Skip to contentSkip to conversation

opp: Don't print error message if getting optional regulator fails

PATCH In review Heiner Kallweit 2 participants 4 messages Mar 6, 2022 – Mar 8, 2022 Operating performance points (OPP)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 opp: Don't print error message if getting optional regulator fails

The regulators are optional, therefore I see no need to bother users with an error level message if -ENODEV is returned.

Inspiration was the following error on my system:
lima d00c0000.gpu: dev_pm_opp_set_regulators: no regulator (mali) found: -19

---
 drivers/opp/core.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/opp/core.c +4 -1
index 740407252..8af5979cc 100644
@@ -2020,7 +2020,10 @@ struct opp_table *dev_pm_opp_set_regulators(struct device *dev,
 		reg = regulator_get_optional(dev, names[i]);
 		if (IS_ERR(reg)) {
 			ret = PTR_ERR(reg);
-			if (ret != -EPROBE_DEFER)
+			if (ret == -ENODEV)
+				dev_info(dev, "%s: no regulator (%s) found\n",
+					 __func__, names[i]);
+			else if (ret != -EPROBE_DEFER)
 				dev_err(dev, "%s: no regulator (%s) found: %d\n",
 					__func__, names[i], ret);
 			goto free_regulators;
Signature · 1 line
2.35.1
#1 lore raw

drivers/opp/core.c struct opp_table *dev_pm_opp_set_regulators(struct device *dev,

Quoted 16 lines · Heiner Kallweit

The regulators are optional, therefore I see no need to bother users with an error level message if -ENODEV is returned.

Inspiration was the following error on my system:
lima d00c0000.gpu: dev_pm_opp_set_regulators: no regulator (mali) found: -19

---
 drivers/opp/core.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/opp/core.c +0 -0
index 740407252..8af5979cc 100644
@@ -2020,7 +2020,10 @@ struct opp_table *dev_pm_opp_set_regulators(struct device *dev,

If this API is called by the platform specific code, then the regulator should be compulsory. Isn't it ?

Maybe we shouldn't use regulator_get_optional() here.

Quoted 11 lines
drivers/opp/core.c +4 -1
 		reg = regulator_get_optional(dev, names[i]);
 		if (IS_ERR(reg)) {
 			ret = PTR_ERR(reg);
-			if (ret != -EPROBE_DEFER)
+			if (ret == -ENODEV)
+				dev_info(dev, "%s: no regulator (%s) found\n",
+					 __func__, names[i]);
+			else if (ret != -EPROBE_DEFER)
 				dev_err(dev, "%s: no regulator (%s) found: %d\n",
 					__func__, names[i], ret);
 			goto free_regulators;
Signature · 1 line
viresh
#2 lore raw
Quoted 21 lines · Viresh Kumar

drivers/opp/core.c struct opp_table *dev_pm_opp_set_regulators(struct device *dev,

Quoted 16 lines · Heiner Kallweit

The regulators are optional, therefore I see no need to bother users with an error level message if -ENODEV is returned.

Inspiration was the following error on my system:
lima d00c0000.gpu: dev_pm_opp_set_regulators: no regulator (mali) found: -19

---
 drivers/opp/core.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/opp/core.c +0 -0
index 740407252..8af5979cc 100644
@@ -2020,7 +2020,10 @@ struct opp_table *dev_pm_opp_set_regulators(struct device *dev,

If this API is called by the platform specific code, then the regulator should be compulsory. Isn't it ?

Quoted 2 lines

Maybe we shouldn't use regulator_get_optional() here.

There are two types of users of dev(m)_pm_opp_set_regulators():
1. Caller ignores -ENODEV and goes on.
2. Caller treats -ENODEV like any other error and bails out.

For type 1 callers a missing regulator should not result in an error message. Switching to regulator_get() wouldn't be too nice because then a missing regulator would result in the dummy regulator being used. This would create a related warning and a functional change that may break something.

Type 2 callers still have the option to print an own error message in case of -ENODEV. AFAICS all of them do so already.

Quoted 12 lines
Quoted 11 lines
drivers/opp/core.c +4 -1
 		reg = regulator_get_optional(dev, names[i]);
 		if (IS_ERR(reg)) {
 			ret = PTR_ERR(reg);
-			if (ret != -EPROBE_DEFER)
+			if (ret == -ENODEV)
+				dev_info(dev, "%s: no regulator (%s) found\n",
+					 __func__, names[i]);
+			else if (ret != -EPROBE_DEFER)
 				dev_err(dev, "%s: no regulator (%s) found: %d\n",
 					__func__, names[i], ret);
 			goto free_regulators;
#3 lore raw
Quoted 2 lines · Heiner Kallweit

There are two types of users of dev(m)_pm_opp_set_regulators():
1. Caller ignores -ENODEV and goes on.

This is abuse of the API as far as I can tell. If the OPP core was doing something like this automatically for everyone, then we can agree that missing regulator isn't necessarily a problem.

But in this case the platforms (lima and panfrost) are asking explicitly to get the regulators added. To me this is a plain and simple error.

Take cpufreq-dt for example, it makes sure beforehand if the supply is available or not, and then only calls this API.

Quoted 7 lines

For type 1 callers a missing regulator should not result in an error message. Switching to regulator_get() wouldn't be too nice because then a missing regulator would result in the dummy regulator being used. This would create a related warning and a functional change that may break something.

Type 2 callers still have the option to print an own error message in case of -ENODEV. AFAICS all of them do so already.

Signature · 1 line
viresh