Skip to contentSkip to conversation

iio: PM-enable leak on probe error path (goto-variant drivers)

PATCH In review Stepan Ionichev 2 participants 10 messages May 29 IIO

lore
Share & tools
lore
#0 lore raw

cover iio: PM-enable leak on probe error path (goto-variant drivers)

This is the goto-variant follow-up to the standalone ti-ads1015 patch and the 4-patch series for the simple drivers. In these four drivers probe calls pm_runtime_enable() but the existing register-failure path jumps to a manual cleanup label (out_err / buffer_cleanup / error_power_down) rather than returning directly. None of those cleanup labels call pm_runtime_disable(), so the runtime PM enable_count leaks on probe failure and on subsequent rebind.

The conversion is the same as in the rest of the class: replace pm_runtime_enable() with devm_pm_runtime_enable() and drop the pm_runtime_disable() / pm_runtime_set_suspended() calls in .remove(). On the new error path from devm_pm_runtime_enable() each patch jumps to the same existing cleanup label that iio_device_register() uses, so the manual unwind for the non-devm resources held earlier in probe still runs. In apds9960 nothing is powered or registered at that point so a plain return is used.

Standalone precedent already on the list:
iio: adc: ti-ads1015 -
https://lore.kernel.org/all/[email protected]/

Companion 4-patch series for the simple-return drivers (isl29028, tsl2583, mlx90614, bmi088-accel) was posted alongside this one.

Stepan Ionichev (4):
iio: light: pa12203001: use devm_pm_runtime_enable() to fix probe
error path
iio: light: us5182d: use devm_pm_runtime_enable() to fix probe error
path
iio: light: apds9960: use devm_pm_runtime_enable() to fix probe error
path
iio: accel: mma8452: use devm_pm_runtime_enable() to fix probe error
path

 drivers/iio/accel/mma8452.c    | 7 +++----
 drivers/iio/light/apds9960.c   | 6 +++---
 drivers/iio/light/pa12203001.c | 7 +++----
 drivers/iio/light/us5182d.c    | 7 +++----
 4 files changed, 12 insertions(+), 15 deletions(-)

base-commit: 7cb1c5b32a2bfde961fff8d5204526b609bcb30a

Signature · 1 line
2.43.0
#1 lore raw

PATCH 1/4 iio: light: pa12203001: use devm_pm_runtime_enable() to fix probe error path

pa12203001_probe() calls pm_runtime_enable() and then iio_device_register(). If iio_device_register() fails the function jumps to out_err but the existing out_err handler does not call pm_runtime_disable(), leaking the runtime PM enable_count on probe failure and on subsequent rebind.

Switch to devm_pm_runtime_enable() so the enable (and the matching dont_use_autosuspend) are torn down automatically. On its new error path the probe jumps to the existing out_err label so the chip-disable cleanup stays in step with the rest of the manual unwind. The pm_runtime_disable() and pm_runtime_set_suspended() calls in pa12203001_remove() are dropped; the devm action runs after .remove() and handles the teardown.

---
 drivers/iio/light/pa12203001.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/iio/light/pa12203001.c +3 -4
index 98a1f1624c75..b9f50c9070a3 100644
@@ -372,7 +372,9 @@ static int pa12203001_probe(struct i2c_client *client)
 	if (ret < 0)
 		goto out_err;
 
-	pm_runtime_enable(&client->dev);
+	ret = devm_pm_runtime_enable(&client->dev);
+	if (ret)
+		goto out_err;
 	pm_runtime_set_autosuspend_delay(&client->dev,
 					 PA12203001_SLEEP_DELAY_MS);
 	pm_runtime_use_autosuspend(&client->dev);
@@ -395,9 +397,6 @@ static void pa12203001_remove(struct i2c_client *client)
 
 	iio_device_unregister(indio_dev);
 
-	pm_runtime_disable(&client->dev);
-	pm_runtime_set_suspended(&client->dev);
-
 	ret = pa12203001_power_chip(indio_dev, PA12203001_CHIP_DISABLE);
 	if (ret)
 		dev_warn(&client->dev, "Failed to power down (%pe)\n",
Signature · 1 line
2.43.0
#3 lore raw

PATCH 2/4 iio: light: us5182d: use devm_pm_runtime_enable() to fix probe error path

In reply to Stepan Ionichev

us5182d_probe() calls pm_runtime_enable() and then iio_device_register(). If iio_device_register() fails the function jumps to out_err but the existing out_err handler does not call pm_runtime_disable(), leaking the runtime PM enable_count on probe failure and on subsequent rebind.

