[PATCH 0/6] backlight: Fix Critical and High issues reported by Sashiko

[PATCH 0/6] backlight: Fix Critical and High issues reported by Sashiko

[Lee Jones]

These issues were reported by Sashiko.

This series resolves the open Critical and High severity bug reports
identified by the Sashiko AI analysis tool (https://sashiko.dev) across
the Backlight subsystem:

  1. lm3630a_bl: Prevent a NULL pointer dereference in lm3630a_pwm_ctrl()
     when PWM mode is enabled by requesting the PWM device before
     registering the backlight device.
  2. aat2870_bl: Prevent a NULL pointer dereference on platform_get_drvdata()
     during devm_backlight_device_register() by populating driver data and
     scaling the initial brightness beforehand.
  3. qcom-wled: Cancel wled->ovp_work on probe failure via
     devm_delayed_work_autocancel() so a scheduled OVP work item cannot
     fire after devm_kzalloc() frees struct wled.
  4. qcom-wled: Manage wled->lock via devm_mutex_init() and remove
     wled_remove() so devres unwinds the backlight device, threaded IRQs,
     delayed work, and mutex in strict reverse registration order without
     calling disable_irq() on unconfigured negative or zero IRQ numbers.
  5. qcom-wled: Guard against calling a NULL
     wled->wled_auto_detection_required() callback on WLED3 hardware when
     an OVP fault interrupt fires.
  6. qcom-wled: Avoid an AB-BA deadlock between wled->lock and the
     threaded OVP IRQ handler by using disable_irq_nosync() in
     wled_module_enable() and verifying under wled->lock that the module
     remains enabled before running wled_auto_string_detection().

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Bjorn Andersson <andersson@kernel.org>
Cc: Daniel Jeong <gshark.jeong@gmail.com>
Cc: Daniel Thompson <danielt@kernel.org>
Cc: David Heidelberg <david@ixit.cz>
Cc: dri-devel@lists.freedesktop.org
Cc: Helge Deller <deller@gmx.de>
Cc: Jingoo Han <jingoohan1@gmail.com>
Cc: Jin Park <jinyoungp@nvidia.com>
Cc: Kiran Gunda <quic_kgunda@quicinc.com>
Cc: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Cc: LDD MLP <ldd-mlp@list.ti.com>
Cc: linux-arm-msm@vger.kernel.org
Cc: linux-fbdev@vger.kernel.org
Cc: Samuel Ortiz <sameo@linux.intel.com>
Cc: Subbaraman Narayanamurthy <quic_subbaram@quicinc.com>

Lee Jones (6):
  backlight: lm3630a: Request PWM device before registering backlight
  backlight: aat2870: Initialise driver data before registering
    backlight
  backlight: qcom-wled: Cancel delayed work on probe failure
  backlight: qcom-wled: Manage teardown via devres and remove
    wled_remove()
  backlight: qcom-wled: Check for auto-detection callback in
    wled_ovp_irq_handler()
  backlight: qcom-wled: Avoid AB-BA deadlock in wled_module_enable()

 drivers/video/backlight/aat2870_bl.c | 43 +++++++++++++------------
 drivers/video/backlight/lm3630a_bl.c | 12 +++----
 drivers/video/backlight/qcom-wled.c  | 47 +++++++++++++++-------------
 3 files changed, 52 insertions(+), 50 deletions(-)

-- 
2.56.0.385.gac671c819a-goog

[PATCH 3/6] backlight: qcom-wled: Cancel delayed work on probe failure

[Lee Jones]

If a short interrupt fires after wled_configure_short_irq() has
registered the interrupt handler, wled_short_irq_handler() calls
wled_module_enable(), which schedules wled->ovp_work.

If wled_probe() subsequently fails during
devm_backlight_device_register(), devres frees the wled structure while
wled->ovp_work remains pending, leading to a use-after-free when
wled_ovp_work() executes.

Configure the OVP interrupt first, then initialise wled->ovp_work via
devm_delayed_work_autocancel() before requesting the short interrupt.
This ensures that on teardown devres frees the short interrupt first,
cancels any pending wled->ovp_work while wled->ovp_irq is still valid,
and finally frees the OVP interrupt and wled structure.

