Skip to contentSkip to conversation

leds: lp5860: fix init error-path locking

PATCH Superseded [email protected] 3 participants 3 messages Jul 20 – Aug 6 LED

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 leds: lp5860: fix init error-path locking

From: surendra <[email protected]>

lp5860_device_init() jumps to err_disable with the mutex held when regmap_update_bits() fails. The error handler calls lp5860_chip_enable(), which tries to acquire the same mutex and deadlocks.

Conversely, an lp5860_init_dt() failure reaches the same handler after the mutex has been released, causing mutex_unlock() to operate on an unlocked mutex.

Release the mutex immediately after the register update and keep the error handler lockless.

---
 drivers/leds/rgb/leds-lp5860-core.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/leds/rgb/leds-lp5860-core.c +1 -2
index fd0e2f6e6e0f..e21d5f2302be 100644
@@ -204,9 +204,9 @@ int lp5860_device_init(struct device *dev)
 	mutex_lock(&lp->lock);
 	ret = regmap_update_bits(lp->regmap, LP5860_REG_DEV_INITIAL, LP5860_MODE_MASK,
 				 LP5860_MODE_1 << LP5860_MODE_SHIFT);
+	mutex_unlock(&lp->lock);
 	if (ret)
 		goto err_disable;
-	mutex_unlock(&lp->lock);
 
 	ret = lp5860_init_dt(lp);
 	if (ret)
@@ -215,7 +215,6 @@ int lp5860_device_init(struct device *dev)
 	return 0;
 
 err_disable:
-	mutex_unlock(&lp->lock);
 	lp5860_chip_enable(lp, LP5860_CHIP_DISABLE);
 	return ret;
 }
Signature · 1 line
2.55.0
#1 lore raw
Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

New issues:
- [Medium] The commit message falsely claims a deadlock existed on the regmap_update_bits() error path.
- [Low] The commit message prefix and description violate subsystem capitalization guidelines.

Pre-existing issues:
- [High] The driver destroys its primary mutex in .remove() before executing hardware teardown and devres cleanup, causing multiple use-after-destroy locking violations.
- [High] LEDs are unregistered via devres after the chip is manually disabled, risking invalid hardware accesses.
--

--- Patch [1]: [PATCH] leds: lp5860: fix init error-path locking --- commit 2979577a641c15da3d90ccaf3a97200e7e476dbb
Author: surendra <[email protected]>

leds: lp5860: fix init error-path locking

This commit addresses a locking issue in the initialization error path for the lp5860 driver. It releases the lock immediately after the regmap update, avoiding potential lock mismanagement in the subsequent error paths.

Quoted 1 line

leds: lp5860: fix init error-path locking

[Severity: Low]
According to the LED subsystem naming conventions, should the description after the subsystem prefix be capitalized? So "fix" would become "Fix".

Quoted 7 lines

lp5860_device_init() jumps to err_disable with the mutex held when regmap_update_bits() fails. The error handler calls lp5860_chip_enable(), which tries to acquire the same mutex and deadlocks.

Conversely, an lp5860_init_dt() failure reaches the same handler after the mutex has been released, causing mutex_unlock() to operate on an unlocked mutex.

[Severity: Medium]
Is this description of a deadlock accurate?

Looking at the original code, the err_disable block explicitly executed a mutex_unlock() before calling lp5860_chip_enable():

err_disable:

	mutex_unlock(&lp->lock);
	lp5860_chip_enable(lp, LP5860_CHIP_DISABLE);
	return ret;

Since lp5860_chip_enable() uses guard(mutex)(&lp->lock), it safely re-locks the just-released mutex. Could the commit message be updated to clarify that this fixes a double-unlock bug when lp5860_init_dt() fails, rather than a deadlock on the regmap_update_bits() path?

Quoted 16 lines
drivers/leds/rgb/leds-lp5860-core.c +1 -1
index fd0e2f6e6e0f4..e21d5f2302be0 100644
@@ -204,9 +204,9 @@ int lp5860_device_init(struct device *dev)
 	mutex_lock(&lp->lock);
 	ret = regmap_update_bits(lp->regmap, LP5860_REG_DEV_INITIAL, LP5860_MODE_MASK,
 				 LP5860_MODE_1 << LP5860_MODE_SHIFT);
+	mutex_unlock(&lp->lock);
 	if (ret)
 		goto err_disable;
-	mutex_unlock(&lp->lock);
 
 	ret = lp5860_init_dt(lp);
 	if (ret)

drivers/leds/rgb/leds-lp5860-core.c int lp5860_device_init(struct device *dev) · line 218

Quoted 5 lines
drivers/leds/rgb/leds-lp5860-core.c +0 -1
@@ -215,7 +215,6 @@ int lp5860_device_init(struct device *dev)
 	return 0;
 
 err_disable:
-	mutex_unlock(&lp->lock);
 	lp5860_chip_enable(lp, LP5860_CHIP_DISABLE);

[Severity: High]
This is a pre-existing issue, but does destroying the mutex in lp5860_remove() before calling hardware teardown lead to a use-after-destroy violation?

In the spi driver, we have:

lp5860_remove() {
	...
	mutex_destroy(&lp5860->lock);
	lp5860_device_remove(&spi->dev);
}

But lp5860_device_remove() calls lp5860_chip_enable(), which acquires the just-destroyed mutex:

lp5860_chip_enable() {
	guard(mutex)(&lp->lock);
	...
}

Will this cause a kernel crash when lock debugging is enabled?

