Re: [PATCH 3/3] i2c: Add Intel USBIO I2C driver

[Hans de Goede <hansg@kernel.org>]

Hi Sakari,

Thank you for your review, note this
is not a full reply.

On 11-Aug-25 9:16 AM, Sakari Ailus wrote:

<snip>

>> +static int usbio_i2c_probe(struct auxiliary_device *adev,
>> +		const struct auxiliary_device_id *adev_id)
>> +{
>> +	struct usbio_i2c_bus_desc *i2c_desc;
>> +	struct device *dev = &adev->dev;
>> +	u8 dummy_read_buf;
>> +	struct i2c_msg dummy_read = {
>> +		.addr = 0x08,
>> +		.flags = I2C_M_RD,
>> +		.len = 1,
>> +		.buf = &dummy_read_buf,
>> +	};
>> +	struct usbio_i2c *i2c;
>> +	u32 max_speed;
>> +	int ret;
>> +
>> +	i2c_desc = dev_get_platdata(dev);
>> +	if (!i2c_desc)
>> +		return -EINVAL;
>> +
>> +	/* Some USBIO chips have caps set to 0, but all chips can do 400KHz */
>> +	if (!i2c_desc->caps)
>> +		max_speed = I2C_MAX_FAST_MODE_FREQ;
>> +	else
>> +		max_speed = usbio_i2c_speeds[i2c_desc->caps & USBIO_I2C_BUS_MODE_CAP_MASK];
>> +
>> +	i2c = devm_kzalloc(dev, sizeof(*i2c), GFP_KERNEL);
>> +	if (!i2c)
>> +		return -ENOMEM;
> 
> Same comment on devm memory allocation than on the GPIO driver: I think you
> need to use the release callback of struct device here.

And more or less the same reply, i2c_del_adapter() ensures
all i2c_clients on the adapters bus are unregistered so
it guarantees that after i2c_del_adapter() none of the adapter
functions will get called.

So freeing the struct containing the adapter after remove()
has run is fine.

I know that the media subsystem does not handle v4l2-subdevs
(which the i2c-clients are) going away very well.

Richard mentioned that after a fw-update the usbio chip will
not come back until a reboot. And I've noticed that after crashing
the usbio fw it will not come back until a full power-cycle.

So we do not need to worry about somehow slotting new i2c-clients
into the media-controller graph after a disconnect + reconnect
since the reconnect will never happen during the current boot.

We do need to somehow make sure that trying to access the v4l2-subdev
after disconnect does not cause oopses or worse.

We will likely need to somehow keep the memory for the v4l2-subdev
around and add something like a disconnected / dead flag to it.

Regards,

Hans