Skip to contentSkip to conversation

regulator: qcom-rpmh: Add support for regulator-off-on-delay-us

PATCH Superseded Saikiran 3 participants 7 messages Jan 27 – Jan 28

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Saikiran

author
#0 lore raw

PATCH v2 1/1 regulator: qcom-rpmh: Add support for regulator-off-on-delay-us

The core regulator framework supports enforcing a physical off-time via standard properties, but the qcom-rpmh-regulator driver currently ignores them. This prevents boards with slow-discharging rails from enforcing safe power-cycling constraints.

On the Lenovo Yoga Slim 7x (Snapdragon X Elite), certain camera regulators rely on passive discharge and require a significant off-time to drop below brownout thresholds. Without this driver support, we cannot enforce this constraint via Device Tree, leading to sensor initialization failures during rapid power cycling.

Add support for parsing the regulator-off-on-delay-us property from the device tree.

I have tested this on the Yoga Slim 7x. When the delay property is present in the device tree, the regulator core correctly blocks re-enable calls until the delay passes, fixing the camera brownout issues.

---
 drivers/regulator/qcom-rpmh-regulator.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/regulator/qcom-rpmh-regulator.c +3 -0
index 6e4cb2871fca..aafba61551b3 100644
@@ -503,6 +503,9 @@ static int rpmh_regulator_init_vreg(struct rpmh_vreg *vreg, struct device *dev,
 	vreg->always_wait_for_ack = of_property_read_bool(node,
 						"qcom,always-wait-for-ack");
 
+	of_property_read_u32(node, "regulator-off-on-delay-us",
+			     &vreg->rdesc.off_on_delay);
+
 	vreg->rdesc.owner	= THIS_MODULE;
 	vreg->rdesc.type	= REGULATOR_VOLTAGE;
 	vreg->rdesc.ops		= vreg->hw_data->ops;
Signature · 1 line
2.51.0
#1 lore raw
Quoted 28 lines · Saikiran

The core regulator framework supports enforcing a physical off-time via standard properties, but the qcom-rpmh-regulator driver currently ignores them. This prevents boards with slow-discharging rails from enforcing safe power-cycling constraints.

On the Lenovo Yoga Slim 7x (Snapdragon X Elite), certain camera regulators rely on passive discharge and require a significant off-time to drop below brownout thresholds. Without this driver support, we cannot enforce this constraint via Device Tree, leading to sensor initialization failures during rapid power cycling.

Add support for parsing the regulator-off-on-delay-us property from the device tree.

I have tested this on the Yoga Slim 7x. When the delay property is present in the device tree, the regulator core correctly blocks re-enable calls until the delay passes, fixing the camera brownout issues.

---
 drivers/regulator/qcom-rpmh-regulator.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/regulator/qcom-rpmh-regulator.c +0 -0
index 6e4cb2871fca..aafba61551b3 100644

drivers/regulator/qcom-rpmh-regulator.c static int rpmh_regulator_init_vreg(struct rpmh_vreg *vreg, struct device *dev, · line 507

Quoted 5 lines
drivers/regulator/qcom-rpmh-regulator.c +2 -0
@@ -503,6 +503,9 @@ static int rpmh_regulator_init_vreg(struct rpmh_vreg *vreg, struct device *dev,
 	vreg->always_wait_for_ack = of_property_read_bool(node,
 						"qcom,always-wait-for-ack");
 
+	of_property_read_u32(node, "regulator-off-on-delay-us",
+			     &vreg->rdesc.off_on_delay);

Would it not be a better fit for of_regulator.c?

Konrad

author
#2 lore raw
Quoted 33 lines · Konrad Dybcio
Quoted 31 lines · Saikiran

The core regulator framework supports enforcing a physical off-time via standard properties, but the qcom-rpmh-regulator driver currently ignores them. This prevents boards with slow-discharging rails from enforcing safe power-cycling constraints.

On the Lenovo Yoga Slim 7x (Snapdragon X Elite), certain camera regulators rely on passive discharge and require a significant off-time to drop below brownout thresholds. Without this driver support, we cannot enforce this constraint via Device Tree, leading to sensor initialization failures during rapid power cycling.

Add support for parsing the regulator-off-on-delay-us property from the device tree.

