[PATCH] staging: iio: ad7606: use iio_device_{claim|release}_direct_mode()

[PATCH] staging: iio: ad7606: use iio_device_{claim|release}_direct_mode()

[Alison Schofield]

Two instances are moved to the new claim/release API:

In the first instance, the driver was using mlock followed by
iio_buffer_enabled(). Replace that code with the new API to guarantee
the device stays in direct mode. There is no change in driver behavior.

In the second instance, the driver was not using mlock to hold the
device in direct mode, but should have been.  Here we introduce the
new API to guarantee direct mode. This is a change in driver behavior.

Signed-off-by: Alison Schofield <amsfield22@gmail.com>
---
 drivers/staging/iio/adc/ad7606_core.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/drivers/staging/iio/adc/ad7606_core.c b/drivers/staging/iio/adc/ad7606_core.c
index 6dbc811..f914b8d 100644
--- a/drivers/staging/iio/adc/ad7606_core.c
+++ b/drivers/staging/iio/adc/ad7606_core.c
@@ -88,12 +88,12 @@ static int ad7606_read_raw(struct iio_dev *indio_dev,
 
 	switch (m) {
 	case IIO_CHAN_INFO_RAW:
-		mutex_lock(&indio_dev->mlock);
-		if (iio_buffer_enabled(indio_dev))
-			ret = -EBUSY;
-		else
-			ret = ad7606_scan_direct(indio_dev, chan->address);
-		mutex_unlock(&indio_dev->mlock);
+		ret = iio_device_claim_direct_mode(indio_dev);
+		if (ret)
+			return ret;
+
+		ret = ad7606_scan_direct(indio_dev, chan->address);
+		iio_device_release_direct_mode(indio_dev);
 
 		if (ret < 0)
 			return ret;
@@ -411,8 +411,9 @@ static irqreturn_t ad7606_interrupt(int irq, void *dev_id)
 	struct iio_dev *indio_dev = dev_id;
 	struct ad7606_state *st = iio_priv(indio_dev);
 
-	if (iio_buffer_enabled(indio_dev)) {
+	if (!iio_device_claim_direct_mode(indio_dev))  {
 		schedule_work(&st->poll_work);
+		iio_device_release_direct_mode(indio_dev);
 	} else {
 		st->done = true;
 		wake_up_interruptible(&st->wq_data_avail);
-- 
2.1.4

Re: [PATCH] staging: iio: ad7606: use iio_device_{claim|release}_direct_mode()

[Jonathan Cameron]

On 01/04/16 17:53, Alison Schofield wrote:
> Two instances are moved to the new claim/release API:
> 
> In the first instance, the driver was using mlock followed by
> iio_buffer_enabled(). Replace that code with the new API to guarantee
> the device stays in direct mode. There is no change in driver behavior.
> 
> In the second instance, the driver was not using mlock to hold the
> device in direct mode, but should have been.  Here we introduce the
> new API to guarantee direct mode. This is a change in driver behavior.
> 
> Signed-off-by: Alison Schofield <amsfield22@gmail.com>
> ---
>  drivers/staging/iio/adc/ad7606_core.c | 15 ++++++++-------
>  1 file changed, 8 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/staging/iio/adc/ad7606_core.c b/drivers/staging/iio/adc/ad7606_core.c
> index 6dbc811..f914b8d 100644
> --- a/drivers/staging/iio/adc/ad7606_core.c
> +++ b/drivers/staging/iio/adc/ad7606_core.c
> @@ -88,12 +88,12 @@ static int ad7606_read_raw(struct iio_dev *indio_dev,
>  
>  	switch (m) {
>  	case IIO_CHAN_INFO_RAW:
> -		mutex_lock(&indio_dev->mlock);
> -		if (iio_buffer_enabled(indio_dev))
> -			ret = -EBUSY;
> -		else
> -			ret = ad7606_scan_direct(indio_dev, chan->address);
> -		mutex_unlock(&indio_dev->mlock);
> +		ret = iio_device_claim_direct_mode(indio_dev);
> +		if (ret)
> +			return ret;
> +
> +		ret = ad7606_scan_direct(indio_dev, chan->address);
> +		iio_device_release_direct_mode(indio_dev);
>  
>  		if (ret < 0)
>  			return ret;
> @@ -411,8 +411,9 @@ static irqreturn_t ad7606_interrupt(int irq, void *dev_id)
>  	struct iio_dev *indio_dev = dev_id;
>  	struct ad7606_state *st = iio_priv(indio_dev);
>  
> -	if (iio_buffer_enabled(indio_dev)) {
> +	if (!iio_device_claim_direct_mode(indio_dev))  {
>  		schedule_work(&st->poll_work);
> +		iio_device_release_direct_mode(indio_dev);
Unfortunately this won't work.  That interrupt is still in traditional non
threaded form.  This will take a mutex in a top half interrupt handler
where a sleep cannot occur.

