On DT platforms, if a regulator init_data is provided in config, it is
silently ignored in favor of the DT parsing done by the framework, if
of_match is set.

of_match is an indication that init_data is expected to be set based on DT
and the parsing should be done by the regulator framework.

If the regulator provider passed init_data it must be because it is useful
somehow, in such case of_match should be clear.

If the driver expects the framework to initialize this data on its
own, it should leave init_data clear.

Warn if both init_data and of_match are set, then default to the provided
init_data.

Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
---
 drivers/regulator/core.c | 57 +++++++++++++++++++++++++++++-------------------
 1 file changed, 34 insertions(+), 23 deletions(-)

diff --git a/drivers/regulator/core.c b/drivers/regulator/core.c
index d0b3879f2746..a58a9db3d9c7 100644
--- a/drivers/regulator/core.c
+++ b/drivers/regulator/core.c
@@ -5681,32 +5681,43 @@ regulator_register(struct device *dev,
 		goto clean;
 	}
 
-	init_data = regulator_of_get_init_data(dev, regulator_desc, config,
-					       &rdev->dev.of_node);
-
-	/*
-	 * Sometimes not all resources are probed already so we need to take
-	 * that into account. This happens most the time if the ena_gpiod comes
-	 * from a gpio extender or something else.
-	 */
-	if (PTR_ERR(init_data) == -EPROBE_DEFER) {
-		ret = -EPROBE_DEFER;
-		goto clean;
-	}
+	if (config->init_data) {
+		/*
+		 * Providing of_match means the framework is expected to parse
+		 * DT to get the init_data. This would conflict with provided
+		 * init_data, if set. Warn if it happens.
+		 */
+		if (regulator_desc->of_match)
+			dev_warn(dev, "Using provided init data - OF match ignored\n");
 
-	/*
-	 * We need to keep track of any GPIO descriptor coming from the
-	 * device tree until we have handled it over to the core. If the
-	 * config that was passed in to this function DOES NOT contain
-	 * a descriptor, and the config after this call DOES contain
-	 * a descriptor, we definitely got one from parsing the device
-	 * tree.
-	 */
-	if (!cfg->ena_gpiod && config->ena_gpiod)
-		dangling_of_gpiod = true;
-	if (!init_data) {
 		init_data = config->init_data;
 		rdev->dev.of_node = of_node_get(config->of_node);
+
+	} else {
+		init_data = regulator_of_get_init_data(dev, regulator_desc,
+						       config,
+						       &rdev->dev.of_node);
+
+		/*
+		 * Sometimes not all resources are probed already so we need to
+		 * take that into account. This happens most the time if the
+		 * ena_gpiod comes from a gpio extender or something else.
+		 */
+		if (PTR_ERR(init_data) == -EPROBE_DEFER) {
+			ret = -EPROBE_DEFER;
+			goto clean;
+		}
+
+		/*
+		 * We need to keep track of any GPIO descriptor coming from the
+		 * device tree until we have handled it over to the core. If the
+		 * config that was passed in to this function DOES NOT contain a
+		 * descriptor, and the config after this call DOES contain a
+		 * descriptor, we definitely got one from parsing the device
+		 * tree.
+		 */
+		if (!cfg->ena_gpiod && config->ena_gpiod)
+			dangling_of_gpiod = true;
 	}
 
 	ww_mutex_init(&rdev->mutex, &regulator_ww_class);

-- 
2.45.2

On dinsdag 8 oktober 2024 18:07:01 Midden-Europese standaardtijd Jerome Brunet wrote:
> On DT platforms, if a regulator init_data is provided in config, it is
> silently ignored in favor of the DT parsing done by the framework, if
> of_match is set.
> 
> of_match is an indication that init_data is expected to be set based on DT
> and the parsing should be done by the regulator framework.
> 
> If the regulator provider passed init_data it must be because it is useful
> somehow, in such case of_match should be clear.
> 
> If the driver expects the framework to initialize this data on its
> own, it should leave init_data clear.
> 
> Warn if both init_data and of_match are set, then default to the provided
> init_data.

Hi Jerome,

This commit is breaking USB on qcom-msm8974-lge-nexus5-hammerhead for me.

I can't easily provide the full log since USB is breaking with this but in
effect it looks like in drivers/usb/chipidea/core.c in ci_get_platdata()
the call devm_regulator_get_optional(dev, "vbus"); is always returning
EPROBE_DEFER, so USB never initializes.

This vbus regulator is provided by ti,bq24192 so the bq24190_charger.c
driver. While the driver does seem to probe correctly, I do also see that
warning "Using provided init data - OF match ignored" in dmesg.