[Severity: High]
This is also a pre-existing issue, but does unregistering the LEDs via devres after manually disabling the chip cause invalid hardware accesses?

The driver uses devm_led_classdev_multicolor_register_ext() to register LEDs, which handles unregistration automatically via devres during device removal. However, the manual lp5860_device_remove() call happens before devres cleanup:

lp5860_remove() {
	...
	lp5860_device_remove(&spi->dev);
}

The automatic devres cleanup will eventually set the brightness to 0, invoking lp5860_set_mc_brightness(). Since the chip was already disabled manually in lp5860_remove(), could this result in regmap I/O to disabled hardware over SPI? Should the hardware teardown be synchronized via a managed devres action instead?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
maintainer
#2 lore raw
Quoted 5 lines · On Mon, 20 Jul 2026, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

New issues:
- [Medium] The commit message falsely claims a deadlock existed on the regmap_update_bits() error path.
- [Low] The commit message prefix and description violate subsystem capitalization guidelines.

Please review, explain or fix the Sashiko review(s).

If this is a set, this request includes the other patches too.

Quoted 113 lines

Pre-existing issues:
- [High] The driver destroys its primary mutex in .remove() before executing hardware teardown and devres cleanup, causing multiple use-after-destroy locking violations.
- [High] LEDs are unregistered via devres after the chip is manually disabled, risking invalid hardware accesses.
--

--- Patch [1]: [PATCH] leds: lp5860: fix init error-path locking --- commit 2979577a641c15da3d90ccaf3a97200e7e476dbb
Author: surendra <[email protected]>

leds: lp5860: fix init error-path locking

This commit addresses a locking issue in the initialization error path for the lp5860 driver. It releases the lock immediately after the regmap update, avoiding potential lock mismanagement in the subsequent error paths.

Quoted 1 line

leds: lp5860: fix init error-path locking

[Severity: Low]
According to the LED subsystem naming conventions, should the description after the subsystem prefix be capitalized? So "fix" would become "Fix".

Quoted 7 lines

lp5860_device_init() jumps to err_disable with the mutex held when regmap_update_bits() fails. The error handler calls lp5860_chip_enable(), which tries to acquire the same mutex and deadlocks.

Conversely, an lp5860_init_dt() failure reaches the same handler after the mutex has been released, causing mutex_unlock() to operate on an unlocked mutex.

[Severity: Medium]
Is this description of a deadlock accurate?

Looking at the original code, the err_disable block explicitly executed a mutex_unlock() before calling lp5860_chip_enable():

err_disable:

	mutex_unlock(&lp->lock);
	lp5860_chip_enable(lp, LP5860_CHIP_DISABLE);
	return ret;

Since lp5860_chip_enable() uses guard(mutex)(&lp->lock), it safely re-locks the just-released mutex. Could the commit message be updated to clarify that this fixes a double-unlock bug when lp5860_init_dt() fails, rather than a deadlock on the regmap_update_bits() path?

Quoted 16 lines
drivers/leds/rgb/leds-lp5860-core.c +1 -1
index fd0e2f6e6e0f4..e21d5f2302be0 100644
@@ -204,9 +204,9 @@ int lp5860_device_init(struct device *dev)
 	mutex_lock(&lp->lock);
 	ret = regmap_update_bits(lp->regmap, LP5860_REG_DEV_INITIAL, LP5860_MODE_MASK,
 				 LP5860_MODE_1 << LP5860_MODE_SHIFT);
+	mutex_unlock(&lp->lock);
 	if (ret)
 		goto err_disable;
-	mutex_unlock(&lp->lock);
 
 	ret = lp5860_init_dt(lp);
 	if (ret)

drivers/leds/rgb/leds-lp5860-core.c int lp5860_device_init(struct device *dev) · line 218

Quoted 5 lines
drivers/leds/rgb/leds-lp5860-core.c +0 -1
@@ -215,7 +215,6 @@ int lp5860_device_init(struct device *dev)
 	return 0;
 
 err_disable:
-	mutex_unlock(&lp->lock);
 	lp5860_chip_enable(lp, LP5860_CHIP_DISABLE);

[Severity: High]
This is a pre-existing issue, but does destroying the mutex in lp5860_remove() before calling hardware teardown lead to a use-after-destroy violation?

In the spi driver, we have:

lp5860_remove() {
	...
	mutex_destroy(&lp5860->lock);
	lp5860_device_remove(&spi->dev);
}

But lp5860_device_remove() calls lp5860_chip_enable(), which acquires the just-destroyed mutex:

lp5860_chip_enable() {
	guard(mutex)(&lp->lock);
	...
}

Will this cause a kernel crash when lock debugging is enabled?

[Severity: High]
This is also a pre-existing issue, but does unregistering the LEDs via devres after manually disabling the chip cause invalid hardware accesses?

The driver uses devm_led_classdev_multicolor_register_ext() to register LEDs, which handles unregistration automatically via devres during device removal. However, the manual lp5860_device_remove() call happens before devres cleanup:

lp5860_remove() {
	...
	lp5860_device_remove(&spi->dev);
}

The automatic devres cleanup will eventually set the brightness to 0, invoking lp5860_set_mc_brightness(). Since the chip was already disabled manually in lp5860_remove(), could this result in regmap I/O to disabled hardware over SPI? Should the hardware teardown be synchronized via a managed devres action instead?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Signature · 1 line
Lee Jones