I'm just wondering how expensive it would be to fix this by moving that over
to a threaded handler.  In the poll_work case (buffer) it would be cleaner to do
so. I'm really confused what the intended interrupt handler
is in here.  I 'think' the sequence is:

Trigger fires the convst pin whether in top half or the bottom half of
a threaded interrupt, but not both - I guess this works, if it is rather
'unusual'.

We then get a interrupt to indicate that it has finished conversion and that
filters through to actually fill the buffer via a traditional top half /
bottom half interrupt handler.

So if we were to convert that to a threaded interrupt (with no top half / non
threaded part), we could drop the schedule_work and just call
ad7606_poll_bh_to_ring from the thread handler.

In the direct read case I doubt we care about the delay in dropping to a
thread prior to signalling the data is ready.

Can't think why this driver is still in staging :)

Lars, any interest from Analog in getting this one cleaned up?  Also
do you have any test hardware, if we mess around with this interrupt handling?

Jonathan


>  	} else {
>  		st->done = true;
>  		wake_up_interruptible(&st->wq_data_avail);
> 

Re: [PATCH] staging: iio: ad7606: use iio_device_{claim|release}_direct_mode()

[Alison Schofield]

On Sun, Apr 03, 2016 at 10:09:13AM +0100, Jonathan Cameron wrote:
> On 01/04/16 17:53, Alison Schofield wrote:
> > Two instances are moved to the new claim/release API:
> > 
> > In the first instance, the driver was using mlock followed by
> > iio_buffer_enabled(). Replace that code with the new API to guarantee
> > the device stays in direct mode. There is no change in driver behavior.
> > 
> > In the second instance, the driver was not using mlock to hold the
> > device in direct mode, but should have been.  Here we introduce the
> > new API to guarantee direct mode. This is a change in driver behavior.
> > 
> > Signed-off-by: Alison Schofield <amsfield22@gmail.com>
> > ---
> >  drivers/staging/iio/adc/ad7606_core.c | 15 ++++++++-------
> >  1 file changed, 8 insertions(+), 7 deletions(-)
> > 
> > diff --git a/drivers/staging/iio/adc/ad7606_core.c b/drivers/staging/iio/adc/ad7606_core.c
> > index 6dbc811..f914b8d 100644
> > --- a/drivers/staging/iio/adc/ad7606_core.c
> > +++ b/drivers/staging/iio/adc/ad7606_core.c
> > @@ -88,12 +88,12 @@ static int ad7606_read_raw(struct iio_dev *indio_dev,
> >  
> >  	switch (m) {
> >  	case IIO_CHAN_INFO_RAW:
> > -		mutex_lock(&indio_dev->mlock);
> > -		if (iio_buffer_enabled(indio_dev))
> > -			ret = -EBUSY;
> > -		else
> > -			ret = ad7606_scan_direct(indio_dev, chan->address);
> > -		mutex_unlock(&indio_dev->mlock);
> > +		ret = iio_device_claim_direct_mode(indio_dev);
> > +		if (ret)
> > +			return ret;
> > +
> > +		ret = ad7606_scan_direct(indio_dev, chan->address);
> > +		iio_device_release_direct_mode(indio_dev);
> >  
> >  		if (ret < 0)
> >  			return ret;
> > @@ -411,8 +411,9 @@ static irqreturn_t ad7606_interrupt(int irq, void *dev_id)
> >  	struct iio_dev *indio_dev = dev_id;
> >  	struct ad7606_state *st = iio_priv(indio_dev);
> >  
> > -	if (iio_buffer_enabled(indio_dev)) {
> > +	if (!iio_device_claim_direct_mode(indio_dev))  {
> >  		schedule_work(&st->poll_work);
> > +		iio_device_release_direct_mode(indio_dev);
> Unfortunately this won't work.  That interrupt is still in traditional non
> threaded form.  This will take a mutex in a top half interrupt handler
> where a sleep cannot occur.
> 
> I'm just wondering how expensive it would be to fix this by moving that over
> to a threaded handler.  In the poll_work case (buffer) it would be cleaner to do
> so. I'm really confused what the intended interrupt handler
> is in here.  I 'think' the sequence is:
> 
> Trigger fires the convst pin whether in top half or the bottom half of
> a threaded interrupt, but not both - I guess this works, if it is rather
> 'unusual'.
> 
> We then get a interrupt to indicate that it has finished conversion and that
> filters through to actually fill the buffer via a traditional top half /
> bottom half interrupt handler.
> 
> So if we were to convert that to a threaded interrupt (with no top half / non
> threaded part), we could drop the schedule_work and just call
> ad7606_poll_bh_to_ring from the thread handler.
> 
> In the direct read case I doubt we care about the delay in dropping to a
> thread prior to signalling the data is ready.
> 
> Can't think why this driver is still in staging :)
> 
> Lars, any interest from Analog in getting this one cleaned up?  Also
> do you have any test hardware, if we mess around with this interrupt handling?
> 
> Jonathan

