Date:   Wed, 13 Feb 2019 16:17:47 +0100
From:   Fabrice Gasnier <fabrice.gasnier@...com>
To:     Thierry Reding <thierry.reding@...il.com>
CC:     <robh+dt@...nel.org>, <u.kleine-koenig@...gutronix.de>,
        <tduszyns@...il.com>, <mark.rutland@....com>,
        <alexandre.torgue@...com>, <mcoquelin.stm32@...il.com>,
        <devicetree@...r.kernel.org>,
        <linux-arm-kernel@...ts.infradead.org>,
        <linux-kernel@...r.kernel.org>, <linux-pwm@...r.kernel.org>,
        <linux-stm32@...md-mailman.stormreply.com>
Subject: Re: [PATCH v3 3/3] pwm: core: add consumer device link

On 2/13/19 1:53 PM, Thierry Reding wrote:
> On Wed, Feb 13, 2019 at 11:50:12AM +0100, Fabrice Gasnier wrote:
>> Add a device link between the PWM consumer and the PWM provider. This
>> enforces the PWM user to get suspended before the PWM provider. It
>> allows proper synchronization of suspend/resume sequences: the PWM user
>> is responsible for properly stopping PWM, before the provider gets
>> suspended: see [1]. Add the device link in:
>> - of_pwm_get()
>> - pwm_get()
>> - devm_ variants
>> as it requires a reference to the device for the PWM consumer.
>>
>> [1] https://lkml.org/lkml/2019/2/5/770
>>
>> Suggested-by: Thierry Reding <thierry.reding@...il.com>
>> Signed-off-by: Fabrice Gasnier <fabrice.gasnier@...com>
>> ---
>> Changes in v3:
>> - add struct device to of_get_pwm() arguments to handle device link from
>>   there.
>> ---
>>  drivers/pwm/core.c  | 14 +++++++++++---
>>  include/linux/pwm.h |  6 ++++--
>>  2 files changed, 15 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/pwm/core.c b/drivers/pwm/core.c
>> index 1581f6a..8cb5d4bc 100644
>> --- a/drivers/pwm/core.c
>> +++ b/drivers/pwm/core.c
>> @@ -638,6 +638,7 @@ static struct pwm_chip *of_node_to_pwmchip(struct device_node *np)
>>  
>>  /**
>>   * of_pwm_get() - request a PWM via the PWM framework
>> + * @dev: device for PWM consumer
>>   * @np: device node to get the PWM from
>>   * @con_id: consumer name
>>   *
>> @@ -655,7 +656,8 @@ static struct pwm_chip *of_node_to_pwmchip(struct device_node *np)
>>   * Returns: A pointer to the requested PWM device or an ERR_PTR()-encoded
>>   * error code on failure.
>>   */
>> -struct pwm_device *of_pwm_get(struct device_node *np, const char *con_id)
>> +struct pwm_device *of_pwm_get(struct device *dev, struct device_node *np,
>> +			      const char *con_id)
>>  {
>>  	struct pwm_device *pwm = NULL;
>>  	struct of_phandle_args args;
>> @@ -689,6 +691,9 @@ struct pwm_device *of_pwm_get(struct device_node *np, const char *con_id)
>>  	if (IS_ERR(pwm))
>>  		goto put;
>>  
>> +	if (!device_link_add(dev, pwm->chip->dev, DL_FLAG_AUTOREMOVE_CONSUMER))
>> +		pr_debug("%s(): device link not added\n", __func__);
> 
> I think it's better to turn this into dev_dbg(dev, ...) and maybe
> mention which supplier it failed to link to, something like:
> 
> 	if (!device_link_add(...))
> 		dev_dbg(dev, "failed to create device link to %s\n",
> 			pwm->chip->dev);

Hi Thierry,

Thanks for reviewing.

I can update this: I used pr_debug() as there are pr_err() calls
elsewhere in this routine.
BTW, do you wish an additional patch to turn pr_err() into dev_err() in
of_pwm_get()?

> 
> Also, I wonder if this should perhaps be dev_err(). Under what
> circumstances does this fail?

Well, here is a comment from "device_link_add()" routine:
"
/*
 * If the supplier has not been fully registered yet or there is a
 * reverse dependency between the consumer and the supplier already in
 * the graph, return NULL.
 */
"

=> Here the PWM supplier is already registered. (It seems a probe defer
can be returned few lines above otherwise.)

Other possibilities: kzalloc() failed, no consumer or supplier has been
provided (or invalid flags, but this is hardcoded here.).