Switch to devm_pm_runtime_enable() so the enable (and the matching dont_use_autosuspend) are torn down automatically. On its new error path the probe jumps to the existing out_err label so the shutdown_en() cleanup stays in step with the rest of the manual unwind. The pm_runtime_disable() and pm_runtime_set_suspended() calls in us5182d_remove() are dropped; the devm action runs after .remove() and handles the teardown.

---
 drivers/iio/light/us5182d.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/iio/light/us5182d.c +3 -4
index d2f5a44892a8..ada75f710bfc 100644
@@ -880,7 +880,9 @@ static int us5182d_probe(struct i2c_client *client)
 			goto out_err;
 	}
 
-	pm_runtime_enable(&client->dev);
+	ret = devm_pm_runtime_enable(&client->dev);
+	if (ret)
+		goto out_err;
 	pm_runtime_set_autosuspend_delay(&client->dev,
 					 US5182D_SLEEP_MS);
 	pm_runtime_use_autosuspend(&client->dev);
@@ -904,9 +906,6 @@ static void us5182d_remove(struct i2c_client *client)
 
 	iio_device_unregister(i2c_get_clientdata(client));
 
-	pm_runtime_disable(&client->dev);
-	pm_runtime_set_suspended(&client->dev);
-
 	ret = us5182d_shutdown_en(data, US5182D_CFG0_SHUTDOWN_EN);
 	if (ret)
 		dev_warn(&client->dev, "Failed to shut down (%pe)\n",
Signature · 1 line
2.43.0
#5 lore raw

PATCH 3/4 iio: light: apds9960: use devm_pm_runtime_enable() to fix probe error path

In reply to Stepan Ionichev

apds9960_probe() calls pm_runtime_enable() and then several operations that may fail through goto error_power_down, ending with iio_device_register(). None of those error paths call pm_runtime_disable(), so the runtime PM enable_count leaks on probe failure and on subsequent rebind.

Switch to devm_pm_runtime_enable() so the enable (and the matching dont_use_autosuspend) are torn down automatically. The probe error path on devm_pm_runtime_enable() itself is a plain return because no hardware has been powered on at that point. The pm_runtime_disable() and pm_runtime_set_suspended() calls in apds9960_remove() are dropped; the devm action runs after .remove() and handles the teardown.

---
 drivers/iio/light/apds9960.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/iio/light/apds9960.c +3 -3
index 785c5dbe2d08..edbf149b931e 100644
@@ -1074,7 +1074,9 @@ static int apds9960_probe(struct i2c_client *client)
 	if (ret)
 		goto error_power_down;
 
-	pm_runtime_enable(&client->dev);
+	ret = devm_pm_runtime_enable(&client->dev);
+	if (ret)
+		return ret;
 	pm_runtime_set_autosuspend_delay(&client->dev, 5000);
 	pm_runtime_use_autosuspend(&client->dev);
 
@@ -1123,8 +1125,6 @@ static void apds9960_remove(struct i2c_client *client)
 	struct apds9960_data *data = iio_priv(indio_dev);
 
 	iio_device_unregister(indio_dev);
-	pm_runtime_disable(&client->dev);
-	pm_runtime_set_suspended(&client->dev);
 	apds9960_set_powermode(data, 0);
 }
 
Signature · 1 line
2.43.0
#7 lore raw

PATCH 4/4 iio: accel: mma8452: use devm_pm_runtime_enable() to fix probe error path

In reply to Stepan Ionichev

mma8452_probe() calls pm_runtime_enable() and then iio_device_register(). If iio_device_register() fails the function jumps to buffer_cleanup but the existing manual unwind does not call pm_runtime_disable(), leaking the runtime PM enable_count on probe failure and on subsequent rebind.

Switch to devm_pm_runtime_enable() so the enable (and the matching dont_use_autosuspend) are torn down automatically. On its new error path the probe jumps to the existing buffer_cleanup label so the triggered buffer, trigger and regulator cleanup chain stays intact. The pm_runtime_disable() and pm_runtime_set_suspended() calls in mma8452_remove() are dropped; the devm action runs after .remove() and handles the teardown.

