[PATCH] reset: Add reset controller API

[PATCH] reset: Add reset controller API

[Amjad Ouled-Ameur]

The current reset framework API does not allow to release what is done by
reset_control_reset(), IOW decrement triggered_count. Add the new
reset_control_resettable() call to do so.

When reset_control_reset() has been called once, the counter
triggered_count, in the reset framework, is incremented i.e the resource
under the reset is in-use and the reset should not be done again.
reset_control_resettable() would be the way to state that the resource is 
no longer used and, that from the caller's perspective, the reset can be 
fired again if necessary.

This patch will fix a usb suspend warning seen on the libretech-cc
related to the shared reset line. This warning was addressed by the 
previously reverted commit 7a410953d1fb ("usb: dwc3: meson-g12a: fix shared
reset control use")

Signed-off-by: Amjad Ouled-Ameur <aouledameur@baylibre.com>
Reported-by: Jerome Brunet <jbrunet@baylibre.com>
---
 drivers/reset/core.c  | 57 +++++++++++++++++++++++++++++++++++++++++++
 include/linux/reset.h |  1 +
 2 files changed, 58 insertions(+)

diff --git a/drivers/reset/core.c b/drivers/reset/core.c
index 01c0c7aa835c..53653d4b55c4 100644
--- a/drivers/reset/core.c
+++ b/drivers/reset/core.c
@@ -207,6 +207,19 @@ static int reset_control_array_reset(struct reset_control_array *resets)
 	return 0;
 }
 
+static int reset_control_array_resettable(struct reset_control_array *resets)
+{
+	int ret, i;
+
+	for (i = 0; i < resets->num_rstcs; i++) {
+		ret = reset_control_resettable(resets->rstc[i]);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
 static int reset_control_array_assert(struct reset_control_array *resets)
 {
 	int ret, i;
@@ -324,6 +337,50 @@ int reset_control_reset(struct reset_control *rstc)
 }
 EXPORT_SYMBOL_GPL(reset_control_reset);
 
+/**
+ * reset_control_resettable - decrements triggered_count of the controlled device
+ * @rstc: reset controller
+ *
+ * On a shared reset line the actual reset pulse is only triggered once for the
+ * lifetime of the reset_control instance, except if this function is used.
+ * In fact, It decrements triggered_count that makes sure of not allowing
+ * a reset if triggered_count is not null.
+ *
+ * This is a no-op in case triggered_count is already null i.e shared reset line
+ * is ready to be triggered.
+ *
+ * Consumers must not use reset_control_(de)assert on shared reset lines when
+ * reset_control_reset has been used.
+ *
+ * If rstc is NULL it is an optional clear and the function will just
+ * return 0.
+ */
+int reset_control_resettable(struct reset_control *rstc)
+{
+	if (!rstc)
+		return 0;
+
+	if (WARN_ON(IS_ERR(rstc)))
+		return -EINVAL;
+
+	if (reset_control_is_array(rstc))
+		return reset_control_array_resettable(rstc_to_array(rstc));
+
+	if (rstc->shared) {
+		if (WARN_ON(atomic_read(&rstc->deassert_count) != 0))
+			return -EINVAL;
+
+		if (atomic_read(&rstc->triggered_count) > 0)
+			atomic_dec(&rstc->triggered_count);
+	} else {
+		if (!rstc->acquired)
+			return -EPERM;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(reset_control_resettable);
+
 /**
  * reset_control_assert - asserts the reset line
  * @rstc: reset controller
diff --git a/include/linux/reset.h b/include/linux/reset.h
index 05aa9f440f48..ffa447d0d1d6 100644
--- a/include/linux/reset.h
+++ b/include/linux/reset.h
@@ -13,6 +13,7 @@ struct reset_control;
 #ifdef CONFIG_RESET_CONTROLLER
 
 int reset_control_reset(struct reset_control *rstc);
+int reset_control_resettable(struct reset_control *rstc);
 int reset_control_assert(struct reset_control *rstc);
 int reset_control_deassert(struct reset_control *rstc);
 int reset_control_status(struct reset_control *rstc);
-- 
2.17.1

Re: [PATCH] reset: Add reset controller API

[Amjad Ouled-Ameur]

An update on the patch title, since we don't add an API but extend it,
The title should rather be: Add a new call to the reset framework

