iio_trigger_poll_chained stop and trig->use_count always equals to 1

iio_trigger_poll_chained stop and trig->use_count always equals to 1

[Denis CIOCCA]

Hi all,

I need your help to understanding my strange issue...
The scenario is that:

- I have one I2C device (microcontroller) that expose some sensors.
- I wrote one driver that create one IIO device for each sensor.
- There is only one trigger associated to all IIO devices, and one buffer f=
or=20
each device.

When interrupt appear (DRDY of one or more sensors), the driver reads a mas=
k=20
from micro to understand how many sensors have new data. After that, the=20
driver reads all new data from micro and save all data to one buffer. This =
is=20
done in a threaded irq function.

When iio_trigger_poll_chained is called, all data are saved to one common=20
buffer, each iio_triggered_buffer_setup functions is called and can split a=
nd=20
push their data to iio_buffer.

The issue is that: after some samples (about 50:100) the=20
iio_trigger_poll_chained doesn't call the iio_triggered_buffer_setup functi=
ons=20
and trig->use_count is always equals to 1.

To well understanding:

static irqreturn_t st_drdy_poll(int irq, void *private)
{
	// stuff statements
	// i2c read mask
	// i2c read sensors data
	iio_trigger_poll_chained(trig, 0);
	return IRQ_HANDLED;
}

static irqreturn_t st_trigger_handler(int irq, void *p)
{
	// stuff statements
	// only check operations
	iio_push_to_buffers(sensor_iio_dev, sdata->buffer_data);
	iio_trigger_notify_done(sensor_iio_dev->trig);
	return IRQ_HANDLED;
}

trigger_allocate()
{
err =3D request_threaded_irq(client->irq,
				NULL,
				st_drdy_poll,
				IRQF_TRIGGER_RISING | IRQF_ONESHOT,
				name,
				trig);
}

buffer_allocate()
{
return iio_triggered_buffer_setup(sensor_iio_dev,
						&iio_pollfunc_store_time,
						&st_trigger_handler,
						&st_buffer_setup_ops);
}



If you have any comments...Thanks for your time!

Denis

Re: iio_trigger_poll_chained stop and trig->use_count always equals to 1

[Lars-Peter Clausen]

On 07/16/2013 11:26 AM, Denis CIOCCA wrote:
> Hi all,
> 
> I need your help to understanding my strange issue...
> The scenario is that:
> 
> - I have one I2C device (microcontroller) that expose some sensors.
> - I wrote one driver that create one IIO device for each sensor.
> - There is only one trigger associated to all IIO devices, and one buffer for 
> each device.
> 
> When interrupt appear (DRDY of one or more sensors), the driver reads a mask 
> from micro to understand how many sensors have new data. After that, the 
> driver reads all new data from micro and save all data to one buffer. This is 
> done in a threaded irq function.
> 
> When iio_trigger_poll_chained is called, all data are saved to one common 
> buffer, each iio_triggered_buffer_setup functions is called and can split and 
> push their data to iio_buffer.
> 
> The issue is that: after some samples (about 50:100) the 
> iio_trigger_poll_chained doesn't call the iio_triggered_buffer_setup functions 
> and trig->use_count is always equals to 1.
> 

Yea, there is a race condition. Try this patch:
http://www.spinics.net/lists/linux-iio/msg08710.html

- Lars

Re: iio_trigger_poll_chained stop and trig->use_count always equals to 1

[Denis CIOCCA]

Hi Lars,

thank you very much, now it works fine!

Only one thing:
- In your patch I think there is a mistake on iio_trigger_poll_chained 
function: generic_handle_irq should be replaced by handle_nested_irq.

Thanks,
Denis


On Tuesday, July 16, 2013 12:28:06 PM Lars-Peter Clausen wrote:
> On 07/16/2013 11:26 AM, Denis CIOCCA wrote:
> > Hi all,
> > 
> > I need your help to understanding my strange issue...
> > The scenario is that:
> > 
> > - I have one I2C device (microcontroller) that expose some sensors.
> > - I wrote one driver that create one IIO device for each sensor.
> > - There is only one trigger associated to all IIO devices, and one buffer
> > for each device.
> > 
> > When interrupt appear (DRDY of one or more sensors), the driver reads a
> > mask from micro to understand how many sensors have new data. After that,
> > the driver reads all new data from micro and save all data to one buffer.
> > This is done in a threaded irq function.
> > 
> > When iio_trigger_poll_chained is called, all data are saved to one common
> > buffer, each iio_triggered_buffer_setup functions is called and can split
> > and push their data to iio_buffer.
> > 
> > The issue is that: after some samples (about 50:100) the
> > iio_trigger_poll_chained doesn't call the iio_triggered_buffer_setup
> > functions and trig->use_count is always equals to 1.
> 
> Yea, there is a race condition. Try this patch:
> http://www.spinics.net/lists/linux-iio/msg08710.html
> 
> - Lars

Re: iio_trigger_poll_chained stop and trig->use_count always equals to 1

[Lars-Peter Clausen]

On 07/16/2013 03:26 PM, Denis CIOCCA wrote:
> Hi Lars,
> 
> thank you very much, now it works fine!
> 
> Only one thing:
> - In your patch I think there is a mistake on iio_trigger_poll_chained 
> function: generic_handle_irq should be replaced by handle_nested_irq.
> 
Yep, thanks. Will fix that in v2. Can you give v2 a try?

- Lars


> 
> 
> On Tuesday, July 16, 2013 12:28:06 PM Lars-Peter Clausen wrote:
>> On 07/16/2013 11:26 AM, Denis CIOCCA wrote:
>>> Hi all,
>>>
>>> I need your help to understanding my strange issue...
>>> The scenario is that:
>>>
>>> - I have one I2C device (microcontroller) that expose some sensors.
>>> - I wrote one driver that create one IIO device for each sensor.
>>> - There is only one trigger associated to all IIO devices, and one buffer
>>> for each device.
>>>
>>> When interrupt appear (DRDY of one or more sensors), the driver reads a
>>> mask from micro to understand how many sensors have new data. After that,
>>> the driver reads all new data from micro and save all data to one buffer.
>>> This is done in a threaded irq function.
>>>
>>> When iio_trigger_poll_chained is called, all data are saved to one common
>>> buffer, each iio_triggered_buffer_setup functions is called and can split
>>> and push their data to iio_buffer.
>>>
>>> The issue is that: after some samples (about 50:100) the
>>> iio_trigger_poll_chained doesn't call the iio_triggered_buffer_setup
>>> functions and trig->use_count is always equals to 1.
>>
>> Yea, there is a race condition. Try this patch:
>> http://www.spinics.net/lists/linux-iio/msg08710.html
>>
>> - Lars