Fixes: 8663c188beea ("backlight: qcom-wled: Add auto string detection logic")
Closes: https://sashiko.dev/#/bug/linux-ca2e48a2-f511-44b8-a7d4-866f360bf107
Cc: stable@vger.kernel.org
Signed-off-by: Lee Jones <lee@kernel.org>
---
Cc: David Heidelberg <david@ixit.cz>
Cc: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

 drivers/video/backlight/qcom-wled.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/drivers/video/backlight/qcom-wled.c b/drivers/video/backlight/qcom-wled.c
index ca350f320e17..f8f41e1d6ccf 100644
--- a/drivers/video/backlight/qcom-wled.c
+++ b/drivers/video/backlight/qcom-wled.c
@@ -3,6 +3,7 @@
  */
 
 #include <linux/delay.h>
+#include <linux/devm-helpers.h>
 #include <linux/interrupt.h>
 #include <linux/ktime.h>
 #include <linux/kernel.h>
@@ -1784,8 +1785,6 @@ static int wled_probe(struct platform_device *pdev)
 		break;
 	}
 
-	INIT_DELAYED_WORK(&wled->ovp_work, wled_ovp_work);
-
 	val = WLED_DEFAULT_BRIGHTNESS;
 	of_property_read_u32(pdev->dev.of_node, "default-brightness", &val);
 
@@ -1800,14 +1799,19 @@ static int wled_probe(struct platform_device *pdev)
 
 	wled->brightness = rc;
 
-	rc = wled_configure_short_irq(wled, pdev);
-	if (rc < 0)
-		return rc;
-
 	rc = wled_configure_ovp_irq(wled, pdev);
 	if (rc < 0)
 		return rc;
 
+	rc = devm_delayed_work_autocancel(&pdev->dev, &wled->ovp_work,
+					  wled_ovp_work);
+	if (rc)
+		return rc;
+
+	rc = wled_configure_short_irq(wled, pdev);
+	if (rc < 0)
+		return rc;
+
 	memset(&props, 0, sizeof(struct backlight_properties));
 	props.type = BACKLIGHT_RAW;
 	props.brightness = wled->brightness ?: val;
-- 
2.56.0.385.gac671c819a-goog

Re: [PATCH 3/6] backlight: qcom-wled: Cancel delayed work on probe failure

[Konrad Dybcio]

On 11/4/26 7:39 PM, Lee Jones wrote:
> If a short interrupt fires after wled_configure_short_irq() has
> registered the interrupt handler, wled_short_irq_handler() calls
> wled_module_enable(), which schedules wled->ovp_work.
> 
> If wled_probe() subsequently fails during
> devm_backlight_device_register(), devres frees the wled structure while
> wled->ovp_work remains pending, leading to a use-after-free when
> wled_ovp_work() executes.
> 
> Configure the OVP interrupt first, then initialise wled->ovp_work via
> devm_delayed_work_autocancel() before requesting the short interrupt.
> This ensures that on teardown devres frees the short interrupt first,
> cancels any pending wled->ovp_work while wled->ovp_irq is still valid,
> and finally frees the OVP interrupt and wled structure.
> 
> Fixes: 8663c188beea ("backlight: qcom-wled: Add auto string detection logic")
> Closes: https://sashiko.dev/#/bug/linux-ca2e48a2-f511-44b8-a7d4-866f360bf107
> Cc: stable@vger.kernel.org
> Signed-off-by: Lee Jones <lee@kernel.org>
> ---

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

Konrad

[PATCH 4/6] backlight: qcom-wled: Manage teardown via devres and remove wled_remove()

[Lee Jones]

wled_remove() currently runs before devres tears down the backlight
device and threaded interrupts, resulting in two teardown bugs:

1. mutex_destroy(&wled->lock) and cancel_delayed_work_sync(&wled->ovp_work)
   execute while the backlight device registered via
   devm_backlight_device_register() and the short interrupt requested
   via devm_request_threaded_irq() are still active.  A concurrent sysfs
   brightness update or short interrupt during unbind can acquire the
   destroyed mutex and re-schedule wled->ovp_work before devres frees
   the wled structure.

2. disable_irq(wled->short_irq) and disable_irq(wled->ovp_irq) are
   called unconditionally even when the optional "short" or "ovp"
   interrupts are absent from the Device Tree (-ENXIO) or unused (0 on
   WLED3), passing invalid IRQ numbers to disable_irq(), despite both
   interrupts already being managed by devm_request_threaded_irq().

Initialise wled->lock with devm_mutex_init() and remove wled_remove()
so that devres tears down the backlight device, short interrupt,
delayed work, OVP interrupt, mutex, and driver data in reverse
registration order.

