On 2020-03-06 14:58, Andy Shevchenko wrote:
> On Fri, Mar 06, 2020 at 12:48:14PM +0100, Peter Rosin wrote:
>> On 2020-03-06 10:54, Andy Shevchenko wrote:
>>> On Thu, Mar 05, 2020 at 09:05:56PM +0000, Peter Rosin wrote:
>>>> On 2020-03-05 16:53, Andy Shevchenko wrote:
>>>>> Device property API allows to gather device resources from different sources,
>>>>> such as ACPI. Convert the drivers to unleash the power of device property API.
>>>
>>> ...
>>>
>>>>> static const struct i2c_device_id pca954x_id[] = {
>>>>> - { "pca9540", pca_9540 },
>>>>> - { "pca9542", pca_9542 },
>>>>> - { "pca9543", pca_9543 },
>>>>> - { "pca9544", pca_9544 },
>>>>> - { "pca9545", pca_9545 },
>>>>> - { "pca9546", pca_9546 },
>>>>> - { "pca9547", pca_9547 },
>>>>> - { "pca9548", pca_9548 },
>>>>> - { "pca9846", pca_9846 },
>>>>> - { "pca9847", pca_9847 },
>>>>> - { "pca9848", pca_9848 },
>>>>> - { "pca9849", pca_9849 },
>>>>> + { "pca9540", .driver_data = (kernel_ulong_t)&chips[pca_9540] },
>>>>> + { "pca9542", .driver_data = (kernel_ulong_t)&chips[pca_9542] },
>>>>> + { "pca9543", .driver_data = (kernel_ulong_t)&chips[pca_9543] },
>>>>> + { "pca9544", .driver_data = (kernel_ulong_t)&chips[pca_9544] },
>>>>> + { "pca9545", .driver_data = (kernel_ulong_t)&chips[pca_9545] },
>>>>> + { "pca9546", .driver_data = (kernel_ulong_t)&chips[pca_9546] },
>>>>> + { "pca9547", .driver_data = (kernel_ulong_t)&chips[pca_9547] },
>>>>> + { "pca9548", .driver_data = (kernel_ulong_t)&chips[pca_9548] },
>>>>> + { "pca9846", .driver_data = (kernel_ulong_t)&chips[pca_9846] },
>>>>> + { "pca9847", .driver_data = (kernel_ulong_t)&chips[pca_9847] },
>>>>> + { "pca9848", .driver_data = (kernel_ulong_t)&chips[pca_9848] },
>>>>> + { "pca9849", .driver_data = (kernel_ulong_t)&chips[pca_9849] },
>>>>
>>>> It feels odd/wrong to specifically name .driver_data when .name is not there.
>>>> None or both...
>>>
>>> I will add .name as well.
>>>
>>>>> + data->chip = device_get_match_data(dev);
>>>>> if (!data->chip)
>>>>> data->chip = &chips[id->driver_data];
>>>>
>>>> These two lines no longer make any sence.
>>>
>>> Please elaborate.
>>>
>>> IIRC Javier explained once that I²C ID table is still good to have to allow
>>> enumeration from user space.
>>
>> id->driver_data is no longer an integer index into chips[]. So, for the I2C
>> ID table case, either device_get_match_data returns the .driver_data as-is
>> from the pca954x_id array, or it returns NULL (I don't know which it is).
>
> No, you took it wrong. device_get_match_data() operates with ACPI/DT tables.
<rant-mode>
What do you mean wrong? I said that either A or B holds but did not know which
(with these definitions):
A. device_get_match_data() digs in the i2c_device_id table and returns the
.driver_data of the matching entry.
B. device_get_match_data() behaves as of_device_get_match_data() and does not
dig in the i2c_device_id table, and therefore returns NULL when the driver
is probed that way.
And that in either of these cases your patch made no sense.
At least that was what I tried to say, using less words...
And then, according to you, B holds. So, I was right: either A or B holds. BTW,
I obviously meant the either/or construct to be in the exclusive sense where
both cannot hold (but my statement is also correct in the inclusive-or sense).
I would only have been wrong if the correct description had been some third
option, which I had not mentioned. But that was apparently not the case.
Cheers,
Peter
>
>> In the first case, if (!data->chip) ...; is useless dead code. In the latter
>> case, it should be
>>
>> if (!data->chip)
>> data->chip = (whatever-type-chip-is *)id->driver_data;
>>
>> (If it's this latter case, I get the feeling that changing the .driver_data
>> of pca954x_id is an orthogonal change that has little to do with using
>> device properties instead of of-specifics.)
>
> But this comment is good, seems I missed that and actually change is not needed
> indeed.
>
