lkml.org 
[lkml]   [2015]   [Mar]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] staging: iio: trigger: Use standard attr for sampling frequency
On 18/03/15 17:25, Lars-Peter Clausen wrote:
> On 03/18/2015 06:21 PM, Daniel Baluta wrote:
>> On Wed, Mar 18, 2015 at 6:47 PM, Lars-Peter Clausen <lars@metafoo.de> wrote:
>>> On 03/18/2015 04:04 PM, Daniel Baluta wrote:
>>>>
>>>> On Sat, Mar 14, 2015 at 7:21 PM, Jonathan Cameron <jic23@kernel.org>
>>>> wrote:
>>>>>
>>>>> On 12/03/15 12:48, Daniel Baluta wrote:
>>>>>>
>>>>>> On Thu, Mar 12, 2015 at 10:40 AM, Lars-Peter Clausen <lars@metafoo.de>
>>>>>> wrote:
>>>>>>>
>>>>>>> On 03/12/2015 09:16 AM, Octavian Purdila wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>> On Wed, Mar 11, 2015 at 6:36 PM, Daniel Baluta
>>>>>>>> <daniel.baluta@intel.com>
>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> As written in Documentation/ABI/testing/sysfs-bus-iio the trigger
>>>>>>>>> attribute for sampling frequency should be sampling_frequency.
>>>>>>>>>
>>>>>>>>> Fix this for iio-trig-periodic-rtc module in order to prepare it
>>>>>>>>> for moving out of staging.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Daniel Baluta <daniel.baluta@intel.com>
>>>>>>>>> ---
>>>>>>>>> Jonathan, this module is very useful for devices that do not have
>>>>>>>>> an interrupt pin.
>>>>>>>>>
>>>>>>>>> We are working on drivers for such devices and would be very nice to
>>>>>>>>> move this driver in advance to the IIO non-staging location.
>>>>>>>>>
>>>>>>>>> What do you say?
>>>>>>>>>
>>>>>>>>
>>>>>>>> Hmm, I wonder what are the advantages of using RTC timers. Couldn't we
>>>>>>>> use a regular kernel timer instead?
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> The long term plan is to get rid of the RTC timer trigger due to its
>>>>>>> various
>>>>>>> limitations (poor resolution, etc).
>>>>>>>
>>>>>>> There is the hrtimer trigger
>>>>>>>
>>>>>>> (https://github.com/analogdevicesinc/linux/blob/xcomm_zynq/drivers/staging/iio/trigger/iio-trig-hrtimer.c)
>>>>>>> but we haven't agreed on a proper interface yet how to instantiate the
>>>>>>> hrtimer trigger.
>>>>>>>
>>>>>>> Check the ml archive for the various discussions on it:
>>>>>>> http://marc.info/?l=linux-iio&w=2&r=1&s=hrtimer&q=b
>>>>>>
>>>>>>
>>>>>>
>>>>>> Hi Lars,
>>>>>>
>>>>>> That was an interesting reading. There were people trying to push
>>>>>> hrtimer based IIO trigger 4 years ago :).
>>>>>>
>>>>>> I think it's now the time to have this upstream.
>>>>>>
>>>>>> I will be back :) (as many others said before) with an RFC patch.
>>>>>>
>>>>>> I think we should keep the following requirements:
>>>>>>
>>>>>> 1) Create a common framework for software based triggers.
>>>>>> 2) User space driven configuration for trigger instances,
>>>>>> as opposed to platform device files used for RTC based trigger
>>>>>> 3) Remove RTC interrupt source, use hrtimers instead
>>>>>>
>>>>>> Still not clear, but I will trying to figure it out during
>>>>>> implementation:
>>>>>>
>>>>>> 4) configfs vs sysfs interface.
>>>>>>
>>>>>> At the first glance, I would say we should stay with sysfs interface in
>>>>>> order
>>>>>> to avoid another dependency. But let's see how it works.
>>>>>
>>>>> This issue with the sysfs only approach (as originally raised by Lars)
>>>>> is that it is actually very poorly suited to instantiating new elements
>>>>> of
>>>>> the device model. Configfs was introduced in the first place exactly to
>>>>> cover this area. We only ended up with the instantiation code in
>>>>> the sysfs trigger via sysfs because at the time (a good long while ago!)
>>>>> I wasn't aware of configfs.
>>>>>
>>>>> I have some initial work on the base elements on an iio configfs
>>>>> interface
>>>>> somewhere that I can dig out if you like. I started working on it in a
>>>>> rare
>>>>> quiet period about a year ago, but never got all that far.
>>>>>
>>>>> There aren't that many examples in tree of how to actually use configfs
>>>>> so it's a bit more of a learning curve than sysfs!
>>>>
>>>>
>>>> First notable problem with using configfs with IIO is boot time modules
>>>> loading.
>>>>
>>>> Because,
>>>>
>>>> * configs uses module_init(configfs_init);
>>>> * IIO uses subsys_initcall(iio_init);
>>>>
>>>> it is guaranteed that IIO will be loaded before configfs. Not fun! :)
>>>>
>>>> So for the moment I will not add configfs support directly into
>>>> industrialiio-core but in a separate module.
>>>
>>>
>>> Just fix configfs, that should clearly not be at the module init level.
>>>
>>> But keeping it in a separate module doesn't hurt either way.
>>
>> Looking at drivers/fs all filesystems modules are at module init level.
>> I'll drop an email to fs folks :).
>
> debugfs is at core_initcall() and in my opinion configfs is on the same level from a infrastructure point of view.
Seems logical. As far as I could tell, last time I looked, configfs still had
relatively few users and I guess most of the are individual drivers.

It's a good idea, but we'll be breaking some new ground using it
extensively in IIO.

It did occur to me that we should be drawing up a list of what IIO elements
it makes sense for.

Quick thoughts from me on this would be

1) Software trigger creation / destruction
2) Mappings to 'soft' in kernel users such as iio_hwmon and iio_input.
Doing these via the device tree has always been a somewhat 'interesting'
corner and rather limiting. I have some thoughts on more advanced
'filter' drivers that do some of what the sensor hubs we are seeing are
doing (step counting for example) but without needing an additional
processor. Will be interesting to see if we save enough overhead to
make this worthwhile, rather than blasting all data out to user space.
3) Perhaps iio on iio drivers (allows for complex data demux / mux operations)
and also allows us to drive the separation between the IIO userspace
interface and the backend. Something that has been on the wish list for
a while.

Quick win would be get (1) done, but we want to keep the layout etc flexible
enough to allow for (2) at least.

Jonathan
>
>>
>> What if IIO core should be at module init core?
>
> It's a subsystem, it should be at subsys_initcall().
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html



\
 
 \ /
  Last update: 2015-03-21 13:01    [W:0.077 / U:0.124 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site