Le jeu. 1 oct. 2020 à 15:28, Amjad Ouled-Ameur
<aouledameur@baylibre.com> a écrit :
>
> The current reset framework API does not allow to release what is done by
> reset_control_reset(), IOW decrement triggered_count. Add the new
> reset_control_resettable() call to do so.
>
> When reset_control_reset() has been called once, the counter
> triggered_count, in the reset framework, is incremented i.e the resource
> under the reset is in-use and the reset should not be done again.
> reset_control_resettable() would be the way to state that the resource is
> no longer used and, that from the caller's perspective, the reset can be
> fired again if necessary.
>
> This patch will fix a usb suspend warning seen on the libretech-cc
> related to the shared reset line. This warning was addressed by the
> previously reverted commit 7a410953d1fb ("usb: dwc3: meson-g12a: fix shared
> reset control use")
>
> Signed-off-by: Amjad Ouled-Ameur <aouledameur@baylibre.com>
> Reported-by: Jerome Brunet <jbrunet@baylibre.com>
> ---
>  drivers/reset/core.c  | 57 +++++++++++++++++++++++++++++++++++++++++++
>  include/linux/reset.h |  1 +
>  2 files changed, 58 insertions(+)
>
> diff --git a/drivers/reset/core.c b/drivers/reset/core.c
> index 01c0c7aa835c..53653d4b55c4 100644
> --- a/drivers/reset/core.c
> +++ b/drivers/reset/core.c
> @@ -207,6 +207,19 @@ static int reset_control_array_reset(struct reset_control_array *resets)
>         return 0;
>  }
>
> +static int reset_control_array_resettable(struct reset_control_array *resets)
> +{
> +       int ret, i;
> +
> +       for (i = 0; i < resets->num_rstcs; i++) {
> +               ret = reset_control_resettable(resets->rstc[i]);
> +               if (ret)
> +                       return ret;
> +       }
> +
> +       return 0;
> +}
> +
>  static int reset_control_array_assert(struct reset_control_array *resets)
>  {
>         int ret, i;
> @@ -324,6 +337,50 @@ int reset_control_reset(struct reset_control *rstc)
>  }
>  EXPORT_SYMBOL_GPL(reset_control_reset);
>
> +/**
> + * reset_control_resettable - decrements triggered_count of the controlled device
> + * @rstc: reset controller
> + *
> + * On a shared reset line the actual reset pulse is only triggered once for the
> + * lifetime of the reset_control instance, except if this function is used.
> + * In fact, It decrements triggered_count that makes sure of not allowing
> + * a reset if triggered_count is not null.
> + *
> + * This is a no-op in case triggered_count is already null i.e shared reset line
> + * is ready to be triggered.
> + *
> + * Consumers must not use reset_control_(de)assert on shared reset lines when
> + * reset_control_reset has been used.
> + *
> + * If rstc is NULL it is an optional clear and the function will just
> + * return 0.
> + */
> +int reset_control_resettable(struct reset_control *rstc)
> +{
> +       if (!rstc)
> +               return 0;
> +
> +       if (WARN_ON(IS_ERR(rstc)))
> +               return -EINVAL;
> +
> +       if (reset_control_is_array(rstc))
> +               return reset_control_array_resettable(rstc_to_array(rstc));
> +
> +       if (rstc->shared) {
> +               if (WARN_ON(atomic_read(&rstc->deassert_count) != 0))
> +                       return -EINVAL;
> +
> +               if (atomic_read(&rstc->triggered_count) > 0)
> +                       atomic_dec(&rstc->triggered_count);
> +       } else {
> +               if (!rstc->acquired)
> +                       return -EPERM;
> +       }
> +
> +       return 0;
> +}
> +EXPORT_SYMBOL_GPL(reset_control_resettable);
> +
>  /**
>   * reset_control_assert - asserts the reset line
>   * @rstc: reset controller
> diff --git a/include/linux/reset.h b/include/linux/reset.h
> index 05aa9f440f48..ffa447d0d1d6 100644
> --- a/include/linux/reset.h
> +++ b/include/linux/reset.h
> @@ -13,6 +13,7 @@ struct reset_control;
>  #ifdef CONFIG_RESET_CONTROLLER
>
>  int reset_control_reset(struct reset_control *rstc);
> +int reset_control_resettable(struct reset_control *rstc);
>  int reset_control_assert(struct reset_control *rstc);
>  int reset_control_deassert(struct reset_control *rstc);
>  int reset_control_status(struct reset_control *rstc);
> --
> 2.17.1
>


