Re: [PATCH v4 15/15] ipu3-cio2: Add cio2-bridge to ipu3-cio2 driver

[Daniel Scally]

Morning Kieran

On 05/01/2021 06:55, Kieran Bingham wrote:
> Hi Dan,
>
> On 04/01/2021 22:02, Daniel Scally wrote:
>>>>>> On 04/01/2021 13:35, Kieran Bingham wrote:
>>>>>>>> +/*
>>>>>>>> + * Extend this array with ACPI Hardware IDs of devices known to be working
>>>>>>>> + * plus the number of link-frequencies expected by their drivers, along with
>>>>>>>> + * the frequency values in hertz. This is somewhat opportunistic way of adding
>>>>>>>> + * support for this for now in the hopes of a better source for the information
>>>>>>>> + * (possibly some encoded value in the SSDB buffer that we're unaware of)
>>>>>>>> + * becoming apparent in the future.
>>>>>>>> + *
>>>>>>>> + * Do not add an entry for a sensor that is not actually supported.
>>>>>>>> + */
>>>>>>>> +static const struct cio2_sensor_config cio2_supported_sensors[] = {
>>>>>>>> +	CIO2_SENSOR_CONFIG("INT33BE", 0),
>>>>>>>> +	CIO2_SENSOR_CONFIG("OVTI2680", 0),
>>>>>>> I don't know if these are expressed anywhere else but would it be
>>>>>>> helpful to add a comment, or indicator as to what the actual sensor is
>>>>>>> that is represented by this HID?
>>>>>>>
>>>>>>> I can make an assumption about what an OVTI2680 might be, but the
>>>>>>> INT33BE is quite opaque. It's not clear what support that adds.
>>>>>>>
>>>>>>> Unless no one cares what the sensor is that is, but I would anticipate
>>>>>>> anyone looking here to add a new sensor might want to investigate what
>>>>>>> was already in the table?
>>>>>> Yeah good point. I'll probably alternate comment and entry then, like:
>>>>>>
>>>>>>
>>>>>> +static const struct cio2_sensor_config cio2_supported_sensors[] = {
>>>>>> +	/* Sensor OVTI5693 */
>>>>>> +	CIO2_SENSOR_CONFIG("INT33BE", 0),
>>>>>> +	/* Sensor OVTI2680 */
>>>>>> +	CIO2_SENSOR_CONFIG("OVTI2680", 0),
>>>>>>
>>>>>> As an inline comment won't fit for the sensors that we know link-frequencies for. That sound ok?
>>>>> I might put the whole vendor name in, and no need to prefix 'Sensor' IMO.
>>>>>
>>>>> +	/* Omnivision OV5693 */
>>>>> +	CIO2_SENSOR_CONFIG("INT33BE", 0),
>>>>> +	/* Omnivision OV2680 */
>>>>> +	CIO2_SENSOR_CONFIG("OVTI2680", 0),
>>>>>
>>>>> but otherwise, yes a comment the line before works for me, as you are
>>>>> right - at the end would not be practical.
>>>> Works for me
>>>>>>>> +static void cio2_bridge_create_fwnode_properties(
>>>>>>>> +	struct cio2_sensor *sensor,
>>>>>>>> +	const struct cio2_sensor_config *cfg)
>>>>>>>> +{
>>>>>>>> +	unsigned int i;
>>>>>>>> +
>>>>>>>> +	sensor->prop_names = prop_names;
>>>>>>>> +
>>>>>>>> +	for (i = 0; i < CIO2_MAX_LANES; i++)
>>>>>>>> +		sensor->data_lanes[i] = i + 1;
>>>>>>> Does something support lane swapping somewhere?
>>>>>>> I assume this is just mapping each lane directly through.
>>>>>> I think Sakari said remapping isn't supported in the CIO2 - so yeah this
>>>>>> is just mapping them directly
>>>>> So is this needed? Or is it some future compatibility thing?
>>>>>
>>>>> I haven't seen where it's used yet, but I'm not too worried about it
>>>>> though, just not sure what value an array of [1, 2, 3, 4] gives if it's
>>>>> constant...
>>>> The endpoints need to have the data-lanes property which passes an array
>>>> of data lanes, but there may well be a better way of doing this. I'm
>>>> just using the lanes member of the ssdb data structure to tell the
>>>> property how many members of the array to look at:
>>>>
>>>>
>>>> +    sensor->cio2_properties[0] = PROPERTY_ENTRY_U32_ARRAY_LEN(
>>>> +                    sensor->prop_names.data_lanes,
>>>> +                    sensor->data_lanes,
>>>> +                    sensor->ssdb.lanes);
>>>>
>>>>
>>>> So if sensor->ssdb.lanes is 2, even though it's passed a pointer to the
>>>> first member of an array of 4 members, the size calculation of that
>>>> macro limits it to just those in use. I.E. if sensor->ssdb.lanes is 2
>>>> then the property will be given the size 2 * sizeof(u32), and so when
>>>> its parsed only [1, 2] will be read.
>>>
>>> Aha, I see, ok - so we are populating an array of [1, 2, 3, 4] for each
>>> sensor that we add.
>>>
>>> What about creating the data_lanes once as a const static array and
>>> mapping to that?
>>>
>>> /*
>>>  * Map the lane arrangement, which is fixed for the IPU3.
>>>  */
>>> static const int data_lanes[CIO2_MAX_LANES] = { 1, 2, 3, 4 };
>>
>> Can't do exactly this; the bridge needs to store everything on heap
>> incase the module is unloaded, but I could move the data_lanes array to
>> the struct cio2_bridge instead of against each sensor and then we're
>> only doing it once.
> Ahh, yes I remember reading about that already.
>
> It maybe worth adding a comment about that in this file, to prevent
> other people from 'optimising' things out in 5 years ...
>
> It probably doesn't make much difference in that case if it's per sensor
> or per bridge. But indeed at least in the bridge it's only created once.
Yep ok; I moved it there and I'll add a comment explaining why it's done
a bit weird.
> --
> Kieran
>
>
>>> ...
>>>
>>>    sensor->cio2_properties[0] = PROPERTY_ENTRY_U32_ARRAY_LEN(
>>>                     sensor->prop_names.data_lanes,
>>>                     data_lanes,
>>>                     sensor->ssdb.lanes);
>>> ...
>>>
>>> Then we don't need the loop to populate the array for each sensor
>>> anymore, or the data_lanes in the sensor struct?
>>>