So, I see two case here:
1 - The caller provided a 'dev' for PWM consumer... So, NULL link is an
error when consumer & supplier has been passed correctly.
=> I can add a check on 'dev' for PWM consumer and report an error here:
   return -EINVAL

2 - The caller can't provide a 'dev' for PWM consumer as you mention
bellow: "to allow code to get at the PWM if they didn't have..."
=> We should probably add a dev_warn() here, with no error ?
Please see here after.

> 
>> +
>>  	/*
>>  	 * If a consumer name was not given, try to look it up from the
>>  	 * "pwm-names" property if it exists. Otherwise use the name of
>> @@ -771,7 +776,7 @@ struct pwm_device *pwm_get(struct device *dev, const char *con_id)
>>  
>>  	/* look up via DT first */
>>  	if (IS_ENABLED(CONFIG_OF) && dev && dev->of_node)
>> -		return of_pwm_get(dev->of_node, con_id);
>> +		return of_pwm_get(dev, dev->of_node, con_id);
>>  
>>  	/*
>>  	 * We look up the provider in the static table typically provided by
>> @@ -851,6 +856,9 @@ struct pwm_device *pwm_get(struct device *dev, const char *con_id)
>>  	pwm->args.period = chosen->period;
>>  	pwm->args.polarity = chosen->polarity;
>>  
>> +	if (!device_link_add(dev, pwm->chip->dev, DL_FLAG_AUTOREMOVE_CONSUMER))
>> +		pr_debug("%s(): device link not added\n", __func__);
> 
> Same here. Also: not sure if we really need to include __func__ in the
> message.
> 
>> +
>>  	return pwm;
>>  }
>>  EXPORT_SYMBOL_GPL(pwm_get);
>> @@ -939,7 +947,7 @@ struct pwm_device *devm_of_pwm_get(struct device *dev, struct device_node *np,
>>  	if (!ptr)
>>  		return ERR_PTR(-ENOMEM);
>>  
>> -	pwm = of_pwm_get(np, con_id);
>> +	pwm = of_pwm_get(dev, np, con_id);
>>  	if (!IS_ERR(pwm)) {
>>  		*ptr = pwm;
>>  		devres_add(dev, ptr);
>> diff --git a/include/linux/pwm.h b/include/linux/pwm.h
>> index d5199b5..895e074 100644
>> --- a/include/linux/pwm.h
>> +++ b/include/linux/pwm.h
>> @@ -406,7 +406,8 @@ struct pwm_device *of_pwm_xlate_with_flags(struct pwm_chip *pc,
>>  		const struct of_phandle_args *args);
>>  
>>  struct pwm_device *pwm_get(struct device *dev, const char *con_id);
>> -struct pwm_device *of_pwm_get(struct device_node *np, const char *con_id);
>> +struct pwm_device *of_pwm_get(struct device *dev, struct device_node *np,
>> +			      const char *con_id);
> 
> I'm slightly concerned about this. I think one of the reasons why this
> was introduced was to allow code to get at the PWM if they didn't have
> a struct device * available. However, it doesn't seem like there are any
> users of that function, so this seems fine.

The git blame pointed out commit 8eb961279960:
"
pwm: core: Rename of_pwm_request() to of_pwm_get() and export it

Allow client driver to use of_pwm_get() to get the PWM they need. This
is needed for drivers which handle more than one PWM separately, like
leds-pwm driver, which have:
"
...

For instance, I tested the leds-pwm driver. It uses the devm_* variant
now (as others), there is a struct device * available. So yes, it seems
fine.

The only thing maybe out of tree code? This is where I have a doubt on
having a mandatory struct device * to enforce consumer link creation...
or make it optional (e.g. behave as a 'legacy' API) and warn the caller.

Please let me know your feeling.
Best regards,
Fabrice

> 
> Thierry
> 
>>  void pwm_put(struct pwm_device *pwm);
>>  
>>  struct pwm_device *devm_pwm_get(struct device *dev, const char *con_id);
>> @@ -494,7 +495,8 @@ static inline struct pwm_device *pwm_get(struct device *dev,
>>  	return ERR_PTR(-ENODEV);
>>  }
>>  
>> -static inline struct pwm_device *of_pwm_get(struct device_node *np,
>> +static inline struct pwm_device *of_pwm_get(struct device *dev,
>> +					    struct device_node *np,
>>  					    const char *con_id)
>>  {
>>  	return ERR_PTR(-ENODEV);
>> -- 
>> 1.9.1
>>

Powered by blists - more mailing lists