-- 
Amjad Ouled-Ameur
Embedded Linux Engineer - Villeneuve Loubet, FR
Engit@BayLibre - At the Heart of Embedded Linux

Re: [PATCH] reset: Add reset controller API

[Philipp Zabel]

Hi Amjad,

Thank you for the patch, comments below:

On Thu, 2020-10-01 at 15:55 +0200, Amjad Ouled-Ameur wrote:
> An update on the patch title, since we don't add an API but extend it,
> The title should rather be: Add a new call to the reset framework

I think it should even say what functionality is added, for example

"reset: make shared pulsed reset controls re-triggerable"

> Le jeu. 1 oct. 2020 à 15:28, Amjad Ouled-Ameur
> <aouledameur@baylibre.com> a écrit :
> > The current reset framework API does not allow to release what is done by
> > reset_control_reset(), IOW decrement triggered_count. Add the new
> > reset_control_resettable() call to do so.
> > 
> > When reset_control_reset() has been called once, the counter
> > triggered_count, in the reset framework, is incremented i.e the resource
> > under the reset is in-use and the reset should not be done again.
> > reset_control_resettable() would be the way to state that the resource is
> > no longer used and, that from the caller's perspective, the reset can be
> > fired again if necessary.
> > 
> > This patch will fix a usb suspend warning seen on the libretech-cc
> > related to the shared reset line. This warning was addressed by the
> > previously reverted commit 7a410953d1fb ("usb: dwc3: meson-g12a: fix shared
> > reset control use")
> > 
> > Signed-off-by: Amjad Ouled-Ameur <aouledameur@baylibre.com>
> > Reported-by: Jerome Brunet <jbrunet@baylibre.com>
> > ---
> >  drivers/reset/core.c  | 57 +++++++++++++++++++++++++++++++++++++++++++
> >  include/linux/reset.h |  1 +
> >  2 files changed, 58 insertions(+)
> > 
> > diff --git a/drivers/reset/core.c b/drivers/reset/core.c
> > index 01c0c7aa835c..53653d4b55c4 100644
> > --- a/drivers/reset/core.c
> > +++ b/drivers/reset/core.c
> > @@ -207,6 +207,19 @@ static int reset_control_array_reset(struct reset_control_array *resets)
> >         return 0;
> >  }
> > 
> > +static int reset_control_array_resettable(struct reset_control_array *resets)
> > +{
> > +       int ret, i;
> > +
> > +       for (i = 0; i < resets->num_rstcs; i++) {
> > +               ret = reset_control_resettable(resets->rstc[i]);
> > +               if (ret)
> > +                       return ret;
> > +       }

This is tricky, as we can't really roll back decrementing
triggered_count in case just one of those fails.

I think reset_control_array_resettable has to be open coded to first
check for errors and only then loop through all controls and decrement
their triggered_count.

> > +
> > +       return 0;
> > +}
> > +
> >  static int reset_control_array_assert(struct reset_control_array *resets)
> >  {
> >         int ret, i;
> > @@ -324,6 +337,50 @@ int reset_control_reset(struct reset_control *rstc)
> >  }
> >  EXPORT_SYMBOL_GPL(reset_control_reset);
> > 
> > +/**
> > + * reset_control_resettable - decrements triggered_count of the controlled device
> > + * @rstc: reset controller

It is more important to document the purpose of the function than the
mechanism by which it is achieved. triggered_count is an implementation
detail.

Maybe "allow shared reset line to be triggered again" or similar. 

> > + *
> > + * On a shared reset line the actual reset pulse is only triggered once for the
> > + * lifetime of the reset_control instance, except if this function is used.
> > + * In fact, It decrements triggered_count that makes sure of not allowing
> > + * a reset if triggered_count is not null.
> > + *
> > + * This is a no-op in case triggered_count is already null i.e shared reset line
> > + * is ready to be triggered.

This is not a good idea IMHO. It would be better to document that calls
to this function must be balanced with calls to reset_control_reset, and
then throw a big warning below in case deassert_count ever dips below 0.

Otherwise nothing stops drivers from silently decrementing other
driver's trigger count by accidentally calling this multiple times.