---
 drivers/iio/accel/mma8452.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/iio/accel/mma8452.c +3 -4
index 15172ba2972c..88ed47d33e03 100644
@@ -1693,7 +1693,9 @@ static int mma8452_probe(struct i2c_client *client)
 	if (ret < 0)
 		goto buffer_cleanup;
 
-	pm_runtime_enable(&client->dev);
+	ret = devm_pm_runtime_enable(&client->dev);
+	if (ret)
+		goto buffer_cleanup;
 	pm_runtime_set_autosuspend_delay(&client->dev,
 					 MMA8452_AUTO_SUSPEND_DELAY_MS);
 	pm_runtime_use_autosuspend(&client->dev);
@@ -1733,9 +1735,6 @@ static void mma8452_remove(struct i2c_client *client)
 
 	iio_device_unregister(indio_dev);
 
-	pm_runtime_disable(&client->dev);
-	pm_runtime_set_suspended(&client->dev);
-
 	iio_triggered_buffer_cleanup(indio_dev);
 	mma8452_trigger_cleanup(indio_dev);
 	mma8452_standby(iio_priv(indio_dev));
Signature · 1 line
2.43.0
maintainer
#9 lore raw
In reply to Stepan Ionichev
Quoted 8 lines · On Fri, 29 May 2026 15:45:40 +0500 Stepan Ionichev <[email protected]> wrote:

This is the goto-variant follow-up to the standalone ti-ads1015 patch and the 4-patch series for the simple drivers. In these four drivers probe calls pm_runtime_enable() but the existing register-failure path jumps to a manual cleanup label (out_err / buffer_cleanup / error_power_down) rather than returning directly. None of those cleanup labels call pm_runtime_disable(), so the runtime PM enable_count leaks on probe failure and on subsequent rebind.

Quoted 8 lines

The conversion is the same as in the rest of the class: replace pm_runtime_enable() with devm_pm_runtime_enable() and drop the pm_runtime_disable() / pm_runtime_set_suspended() calls in .remove(). On the new error path from devm_pm_runtime_enable() each patch jumps to the same existing cleanup label that iio_device_register() uses, so the manual unwind for the non-devm resources held earlier in probe still runs. In apds9960 nothing is powered or registered at that point so a plain return is used.

This is v2. Make sure you increment version numbers even for the sort of commit message reformat you made here. Sending as the same version confuses readers and potentially scripts.

Note this is a clarification not a request that you resend now as v2! J

Quoted 26 lines

Standalone precedent already on the list:
iio: adc: ti-ads1015 -
https://lore.kernel.org/all/[email protected]/

Companion 4-patch series for the simple-return drivers (isl29028, tsl2583, mlx90614, bmi088-accel) was posted alongside this one.

Stepan Ionichev (4):
iio: light: pa12203001: use devm_pm_runtime_enable() to fix probe
error path
iio: light: us5182d: use devm_pm_runtime_enable() to fix probe error
path
iio: light: apds9960: use devm_pm_runtime_enable() to fix probe error
path
iio: accel: mma8452: use devm_pm_runtime_enable() to fix probe error
path

 drivers/iio/accel/mma8452.c    | 7 +++----
 drivers/iio/light/apds9960.c   | 6 +++---
 drivers/iio/light/pa12203001.c | 7 +++----
 drivers/iio/light/us5182d.c    | 7 +++----
 4 files changed, 12 insertions(+), 15 deletions(-)

base-commit: 7cb1c5b32a2bfde961fff8d5204526b609bcb30a

maintainer
#2 lore raw

Re: [PATCH 1/4] iio: light: pa12203001: use devm_pm_runtime_enable() to fix probe error path

In reply to Stepan Ionichev
Quoted 23 lines · On Fri, 29 May 2026 15:45:41 +0500 Stepan Ionichev <[email protected]> wrote:

pa12203001_probe() calls pm_runtime_enable() and then iio_device_register(). If iio_device_register() fails the function jumps to out_err but the existing out_err handler does not call pm_runtime_disable(), leaking the runtime PM enable_count on probe failure and on subsequent rebind.