I see the problem. Thanks for the review and the details for the
redesign.  Plan to v2 this patch leaving out this piece. I'll keep
the interrupt handler rework on my radar pending Lars's comments.

alisons

> 
> 
> >  	} else {
> >  		st->done = true;
> >  		wake_up_interruptible(&st->wq_data_avail);
> > 
> 

Re: [PATCH] staging: iio: ad7606: use iio_device_{claim|release}_direct_mode()

[Lars-Peter Clausen]

On 04/03/2016 11:09 AM, Jonathan Cameron wrote:
> On 01/04/16 17:53, Alison Schofield wrote:
>> Two instances are moved to the new claim/release API:
>>
>> In the first instance, the driver was using mlock followed by
>> iio_buffer_enabled(). Replace that code with the new API to guarantee
>> the device stays in direct mode. There is no change in driver behavior.
>>
>> In the second instance, the driver was not using mlock to hold the
>> device in direct mode, but should have been.  Here we introduce the
>> new API to guarantee direct mode. This is a change in driver behavior.
>>
>> Signed-off-by: Alison Schofield <amsfield22@gmail.com>
>> ---
>>  drivers/staging/iio/adc/ad7606_core.c | 15 ++++++++-------
>>  1 file changed, 8 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/staging/iio/adc/ad7606_core.c
b/drivers/staging/iio/adc/ad7606_core.c
>> index 6dbc811..f914b8d 100644
>> --- a/drivers/staging/iio/adc/ad7606_core.c
>> +++ b/drivers/staging/iio/adc/ad7606_core.c
>> @@ -88,12 +88,12 @@ static int ad7606_read_raw(struct iio_dev *indio_dev,
>>
>>  	switch (m) {
>>  	case IIO_CHAN_INFO_RAW:
>> -		mutex_lock(&indio_dev->mlock);
>> -		if (iio_buffer_enabled(indio_dev))
>> -			ret = -EBUSY;
>> -		else
>> -			ret = ad7606_scan_direct(indio_dev, chan->address);
>> -		mutex_unlock(&indio_dev->mlock);
>> +		ret = iio_device_claim_direct_mode(indio_dev);
>> +		if (ret)
>> +			return ret;
>> +
>> +		ret = ad7606_scan_direct(indio_dev, chan->address);
>> +		iio_device_release_direct_mode(indio_dev);
>>
>>  		if (ret < 0)
>>  			return ret;
>> @@ -411,8 +411,9 @@ static irqreturn_t ad7606_interrupt(int irq, void
*dev_id)
>>  	struct iio_dev *indio_dev = dev_id;
>>  	struct ad7606_state *st = iio_priv(indio_dev);
>>
>> -	if (iio_buffer_enabled(indio_dev)) {
>> +	if (!iio_device_claim_direct_mode(indio_dev))  {
>>  		schedule_work(&st->poll_work);
>> +		iio_device_release_direct_mode(indio_dev);
> Unfortunately this won't work.  That interrupt is still in traditional non
> threaded form.  This will take a mutex in a top half interrupt handler
> where a sleep cannot occur.
>
> I'm just wondering how expensive it would be to fix this by moving that over
> to a threaded handler.  In the poll_work case (buffer) it would be cleaner
to do
> so. I'm really confused what the intended interrupt handler
> is in here.  I 'think' the sequence is:
>
> Trigger fires the convst pin whether in top half or the bottom half of
> a threaded interrupt, but not both - I guess this works, if it is rather
> 'unusual'.
>
> We then get a interrupt to indicate that it has finished conversion and that
> filters through to actually fill the buffer via a traditional top half /
> bottom half interrupt handler.
>
> So if we were to convert that to a threaded interrupt (with no top half / non
> threaded part), we could drop the schedule_work and just call
> ad7606_poll_bh_to_ring from the thread handler.
>
> In the direct read case I doubt we care about the delay in dropping to a
> thread prior to signalling the data is ready.
>
> Can't think why this driver is still in staging :)

