>>
>>
>> AI taking everything literally :-(. I'll have to reword that guidance again
>> to include the probe error path.
>>
> 
> Yeah and it is annoying it always comes up with something that could
> have been said before. Even more annoying some of are legit issues :)
> 

Agreed, but I think it it is better that it finds the issues in multiple
rounds than not finding them at all. And I have to say it is very human -
the same happens to me as well when I do code reviews of complex patches.
A major difference is that, unlike me, AI doesn't get tired.

Guenter