Switch to devm_pm_runtime_enable() so the enable (and the matching dont_use_autosuspend) are torn down automatically. On its new error path the probe jumps to the existing out_err label so the chip-disable cleanup stays in step with the rest of the manual unwind. The pm_runtime_disable() and pm_runtime_set_suspended() calls in pa12203001_remove() are dropped; the devm action runs after .remove() and handles the teardown.

---
 drivers/iio/light/pa12203001.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/iio/light/pa12203001.c +0 -0
index 98a1f1624c75..b9f50c9070a3 100644

drivers/iio/light/pa12203001.c static int pa12203001_probe(struct i2c_client *client) · line 377

Quoted 7 lines
drivers/iio/light/pa12203001.c +3 -1
@@ -372,7 +372,9 @@ static int pa12203001_probe(struct i2c_client *client)
 	if (ret < 0)
 		goto out_err;
 
-	pm_runtime_enable(&client->dev);
+	ret = devm_pm_runtime_enable(&client->dev);
+	if (ret)
+		goto out_err;

A goto after a devm call is a big warning flag.

A quick look suggests easy enough to solve with a devm_add_action_or_reset(); Probably needs to be a precursor patch to this one.

Quoted 13 lines
drivers/iio/light/pa12203001.c +0 -3
 	pm_runtime_set_autosuspend_delay(&client->dev,
 					 PA12203001_SLEEP_DELAY_MS);
 	pm_runtime_use_autosuspend(&client->dev);
@@ -395,9 +397,6 @@ static void pa12203001_remove(struct i2c_client *client)
 
 	iio_device_unregister(indio_dev);
 
-	pm_runtime_disable(&client->dev);
-	pm_runtime_set_suspended(&client->dev);
-
 	ret = pa12203001_power_chip(indio_dev, PA12203001_CHIP_DISABLE);
 	if (ret)
 		dev_warn(&client->dev, "Failed to power down (%pe)\n",
maintainer
#4 lore raw

Re: [PATCH 2/4] iio: light: us5182d: use devm_pm_runtime_enable() to fix probe error path

In reply to Stepan Ionichev
Quoted 5 lines · On Fri, 29 May 2026 15:45:42 +0500 Stepan Ionichev <[email protected]> wrote:

us5182d_probe() calls pm_runtime_enable() and then iio_device_register(). If iio_device_register() fails the function jumps to out_err but the existing out_err handler does not call pm_runtime_disable(), leaking the runtime PM enable_count on probe failure and on subsequent rebind.

Quoted 8 lines

Switch to devm_pm_runtime_enable() so the enable (and the matching dont_use_autosuspend) are torn down automatically. On its new error path the probe jumps to the existing out_err label so the shutdown_en() cleanup stays in step with the rest of the manual unwind. The pm_runtime_disable() and pm_runtime_set_suspended() calls in us5182d_remove() are dropped; the devm action runs after .remove() and handles the teardown.

Same a for patch 1. Needs more work so there is no goto.

Quoted 29 lines
---
 drivers/iio/light/us5182d.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/iio/light/us5182d.c +3 -4
index d2f5a44892a8..ada75f710bfc 100644
@@ -880,7 +880,9 @@ static int us5182d_probe(struct i2c_client *client)
 			goto out_err;
 	}
 
-	pm_runtime_enable(&client->dev);
+	ret = devm_pm_runtime_enable(&client->dev);
+	if (ret)
+		goto out_err;
 	pm_runtime_set_autosuspend_delay(&client->dev,
 					 US5182D_SLEEP_MS);
 	pm_runtime_use_autosuspend(&client->dev);
@@ -904,9 +906,6 @@ static void us5182d_remove(struct i2c_client *client)
 
 	iio_device_unregister(i2c_get_clientdata(client));
 
-	pm_runtime_disable(&client->dev);
-	pm_runtime_set_suspended(&client->dev);
-
 	ret = us5182d_shutdown_en(data, US5182D_CFG0_SHUTDOWN_EN);
 	if (ret)
 		dev_warn(&client->dev, "Failed to shut down (%pe)\n",
maintainer
#6 lore raw

Re: [PATCH 3/4] iio: light: apds9960: use devm_pm_runtime_enable() to fix probe error path

In reply to Stepan Ionichev
Quoted 23 lines · On Fri, 29 May 2026 15:45:43 +0500 Stepan Ionichev <[email protected]> wrote:

apds9960_probe() calls pm_runtime_enable() and then several operations that may fail through goto error_power_down, ending with iio_device_register(). None of those error paths call pm_runtime_disable(), so the runtime PM enable_count leaks on probe failure and on subsequent rebind.