Yeah, we should leave this driver out from the conversion for now. The whole
convst pin handling need a major rework. It shouldn't really be in the
driver and usually you wouldn't want to use to use a GPIO and software timer
since that gives you way to much jitter for good results. You'd probably use
something like a PWM.

>
> Lars, any interest from Analog in getting this one cleaned up?  Also
> do you have any test hardware, if we mess around with this interrupt handling?

I have the hardware somewhere in some storage bay, but just converting this
over to threaded interrupt handling is not really a solution. So, if you
want to get rid of the iio_buffer_enabled() in the interrupt handler a
simple solution is to register preenable and postdisable callbacks where you
set a flag in the driver struct to indicate that it is in buffered mode or not.

Re: [PATCH] staging: iio: ad7606: use iio_device_{claim|release}_direct_mode()

[Jonathan Cameron]

On 06/04/16 19:06, Lars-Peter Clausen wrote:
> On 04/03/2016 11:09 AM, Jonathan Cameron wrote:
>> On 01/04/16 17:53, Alison Schofield wrote:
>>> Two instances are moved to the new claim/release API:
>>>
>>> In the first instance, the driver was using mlock followed by
>>> iio_buffer_enabled(). Replace that code with the new API to guarantee
>>> the device stays in direct mode. There is no change in driver behavior.
>>>
>>> In the second instance, the driver was not using mlock to hold the
>>> device in direct mode, but should have been.  Here we introduce the
>>> new API to guarantee direct mode. This is a change in driver behavior.
>>>
>>> Signed-off-by: Alison Schofield <amsfield22@gmail.com>
>>> ---
>>>  drivers/staging/iio/adc/ad7606_core.c | 15 ++++++++-------
>>>  1 file changed, 8 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/drivers/staging/iio/adc/ad7606_core.c
> b/drivers/staging/iio/adc/ad7606_core.c
>>> index 6dbc811..f914b8d 100644
>>> --- a/drivers/staging/iio/adc/ad7606_core.c
>>> +++ b/drivers/staging/iio/adc/ad7606_core.c
>>> @@ -88,12 +88,12 @@ static int ad7606_read_raw(struct iio_dev *indio_dev,
>>>
>>>  	switch (m) {
>>>  	case IIO_CHAN_INFO_RAW:
>>> -		mutex_lock(&indio_dev->mlock);
>>> -		if (iio_buffer_enabled(indio_dev))
>>> -			ret = -EBUSY;
>>> -		else
>>> -			ret = ad7606_scan_direct(indio_dev, chan->address);
>>> -		mutex_unlock(&indio_dev->mlock);
>>> +		ret = iio_device_claim_direct_mode(indio_dev);
>>> +		if (ret)
>>> +			return ret;
>>> +
>>> +		ret = ad7606_scan_direct(indio_dev, chan->address);
>>> +		iio_device_release_direct_mode(indio_dev);
>>>
>>>  		if (ret < 0)
>>>  			return ret;
>>> @@ -411,8 +411,9 @@ static irqreturn_t ad7606_interrupt(int irq, void
> *dev_id)
>>>  	struct iio_dev *indio_dev = dev_id;
>>>  	struct ad7606_state *st = iio_priv(indio_dev);
>>>
>>> -	if (iio_buffer_enabled(indio_dev)) {
>>> +	if (!iio_device_claim_direct_mode(indio_dev))  {
>>>  		schedule_work(&st->poll_work);
>>> +		iio_device_release_direct_mode(indio_dev);
>> Unfortunately this won't work.  That interrupt is still in traditional non
>> threaded form.  This will take a mutex in a top half interrupt handler
>> where a sleep cannot occur.
>>
>> I'm just wondering how expensive it would be to fix this by moving that over