> > + *
> > + * Consumers must not use reset_control_(de)assert on shared reset lines when
> > + * reset_control_reset has been used.
> > + *
> > + * If rstc is NULL it is an optional clear and the function will just
> > + * return 0.
> > + */
> > +int reset_control_resettable(struct reset_control *rstc)
> > +{
> > +       if (!rstc)
> > +               return 0;
> > +
> > +       if (WARN_ON(IS_ERR(rstc)))
> > +               return -EINVAL;
> > +
> > +       if (reset_control_is_array(rstc))
> > +               return reset_control_array_resettable(rstc_to_array(rstc));
> > +
> > +       if (rstc->shared) {
> > +               if (WARN_ON(atomic_read(&rstc->deassert_count) != 0))
> > +                       return -EINVAL;
> > +
> > +               if (atomic_read(&rstc->triggered_count) > 0)
> > +                       atomic_dec(&rstc->triggered_count);

I think this should be

		WARN_ON(atomic_dec_return(&rstc->triggered_count) < 0);

regards
Philipp

Re: [PATCH] reset: Add reset controller API

[Amjad Ouled-Ameur]

Hi Philipp,

Thank you very much for the review.

Please find my comments below:

On 02/10/2020 13:14, Philipp Zabel wrote:

> Hi Amjad,
>
> Thank you for the patch, comments below:
>
> On Thu, 2020-10-01 at 15:55 +0200, Amjad Ouled-Ameur wrote:
>> An update on the patch title, since we don't add an API but extend it,
>> The title should rather be: Add a new call to the reset framework
> I think it should even say what functionality is added, for example
>
> "reset: make shared pulsed reset controls re-triggerable"

Will do !

>> Le jeu. 1 oct. 2020 à 15:28, Amjad Ouled-Ameur
>> <aouledameur@baylibre.com> a écrit :
>>> The current reset framework API does not allow to release what is done by
>>> reset_control_reset(), IOW decrement triggered_count. Add the new
>>> reset_control_resettable() call to do so.
>>>
>>> When reset_control_reset() has been called once, the counter
>>> triggered_count, in the reset framework, is incremented i.e the resource
>>> under the reset is in-use and the reset should not be done again.
>>> reset_control_resettable() would be the way to state that the resource is
>>> no longer used and, that from the caller's perspective, the reset can be
>>> fired again if necessary.
>>>
>>> This patch will fix a usb suspend warning seen on the libretech-cc
>>> related to the shared reset line. This warning was addressed by the
>>> previously reverted commit 7a410953d1fb ("usb: dwc3: meson-g12a: fix shared
>>> reset control use")
>>>
>>> Signed-off-by: Amjad Ouled-Ameur <aouledameur@baylibre.com>
>>> Reported-by: Jerome Brunet <jbrunet@baylibre.com>
>>> ---
>>>   drivers/reset/core.c  | 57 +++++++++++++++++++++++++++++++++++++++++++
>>>   include/linux/reset.h |  1 +
>>>   2 files changed, 58 insertions(+)
>>>
>>> diff --git a/drivers/reset/core.c b/drivers/reset/core.c
>>> index 01c0c7aa835c..53653d4b55c4 100644
>>> --- a/drivers/reset/core.c
>>> +++ b/drivers/reset/core.c
>>> @@ -207,6 +207,19 @@ static int reset_control_array_reset(struct reset_control_array *resets)
>>>          return 0;
>>>   }
>>>
>>> +static int reset_control_array_resettable(struct reset_control_array *resets)
>>> +{
>>> +       int ret, i;
>>> +
>>> +       for (i = 0; i < resets->num_rstcs; i++) {
>>> +               ret = reset_control_resettable(resets->rstc[i]);
>>> +               if (ret)
>>> +                       return ret;
>>> +       }
> This is tricky, as we can't really roll back decrementing
> triggered_count in case just one of those fails.
>
> I think reset_control_array_resettable has to be open coded to first
> check for errors and only then loop through all controls and decrement
> their triggered_count.

I agree with this, it is risky to start decrementing before checking for
errors. The V2 will include an open coded version of this function.

>>> +
>>> +       return 0;
>>> +}
>>> +
>>>   static int reset_control_array_assert(struct reset_control_array *resets)
>>>   {
>>>          int ret, i;
>>> @@ -324,6 +337,50 @@ int reset_control_reset(struct reset_control *rstc)
>>>   }
>>>   EXPORT_SYMBOL_GPL(reset_control_reset);
>>>
>>> +/**
>>> + * reset_control_resettable - decrements triggered_count of the controlled device
>>> + * @rstc: reset controller
> It is more important to document the purpose of the function than the
> mechanism by which it is achieved. triggered_count is an implementation
> detail.
>
> Maybe "allow shared reset line to be triggered again" or similar.