I have tested this on the Yoga Slim 7x. When the delay property is present in the device tree, the regulator core correctly blocks re-enable calls until the delay passes, fixing the camera brownout issues.

---
 drivers/regulator/qcom-rpmh-regulator.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/regulator/qcom-rpmh-regulator.c +0 -0
index 6e4cb2871fca..aafba61551b3 100644
@@ -503,6 +503,9 @@ static int rpmh_regulator_init_vreg(struct rpmh_vreg *vreg, struct device *dev,
      vreg->always_wait_for_ack = of_property_read_bool(node,
                                              "qcom,always-wait-for-ack");
 
Quoted 6 lines

drivers/regulator/qcom-rpmh-regulator.c static int rpmh_regulator_init_vreg(struct rpmh_vreg *vreg, struct device *dev, · line 507

Quoted 2 lines
drivers/regulator/qcom-rpmh-regulator.c +2 -0
+     of_property_read_u32(node, "regulator-off-on-delay-us",
+                          &vreg->rdesc.off_on_delay);

Would it not be a better fit for of_regulator.c?

Konrad

Hi Konrad,

Quoted 1 line · Konrad Dybcio

Would it not be a better fit for of_regulator.c?

That was my initial thought as well, but there is a limitation:

The off_on_delay field resides in struct regulator_desc. For most regulator drivers, this structure is static const (immutable) as it describes fixed silicon characteristics. Generic parsing code in of_regulator.c cannot blindly write to desc->off_on_delay without risking a write to read-only memory.

The qcom-rpmh-regulator driver is unique in that it allocates vreg->rdesc dynamically at runtime, which allows us to safely populate this field from DT.

To support this generically in of_regulator.c, we would likely need to introduce off_on_delay into struct regulator_constraints instead, and then update the core regulator handling to check both sources.

I opted for this driver-specific approach to minimize impact on the core subsystem, given that qcom-rpmh is already set up to handle dynamic descriptors.

Let me know if you would prefer I attempt the core framework change instead.

Regards,
Saikiran

#3 lore raw
Quoted 63 lines · Saikiran B
Quoted 33 lines · Konrad Dybcio
Quoted 31 lines · Saikiran

The core regulator framework supports enforcing a physical off-time via standard properties, but the qcom-rpmh-regulator driver currently ignores them. This prevents boards with slow-discharging rails from enforcing safe power-cycling constraints.

On the Lenovo Yoga Slim 7x (Snapdragon X Elite), certain camera regulators rely on passive discharge and require a significant off-time to drop below brownout thresholds. Without this driver support, we cannot enforce this constraint via Device Tree, leading to sensor initialization failures during rapid power cycling.

Add support for parsing the regulator-off-on-delay-us property from the device tree.

I have tested this on the Yoga Slim 7x. When the delay property is present in the device tree, the regulator core correctly blocks re-enable calls until the delay passes, fixing the camera brownout issues.

---
 drivers/regulator/qcom-rpmh-regulator.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/regulator/qcom-rpmh-regulator.c +0 -0
index 6e4cb2871fca..aafba61551b3 100644
@@ -503,6 +503,9 @@ static int rpmh_regulator_init_vreg(struct rpmh_vreg *vreg, struct device *dev,
      vreg->always_wait_for_ack = of_property_read_bool(node,
                                              "qcom,always-wait-for-ack");
 
Quoted 6 lines

drivers/regulator/qcom-rpmh-regulator.c static int rpmh_regulator_init_vreg(struct rpmh_vreg *vreg, struct device *dev, · line 507

Quoted 2 lines
drivers/regulator/qcom-rpmh-regulator.c +2 -0
+     of_property_read_u32(node, "regulator-off-on-delay-us",
+                          &vreg->rdesc.off_on_delay);

Would it not be a better fit for of_regulator.c?

Konrad

Hi Konrad,

Quoted 1 line · Konrad Dybcio

Would it not be a better fit for of_regulator.c?

That was my initial thought as well, but there is a limitation:

The off_on_delay field resides in struct regulator_desc. For most regulator drivers, this structure is static const (immutable) as it describes fixed silicon characteristics. Generic parsing code in of_regulator.c cannot blindly write to desc->off_on_delay without risking a write to read-only memory.