>> to a threaded handler.  In the poll_work case (buffer) it would be cleaner
> to do
>> so. I'm really confused what the intended interrupt handler
>> is in here.  I 'think' the sequence is:
>>
>> Trigger fires the convst pin whether in top half or the bottom half of
>> a threaded interrupt, but not both - I guess this works, if it is rather
>> 'unusual'.
>>
>> We then get a interrupt to indicate that it has finished conversion and that
>> filters through to actually fill the buffer via a traditional top half /
>> bottom half interrupt handler.
>>
>> So if we were to convert that to a threaded interrupt (with no top half / non
>> threaded part), we could drop the schedule_work and just call
>> ad7606_poll_bh_to_ring from the thread handler.
>>
>> In the direct read case I doubt we care about the delay in dropping to a
>> thread prior to signalling the data is ready.
>>
>> Can't think why this driver is still in staging :)
> 
> Yeah, we should leave this driver out from the conversion for now. The whole
> convst pin handling need a major rework. It shouldn't really be in the
> driver and usually you wouldn't want to use to use a GPIO and software timer
> since that gives you way to much jitter for good results. You'd probably use
> something like a PWM.
Agreed, you would normally move this into hardware, but we'd still need to allow
for the possibility of it being software triggered - even if ugly.

Anyhow, lets park this for now.

J
> 
>>
>> Lars, any interest from Analog in getting this one cleaned up?  Also
>> do you have any test hardware, if we mess around with this interrupt handling?
> 
> I have the hardware somewhere in some storage bay, but just converting this
> over to threaded interrupt handling is not really a solution. So, if you
> want to get rid of the iio_buffer_enabled() in the interrupt handler a
> simple solution is to register preenable and postdisable callbacks where you
> set a flag in the driver struct to indicate that it is in buffered mode or not.
> 

[PATCH v2] staging: iio: ad7606: use iio_device_{claim|release}_direct_mode()

[Alison Schofield]

Replace the code that guarantees the device stays in direct mode with
iio_device_{claim|release}_direct_mode() which does same.

Signed-off-by: Alison Schofield <amsfield22@gmail.com>
---
Changed in v2:
 - removed improper application of claim/release from intr handler
 - updated changelog

 drivers/staging/iio/adc/ad7606_core.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/staging/iio/adc/ad7606_core.c b/drivers/staging/iio/adc/ad7606_core.c
index 6dbc811..f79ee61 100644
--- a/drivers/staging/iio/adc/ad7606_core.c
+++ b/drivers/staging/iio/adc/ad7606_core.c
@@ -88,12 +88,12 @@ static int ad7606_read_raw(struct iio_dev *indio_dev,
 
 	switch (m) {
 	case IIO_CHAN_INFO_RAW:
-		mutex_lock(&indio_dev->mlock);
-		if (iio_buffer_enabled(indio_dev))
-			ret = -EBUSY;
-		else
-			ret = ad7606_scan_direct(indio_dev, chan->address);
-		mutex_unlock(&indio_dev->mlock);
+		ret = iio_device_claim_direct_mode(indio_dev);
+		if (ret)
+			return ret;
+
+		ret = ad7606_scan_direct(indio_dev, chan->address);
+		iio_device_release_direct_mode(indio_dev);
 
 		if (ret < 0)
 			return ret;
-- 
2.1.4

Re: [PATCH v2] staging: iio: ad7606: use iio_device_{claim|release}_direct_mode()

