Re: [PATCH RFC 1/3] iio: Add symlink to triggers in the device's trigger folder

[Lars-Peter Clausen <lars@metafoo.de>]

On 05/13/2015 08:05 PM, Lars-Peter Clausen wrote:
> On 05/13/2015 08:03 PM, Robert Dolca wrote:
>> On Wed, May 13, 2015 at 8:42 PM, Jonathan Cameron <jic23@kernel.org> wrote:
>>>
>>> On 13/05/15 08:28, Lars-Peter Clausen wrote:
>>>> On 05/12/2015 09:06 PM, Jonathan Cameron wrote:
>>>>> On 12/05/15 17:56, Lars-Peter Clausen wrote:
>>>>>> On 05/08/2015 05:11 PM, Jonathan Cameron wrote:
>>>>>>> On 16/04/15 05:01, Robert Dolca wrote:
>>>>>>>> This patch adds a new function called iio_trigger_register_with_dev
>>>>>>>> which is a wrapper for iio_trigger_register. Besides the iio_trigger
>>>>>>>> struct this function requires iio_dev struct. It adds the trigger in
>>>>>>>> the device's trigger list and saves a reference to the device in the
>>>>>>>> trigger's struct.
>>>>>>>>
>>>>>>>> When the device is registered, in the trigger folder of the device
>>>>>>>> (where current_trigger file resides) a symlink is being created for
>>>>>>>> each trigger that was registered width iio_trigger_register_with_dev.
>>>>>>>>
>>>>>>>>     # ls -l /sys/bus/iio/devices/iio:device0/trigger/
>>>>>>>> total 0
>>>>>>>> -rw-r--r--    1 root     root          4096 Apr 16 08:33
>>>>>>>> current_trigger
>>>>>>>> lrwxrwxrwx    1 root     root             0 Apr 16 08:33 trigger0 ->
>>>>>>>> ../../trigg
>>>>>>>> er0
>>>>>>>>
>>>>>>>> This should be used for device specific triggers. Doing this the
>>>>>>>> user space
>>>>>>>> applications can figure out what if the trigger registered by a
>>>>>>>> specific device
>>>>>>>> and what should they write in the current_trigger file. Currently some
>>>>>>>> applications rely on the trigger name and this does not always work.
>>>>>>>>
>>>>>>>> This implementation assumes that the trigger is registered before
>>>>>>>> the device is
>>>>>>>> registered. If the order is not this the symlink will not be created
>>>>>>>> but
>>>>>>>> everything else will work as before.
>>>>>>>>
>>>>>>>> Signed-off-by: Robert Dolca <robert.dolca@intel.com>
>>>>>>> I was rather hoping we'd get a few more comments on this.
>>>>>>> In principle I like the idea, but it's new ABI and does make life
>>>>>>> a tiny bit more complex, so what do people think?
>>>>>>>
>>>>>>> Few trivial code comments inline.
>>>>>>
>>>>>> I don't think it adds more information. Both the trigger and the
>>>>>> device get registered for the same parent device, so you can already
>>>>>> easily find the trigger for a device by going to the parent device
>>>>>> and taking a look at the triggers registered by the parent device.
>>>>> I had the same thought. The question is whether the slightly gain in
>>>>> simplicity for userspace is worth it...  I'm undecided at the moment.
>>>>
>>>> As you may have guessed by now I'm always quite conservative when it
>>>> comes to introducing new ABI. Simply because we have to maintain it
>>>> forever, the less stuff to maintain forever the better.
>>>>
>>>> Hence I think all new ABI needs a compelling reason, e.g. like a
>>>> improvement in performance. And of course this patch slightly
>>>> simplifies things, but in my opinion not enough to justify a ABI
>>>> extension. We can always find ways to simplify the interface, but the
>>>> metric for ABI should be whether the simplification actually matters.
>>>> In this case I don't think it does, finding the trigger for a device
>>>> is not really hot-path. The amount of time saved will be disappear in
>>>> the noise.
>>>>
>>>> And in my opinion applications shouldn't directly use the low-level
>>>> ABI but rather use middle-ware libraries/frameworks, like e.g.
>>>> libiio, and that's where you'd hide the complexities of a operation.
>>>>
>>>> - Lars
>>> I'll go with Lars response on this one.  Not worth the hassle.
>>> That's the nature of an RFC of course!
>>>
>>> Jonathan
>>
>> Would it be acceptable to add the symlinks without adding a new API?
>> When a trigger is registered you could use the common parent to get a
>> pointer to the iio_dev and then create the symlink. This is a little
>> bit complicated but I think it can be done.
>
> The concerns are with the symlink and with he symlink only. Adding new API
> inside the kernel is generally not as much of a problem as external ABI.

Sorry, I should have added you can easily find out which triggers and which 
devices belong together by basically doing a dirname `readlink 
/sys/bus/iio/devices/X`. Those that are at the same place in the global 
hierarchy belong to the same device.