Reverting this patch on top of v6.13.2 fixes the issue and makes USB work
again.

Regards
Luca

> 
> Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
> ---
>  drivers/regulator/core.c | 57 +++++++++++++++++++++++++++++-------------------
>  1 file changed, 34 insertions(+), 23 deletions(-)
> 
> diff --git a/drivers/regulator/core.c b/drivers/regulator/core.c
> index d0b3879f2746..a58a9db3d9c7 100644
> --- a/drivers/regulator/core.c
> +++ b/drivers/regulator/core.c
> @@ -5681,32 +5681,43 @@ regulator_register(struct device *dev,
>  		goto clean;
>  	}
>  
> -	init_data = regulator_of_get_init_data(dev, regulator_desc, config,
> -					       &rdev->dev.of_node);
> -
> -	/*
> -	 * Sometimes not all resources are probed already so we need to take
> -	 * that into account. This happens most the time if the ena_gpiod comes
> -	 * from a gpio extender or something else.
> -	 */
> -	if (PTR_ERR(init_data) == -EPROBE_DEFER) {
> -		ret = -EPROBE_DEFER;
> -		goto clean;
> -	}
> +	if (config->init_data) {
> +		/*
> +		 * Providing of_match means the framework is expected to parse
> +		 * DT to get the init_data. This would conflict with provided
> +		 * init_data, if set. Warn if it happens.
> +		 */
> +		if (regulator_desc->of_match)
> +			dev_warn(dev, "Using provided init data - OF match ignored\n");
>  
> -	/*
> -	 * We need to keep track of any GPIO descriptor coming from the
> -	 * device tree until we have handled it over to the core. If the
> -	 * config that was passed in to this function DOES NOT contain
> -	 * a descriptor, and the config after this call DOES contain
> -	 * a descriptor, we definitely got one from parsing the device
> -	 * tree.
> -	 */
> -	if (!cfg->ena_gpiod && config->ena_gpiod)
> -		dangling_of_gpiod = true;
> -	if (!init_data) {
>  		init_data = config->init_data;
>  		rdev->dev.of_node = of_node_get(config->of_node);
> +
> +	} else {
> +		init_data = regulator_of_get_init_data(dev, regulator_desc,
> +						       config,
> +						       &rdev->dev.of_node);
> +
> +		/*
> +		 * Sometimes not all resources are probed already so we need to
> +		 * take that into account. This happens most the time if the
> +		 * ena_gpiod comes from a gpio extender or something else.
> +		 */
> +		if (PTR_ERR(init_data) == -EPROBE_DEFER) {
> +			ret = -EPROBE_DEFER;
> +			goto clean;
> +		}
> +
> +		/*
> +		 * We need to keep track of any GPIO descriptor coming from the
> +		 * device tree until we have handled it over to the core. If the
> +		 * config that was passed in to this function DOES NOT contain a
> +		 * descriptor, and the config after this call DOES contain a
> +		 * descriptor, we definitely got one from parsing the device
> +		 * tree.
> +		 */
> +		if (!cfg->ena_gpiod && config->ena_gpiod)
> +			dangling_of_gpiod = true;
>  	}
>  
>  	ww_mutex_init(&rdev->mutex, &regulator_ww_class);
> 
>

On Sun 09 Feb 2025 at 15:16, Luca Weiss <luca@lucaweiss.eu> wrote:

> On dinsdag 8 oktober 2024 18:07:01 Midden-Europese standaardtijd Jerome Brunet wrote:
>> On DT platforms, if a regulator init_data is provided in config, it is
>> silently ignored in favor of the DT parsing done by the framework, if
>> of_match is set.
>> 
>> of_match is an indication that init_data is expected to be set based on DT
>> and the parsing should be done by the regulator framework.
>> 
>> If the regulator provider passed init_data it must be because it is useful
>> somehow, in such case of_match should be clear.
>> 
>> If the driver expects the framework to initialize this data on its
>> own, it should leave init_data clear.
>> 
>> Warn if both init_data and of_match are set, then default to the provided
>> init_data.
>
> Hi Jerome,
>
> This commit is breaking USB on qcom-msm8974-lge-nexus5-hammerhead for me.
>
> I can't easily provide the full log since USB is breaking with this but in
> effect it looks like in drivers/usb/chipidea/core.c in ci_get_platdata()
> the call devm_regulator_get_optional(dev, "vbus"); is always returning
> EPROBE_DEFER, so USB never initializes.

Sorry about that.