[Lars-Peter Clausen]

On 04/06/2016 06:11 AM, Alison Schofield wrote:
> Replace the code that guarantees the device stays in direct mode with
> iio_device_{claim|release}_direct_mode() which does same.
> 
> Signed-off-by: Alison Schofield <amsfield22@gmail.com>

Looks good, thanks.

Acked-by: Lars-Peter Clausen <lars@metafoo.de>

> ---
> Changed in v2:
>  - removed improper application of claim/release from intr handler
>  - updated changelog
> 
>  drivers/staging/iio/adc/ad7606_core.c | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/staging/iio/adc/ad7606_core.c b/drivers/staging/iio/adc/ad7606_core.c
> index 6dbc811..f79ee61 100644
> --- a/drivers/staging/iio/adc/ad7606_core.c
> +++ b/drivers/staging/iio/adc/ad7606_core.c
> @@ -88,12 +88,12 @@ static int ad7606_read_raw(struct iio_dev *indio_dev,
>  
>  	switch (m) {
>  	case IIO_CHAN_INFO_RAW:
> -		mutex_lock(&indio_dev->mlock);
> -		if (iio_buffer_enabled(indio_dev))
> -			ret = -EBUSY;
> -		else
> -			ret = ad7606_scan_direct(indio_dev, chan->address);
> -		mutex_unlock(&indio_dev->mlock);
> +		ret = iio_device_claim_direct_mode(indio_dev);
> +		if (ret)
> +			return ret;
> +
> +		ret = ad7606_scan_direct(indio_dev, chan->address);
> +		iio_device_release_direct_mode(indio_dev);
>  
>  		if (ret < 0)
>  			return ret;
> 

Re: [PATCH v2] staging: iio: ad7606: use iio_device_{claim|release}_direct_mode()

[Jonathan Cameron]

On 06/04/16 22:06, Lars-Peter Clausen wrote:
> On 04/06/2016 06:11 AM, Alison Schofield wrote:
>> Replace the code that guarantees the device stays in direct mode with
>> iio_device_{claim|release}_direct_mode() which does same.
>>
>> Signed-off-by: Alison Schofield <amsfield22@gmail.com>
> 
> Looks good, thanks.
> 
> Acked-by: Lars-Peter Clausen <lars@metafoo.de>
> 
Applied to the togreg branch of iio.git - thanks.

Jonathan
>> ---
>> Changed in v2:
>>  - removed improper application of claim/release from intr handler
>>  - updated changelog
>>
>>  drivers/staging/iio/adc/ad7606_core.c | 12 ++++++------
>>  1 file changed, 6 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/staging/iio/adc/ad7606_core.c b/drivers/staging/iio/adc/ad7606_core.c
>> index 6dbc811..f79ee61 100644
>> --- a/drivers/staging/iio/adc/ad7606_core.c
>> +++ b/drivers/staging/iio/adc/ad7606_core.c
>> @@ -88,12 +88,12 @@ static int ad7606_read_raw(struct iio_dev *indio_dev,
>>  
>>  	switch (m) {
>>  	case IIO_CHAN_INFO_RAW:
>> -		mutex_lock(&indio_dev->mlock);
>> -		if (iio_buffer_enabled(indio_dev))
>> -			ret = -EBUSY;
>> -		else
>> -			ret = ad7606_scan_direct(indio_dev, chan->address);
>> -		mutex_unlock(&indio_dev->mlock);
>> +		ret = iio_device_claim_direct_mode(indio_dev);
>> +		if (ret)
>> +			return ret;
>> +
>> +		ret = ad7606_scan_direct(indio_dev, chan->address);
>> +		iio_device_release_direct_mode(indio_dev);
>>  
>>  		if (ret < 0)
>>  			return ret;
>>
>