Switch to devm_pm_runtime_enable() so the enable (and the matching dont_use_autosuspend) are torn down automatically. The probe error path on devm_pm_runtime_enable() itself is a plain return because no hardware has been powered on at that point. The pm_runtime_disable() and pm_runtime_set_suspended() calls in apds9960_remove() are dropped; the devm action runs after .remove() and handles the teardown.

---
 drivers/iio/light/apds9960.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/iio/light/apds9960.c +0 -0
index 785c5dbe2d08..edbf149b931e 100644

drivers/iio/light/apds9960.c static int apds9960_probe(struct i2c_client *client) · line 1079

Quoted 7 lines
drivers/iio/light/apds9960.c +3 -1
@@ -1074,7 +1074,9 @@ static int apds9960_probe(struct i2c_client *client)
 	if (ret)
 		goto error_power_down;
 
-	pm_runtime_enable(&client->dev);
+	ret = devm_pm_runtime_enable(&client->dev);
+	if (ret)
+		return ret;

Leaves the power on. And resolving that would have the same issue as the previous two patches with being a mix of goto and devm cleanup.

Very strong advice on this is the moment anything at all needs to be in remove() you have to stop using devm for remaining calls and unwind them by hand.

Quoted 12 lines
drivers/iio/light/apds9960.c +0 -2
 	pm_runtime_set_autosuspend_delay(&client->dev, 5000);
 	pm_runtime_use_autosuspend(&client->dev);
 
@@ -1123,8 +1125,6 @@ static void apds9960_remove(struct i2c_client *client)
 	struct apds9960_data *data = iio_priv(indio_dev);
 
 	iio_device_unregister(indio_dev);
-	pm_runtime_disable(&client->dev);
-	pm_runtime_set_suspended(&client->dev);
 	apds9960_set_powermode(data, 0);
 }
 
maintainer
#8 lore raw

Re: [PATCH 4/4] iio: accel: mma8452: use devm_pm_runtime_enable() to fix probe error path

In reply to Stepan Ionichev
Quoted 13 lines · On Fri, 29 May 2026 15:45:44 +0500 Stepan Ionichev <[email protected]> wrote:

mma8452_probe() calls pm_runtime_enable() and then iio_device_register(). If iio_device_register() fails the function jumps to buffer_cleanup but the existing manual unwind does not call pm_runtime_disable(), leaking the runtime PM enable_count on probe failure and on subsequent rebind.

Switch to devm_pm_runtime_enable() so the enable (and the matching dont_use_autosuspend) are torn down automatically. On its new error path the probe jumps to the existing buffer_cleanup label so the triggered buffer, trigger and regulator cleanup chain stays intact. The pm_runtime_disable() and pm_runtime_set_suspended() calls in mma8452_remove() are dropped; the devm action runs after .remove() and handles the teardown.

Quoted 1 line

Same again on not mixing devm with goto.

Side effect here is that the tear down order is completely different from reverse of setup.

Quoted 29 lines
---
 drivers/iio/accel/mma8452.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/iio/accel/mma8452.c +3 -4
index 15172ba2972c..88ed47d33e03 100644
@@ -1693,7 +1693,9 @@ static int mma8452_probe(struct i2c_client *client)
 	if (ret < 0)
 		goto buffer_cleanup;
 
-	pm_runtime_enable(&client->dev);
+	ret = devm_pm_runtime_enable(&client->dev);
+	if (ret)
+		goto buffer_cleanup;
 	pm_runtime_set_autosuspend_delay(&client->dev,
 					 MMA8452_AUTO_SUSPEND_DELAY_MS);
 	pm_runtime_use_autosuspend(&client->dev);
@@ -1733,9 +1735,6 @@ static void mma8452_remove(struct i2c_client *client)
 
 	iio_device_unregister(indio_dev);
 
-	pm_runtime_disable(&client->dev);
-	pm_runtime_set_suspended(&client->dev);
-
 	iio_triggered_buffer_cleanup(indio_dev);
 	mma8452_trigger_cleanup(indio_dev);
 	mma8452_standby(iio_priv(indio_dev));