>
> This vbus regulator is provided by ti,bq24192 so the bq24190_charger.c
> driver. While the driver does seem to probe correctly, I do also see that
> warning "Using provided init data - OF match ignored" in dmesg.

Well the driver does this funny thing of passing init_data but also
setting the of_match.

Looking at it more, your driver provide init_data/constraint here:

https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/drivers/power/supply/bq24190_charger.c?h=v6.12.13#n718

Apparently this constraint is only meant as a backup/default in case
nothing is matched by the platform, which would explain why your
platform through the trouble of passing an empty regulator node to
override it.

https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/arch/arm/boot/dts/qcom/qcom-msm8974-lge-nexus5-hammerhead.dts#n116

I did not really expect that but it seems intended indeed.
Revert is probably the sane thing to do but it would be nice to have
comment about that and maybe a debug print.

Mark, do you want to revert this directly or shall I submit the change
for you to apply ?

>
> Reverting this patch on top of v6.13.2 fixes the issue and makes USB work
> again.
>
> Regards
> Luca
>
>> 
>> Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
>> ---
>>  drivers/regulator/core.c | 57 +++++++++++++++++++++++++++++-------------------
>>  1 file changed, 34 insertions(+), 23 deletions(-)
>> 
>> diff --git a/drivers/regulator/core.c b/drivers/regulator/core.c
>> index d0b3879f2746..a58a9db3d9c7 100644
>> --- a/drivers/regulator/core.c
>> +++ b/drivers/regulator/core.c
>> @@ -5681,32 +5681,43 @@ regulator_register(struct device *dev,
>>  		goto clean;
>>  	}
>>  
>> -	init_data = regulator_of_get_init_data(dev, regulator_desc, config,
>> -					       &rdev->dev.of_node);
>> -
>> -	/*
>> -	 * Sometimes not all resources are probed already so we need to take
>> -	 * that into account. This happens most the time if the ena_gpiod comes
>> -	 * from a gpio extender or something else.
>> -	 */
>> -	if (PTR_ERR(init_data) == -EPROBE_DEFER) {
>> -		ret = -EPROBE_DEFER;
>> -		goto clean;
>> -	}
>> +	if (config->init_data) {
>> +		/*
>> +		 * Providing of_match means the framework is expected to parse
>> +		 * DT to get the init_data. This would conflict with provided
>> +		 * init_data, if set. Warn if it happens.
>> +		 */
>> +		if (regulator_desc->of_match)
>> +			dev_warn(dev, "Using provided init data - OF match ignored\n");
>>  
>> -	/*
>> -	 * We need to keep track of any GPIO descriptor coming from the
>> -	 * device tree until we have handled it over to the core. If the
>> -	 * config that was passed in to this function DOES NOT contain
>> -	 * a descriptor, and the config after this call DOES contain
>> -	 * a descriptor, we definitely got one from parsing the device
>> -	 * tree.
>> -	 */
>> -	if (!cfg->ena_gpiod && config->ena_gpiod)
>> -		dangling_of_gpiod = true;
>> -	if (!init_data) {
>>  		init_data = config->init_data;
>>  		rdev->dev.of_node = of_node_get(config->of_node);
>> +
>> +	} else {
>> +		init_data = regulator_of_get_init_data(dev, regulator_desc,
>> +						       config,
>> +						       &rdev->dev.of_node);
>> +
>> +		/*
>> +		 * Sometimes not all resources are probed already so we need to
>> +		 * take that into account. This happens most the time if the
>> +		 * ena_gpiod comes from a gpio extender or something else.
>> +		 */
>> +		if (PTR_ERR(init_data) == -EPROBE_DEFER) {
>> +			ret = -EPROBE_DEFER;
>> +			goto clean;
>> +		}
>> +
>> +		/*
>> +		 * We need to keep track of any GPIO descriptor coming from the
>> +		 * device tree until we have handled it over to the core. If the
>> +		 * config that was passed in to this function DOES NOT contain a
>> +		 * descriptor, and the config after this call DOES contain a
>> +		 * descriptor, we definitely got one from parsing the device
>> +		 * tree.
>> +		 */
>> +		if (!cfg->ena_gpiod && config->ena_gpiod)
>> +			dangling_of_gpiod = true;
>>  	}
>>  
>>  	ww_mutex_init(&rdev->mutex, &regulator_ww_class);
>> 
>> 

-- 
Jerome

On Sun, Feb 09, 2025 at 08:05:05PM +0100, Jerome Brunet wrote:

> Mark, do you want to revert this directly or shall I submit the change
> for you to apply ?

Could someone please send a patch describing the change.