The qcom-rpmh-regulator driver is unique in that it allocates vreg->rdesc dynamically at runtime, which allows us to safely populate this field from DT.

To support this generically in of_regulator.c, we would likely need to introduce off_on_delay into struct regulator_constraints instead, and then update the core regulator handling to check both sources.

Quoted 5 lines

I opted for this driver-specific approach to minimize impact on the core subsystem, given that qcom-rpmh is already set up to handle dynamic descriptors.

Let me know if you would prefer I attempt the core framework change instead.

I'm a fly-by reviewer for this sort of thing (as you can see by me not knowing this reasoning..)

Mark (the maintainer) should be able to give you a more insightful answer

Konrad

maintainer
#4 lore raw
In reply to Saikiran B
Quoted 5 lines · Saikiran B

The off_on_delay field resides in struct regulator_desc. For most regulator drivers, this structure is static const (immutable) as it describes fixed silicon characteristics. Generic parsing code in of_regulator.c cannot blindly write to desc->off_on_delay without risking a write to read-only memory.

Your reading there is right, generally the expectation was that this should be a property of the regulator rather than the system. This case seems pretty unusual thus far but perhaps we'll see more such cases in future and should move the property but for now having it enabled per driver seems safer.

Quoted 7 lines

The qcom-rpmh-regulator driver is unique in that it allocates vreg->rdesc dynamically at runtime, which allows us to safely populate this field from DT.

To support this generically in of_regulator.c, we would likely need to introduce off_on_delay into struct regulator_constraints instead, and then update the core regulator handling to check both sources.

Yes, indeed - if we were supporting the property completely generically we'd just not have the driver fill in in the information and store the actual value used separately with the DT property overriding the driver value (possibly only if it was larger).

author
#5 lore raw
Quoted 19 lines · Mark Brown
Quoted 5 lines · Saikiran B

The off_on_delay field resides in struct regulator_desc. For most regulator drivers, this structure is static const (immutable) as it describes fixed silicon characteristics. Generic parsing code in of_regulator.c cannot blindly write to desc->off_on_delay without risking a write to read-only memory.

Your reading there is right, generally the expectation was that this should be a property of the regulator rather than the system. This case seems pretty unusual thus far but perhaps we'll see more such cases in future and should move the property but for now having it enabled per driver seems safer.

Quoted 3 lines

The qcom-rpmh-regulator driver is unique in that it allocates vreg->rdesc dynamically at runtime, which allows us to safely populate this field from DT.

Quoted 8 lines
Quoted 3 lines

To support this generically in of_regulator.c, we would likely need to introduce off_on_delay into struct regulator_constraints instead, and then update the core regulator handling to check both sources.

Yes, indeed - if we were supporting the property completely generically we'd just not have the driver fill in in the information and store the actual value used separately with the DT property overriding the driver value (possibly only if it was larger).

Thanks for the confirmation, Mark.

Since the driver-specific approach is architecturally acceptable, I will keep this implementation in mind as a fallback.

However, to address the root cause of the camera brownout once and for all, I am now investigating porting the Active Discharge logic (sending specific voltage/mode resource commands) from similar Qualcomm sister boards into qcom-rpmh-regulator.

That would be the superior hardware solution, eliminating the need for arbitrary delays. Since that involves complex RPMh resource handling and reverse-engineering the specific commands for this PMIC, it will take some time to implement and validate.

I will hold off on pursuing this delay patch further until I have exhausted the Active Discharge path.

I thank you and konrad for providing the feedback.

Regards,
Saikiran

maintainer
#6 lore raw
Quoted 5 lines · Saikiran B

However, to address the root cause of the camera brownout once and for all, I am now investigating porting the Active Discharge logic (sending specific voltage/mode resource commands) from similar Qualcomm sister boards into qcom-rpmh-regulator.

Quoted 7 lines

That would be the superior hardware solution, eliminating the need for arbitrary delays. Since that involves complex RPMh resource handling and reverse-engineering the specific commands for this PMIC, it will take some time to implement and validate.

I will hold off on pursuing this delay patch further until I have exhausted the Active Discharge path.

Yes, active discharge would definitely be a better solution.