Roger that, will do in V2.

>
>>> + *
>>> + * On a shared reset line the actual reset pulse is only triggered once for the
>>> + * lifetime of the reset_control instance, except if this function is used.
>>> + * In fact, It decrements triggered_count that makes sure of not allowing
>>> + * a reset if triggered_count is not null.
>>> + *
>>> + * This is a no-op in case triggered_count is already null i.e shared reset line
>>> + * is ready to be triggered.
> This is not a good idea IMHO. It would be better to document that calls
> to this function must be balanced with calls to reset_control_reset, and
> then throw a big warning below in case deassert_count ever dips below 0.
>
> Otherwise nothing stops drivers from silently decrementing other
> driver's trigger count by accidentally calling this multiple times.

I do agree, accidental calls should be reported by warnings.

>>> + *
>>> + * Consumers must not use reset_control_(de)assert on shared reset lines when
>>> + * reset_control_reset has been used.
>>> + *
>>> + * If rstc is NULL it is an optional clear and the function will just
>>> + * return 0.
>>> + */
>>> +int reset_control_resettable(struct reset_control *rstc)
>>> +{
>>> +       if (!rstc)
>>> +               return 0;
>>> +
>>> +       if (WARN_ON(IS_ERR(rstc)))
>>> +               return -EINVAL;
>>> +
>>> +       if (reset_control_is_array(rstc))
>>> +               return reset_control_array_resettable(rstc_to_array(rstc));
>>> +
>>> +       if (rstc->shared) {
>>> +               if (WARN_ON(atomic_read(&rstc->deassert_count) != 0))
>>> +                       return -EINVAL;
>>> +
>>> +               if (atomic_read(&rstc->triggered_count) > 0)
>>> +                       atomic_dec(&rstc->triggered_count);
> I think this should be
>
> 		WARN_ON(atomic_dec_return(&rstc->triggered_count) < 0);

That is even better, having this warning means that the call has not

been properly used.

>
> regards
> Philipp

Next version of the patch will be sent soon, include everything we

have discussed.

Sincerely,

Amjad

Re: [PATCH] reset: Add reset controller API

[Kevin Hilman]

Amjad Ouled-Ameur <aouledameur@baylibre.com> writes:

> The current reset framework API does not allow to release what is done by
> reset_control_reset(), IOW decrement triggered_count. Add the new
> reset_control_resettable() call to do so.
>
> When reset_control_reset() has been called once, the counter
> triggered_count, in the reset framework, is incremented i.e the resource
> under the reset is in-use and the reset should not be done again.
> reset_control_resettable() would be the way to state that the resource is 
> no longer used and, that from the caller's perspective, the reset can be 
> fired again if necessary.
>
> This patch will fix a usb suspend warning seen on the libretech-cc
> related to the shared reset line. This warning was addressed by the 
> previously reverted commit 7a410953d1fb ("usb: dwc3: meson-g12a: fix shared
> reset control use")

Could you also send a patch that shows how your new feature can be used
to fix the problem that was originally fixed by that patch (and still
exists, now that it was reverted.)

Thanks,

Kevin

Re: [PATCH] reset: Add reset controller API

[Amjad Ouled-Ameur]

On 03/10/2020 01:00, Kevin Hilman wrote:
> Amjad Ouled-Ameur <aouledameur@baylibre.com> writes:
>
>> The current reset framework API does not allow to release what is done by
>> reset_control_reset(), IOW decrement triggered_count. Add the new
>> reset_control_resettable() call to do so.
>>
>> When reset_control_reset() has been called once, the counter
>> triggered_count, in the reset framework, is incremented i.e the resource
>> under the reset is in-use and the reset should not be done again.
>> reset_control_resettable() would be the way to state that the resource is
>> no longer used and, that from the caller's perspective, the reset can be
>> fired again if necessary.
>>
>> This patch will fix a usb suspend warning seen on the libretech-cc
>> related to the shared reset line. This warning was addressed by the
>> previously reverted commit 7a410953d1fb ("usb: dwc3: meson-g12a: fix shared
>> reset control use")
> Could you also send a patch that shows how your new feature can be used
> to fix the problem that was originally fixed by that patch (and still
> exists, now that it was reverted.)
>
> Thanks,
>
> Kevin

Hello Kevin,


Will do soon !


Sincerely,

Amjad