Re: [PATCH 01/12] misc: add driver for sequencer serial port

[Grant Likely <grant.likely-s3s/WqlpOiPyB63q8FvJNQ@public.gmane.org>]

On Mon, Oct 18, 2010 at 7:56 AM, Cyril Chemparathy <cyril-l0cyMroinI0@public.gmane.org> wrote:
> Hi Grant,
>
> Thanks for the feedback, I will send out an updated v2 with the proposed
> changes.
>
> [...]
>>> +EXPORT_SYMBOL(ti_ssp_open);
>>
>> I'm not thrilled with the ti_ssp_open()/ti_ssp_close() usage model.
>> It appears that the usage model is the board code registers an pile
>> of platform_devices, one for the ssp, and one for each of the
>> behaviours on top of it.  Then the various driver instances have to
>> figure out how to find each other and whether or not they are related.
>> Am I correct?
>>
>> Rather than doing an end-run around the Linux driver model, I strongly
>> recommend using the model to solve your problem.  Register only the
>> ssp platform_device in the board support code and pass it data about
>> the intended behaviour (via platform data).  When it gets probed, it
>> can then register additional platform devices which will get probed by
>> the requested driver.  That way the specific ssp device instance data
>> can be passed reliably to the spi/i2c/whatever driver without any
>> ambiguity, without any uncertainty about whether a port is 'busy', and
>> without the need of these open/close routines.
>>
>> (Hint: The trick is to set the platform_device's pdev->dev.parent
>> pointer to make use of the Linux device model's hierarchy).
>
> I am not very thrilled with this approach either :-)  I had looked at a
> few other instances where something similar was being done (spi, i2c,
> mdio, etc.), but felt that defining a new bus type would be an overkill
> for this.

Nope!  you don't need to define a new bus type.  Just continue to use
platform bus.  If you set the dev->parent pointer before registering
the device then the device model with have the appropriate hierarchy.

> For reference, could you please point me to some other place where
> something similar is being done?

Most of the multifunction devices operating in this way (though those
implementations are probably more convoluted than they need to be.
Take a look.  If you still have trouble with it let me know and I'll
draft an example for you.

g.