Fixes: feeab87b3072 ("backlight: qcom-wled: Add support for short circuit handling")
Fixes: 8663c188beea ("backlight: qcom-wled: Add auto string detection logic")
Closes: https://sashiko.dev/#/bug/linux-51cd5e35-3cae-4e4c-986b-7261263507cd
Closes: https://sashiko.dev/#/bug/linux-44e5ed80-c35c-4786-8bf8-9ee955967d0b
Cc: stable@vger.kernel.org
Signed-off-by: Lee Jones <lee@kernel.org>
---
Cc: David Heidelberg <david@ixit.cz>
Cc: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

 drivers/video/backlight/qcom-wled.c | 17 ++++-------------
 1 file changed, 4 insertions(+), 13 deletions(-)

diff --git a/drivers/video/backlight/qcom-wled.c b/drivers/video/backlight/qcom-wled.c
index f8f41e1d6ccf..8d4d7fe5e84d 100644
--- a/drivers/video/backlight/qcom-wled.c
+++ b/drivers/video/backlight/qcom-wled.c
@@ -1732,7 +1732,6 @@ static int wled_probe(struct platform_device *pdev)
 
 	wled->regmap = regmap;
 	wled->dev = &pdev->dev;
-	platform_set_drvdata(pdev, wled);
 
 	wled->version = (uintptr_t)of_device_get_match_data(&pdev->dev);
 	if (!wled->version) {
@@ -1740,7 +1739,10 @@ static int wled_probe(struct platform_device *pdev)
 		return -ENODEV;
 	}
 
-	mutex_init(&wled->lock);
+	rc = devm_mutex_init(&pdev->dev, &wled->lock);
+	if (rc)
+		return rc;
+
 	rc = wled_configure(wled);
 	if (rc)
 		return rc;
@@ -1822,16 +1824,6 @@ static int wled_probe(struct platform_device *pdev)
 	return PTR_ERR_OR_ZERO(bl);
 };
 
-static void wled_remove(struct platform_device *pdev)
-{
-	struct wled *wled = platform_get_drvdata(pdev);
-
-	mutex_destroy(&wled->lock);
-	cancel_delayed_work_sync(&wled->ovp_work);
-	disable_irq(wled->short_irq);
-	disable_irq(wled->ovp_irq);
-}
-
 static const struct of_device_id wled_match_table[] = {
 	{ .compatible = "qcom,pm8941-wled", .data = (void *)3 },
 	{ .compatible = "qcom,pmi8950-wled", .data = (void *)4 },
@@ -1846,7 +1838,6 @@ MODULE_DEVICE_TABLE(of, wled_match_table);
 
 static struct platform_driver wled_driver = {
 	.probe = wled_probe,
-	.remove = wled_remove,
 	.driver	= {
 		.name = "qcom,wled",
 		.of_match_table	= wled_match_table,
-- 
2.56.0.385.gac671c819a-goog

Re: [PATCH 4/6] backlight: qcom-wled: Manage teardown via devres and remove wled_remove()

[Konrad Dybcio]

On 11/4/26 7:39 PM, Lee Jones wrote:
> wled_remove() currently runs before devres tears down the backlight
> device and threaded interrupts, resulting in two teardown bugs:
> 
> 1. mutex_destroy(&wled->lock) and cancel_delayed_work_sync(&wled->ovp_work)
>    execute while the backlight device registered via
>    devm_backlight_device_register() and the short interrupt requested
>    via devm_request_threaded_irq() are still active.  A concurrent sysfs
>    brightness update or short interrupt during unbind can acquire the
>    destroyed mutex and re-schedule wled->ovp_work before devres frees
>    the wled structure.
> 
> 2. disable_irq(wled->short_irq) and disable_irq(wled->ovp_irq) are
>    called unconditionally even when the optional "short" or "ovp"
>    interrupts are absent from the Device Tree (-ENXIO) or unused (0 on
>    WLED3), passing invalid IRQ numbers to disable_irq(), despite both
>    interrupts already being managed by devm_request_threaded_irq().
> 
> Initialise wled->lock with devm_mutex_init() and remove wled_remove()
> so that devres tears down the backlight device, short interrupt,
> delayed work, OVP interrupt, mutex, and driver data in reverse
> registration order.
> 
> Fixes: feeab87b3072 ("backlight: qcom-wled: Add support for short circuit handling")
> Fixes: 8663c188beea ("backlight: qcom-wled: Add auto string detection logic")
> Closes: https://sashiko.dev/#/bug/linux-51cd5e35-3cae-4e4c-986b-7261263507cd
> Closes: https://sashiko.dev/#/bug/linux-44e5ed80-c35c-4786-8bf8-9ee955967d0b
> Cc: stable@vger.kernel.org
> Signed-off-by: Lee Jones <lee@kernel.org>
> ---

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

Konrad

[PATCH 5/6] backlight: qcom-wled: Check for auto-detection callback in wled_ovp_irq_handler()

[Lee Jones]

The wled_auto_detection_required callback is only populated for WLED4
and WLED5 hardware, remaining NULL on WLED3 (where automatic string
detection is not supported).

However, wled_configure_ovp_irq() still registers wled_ovp_irq_handler()
on WLED3 when an "ovp" interrupt is present in the Device Tree.  If an
OVP fault interrupt fires on WLED3, wled_ovp_irq_handler()
unconditionally invokes wled->wled_auto_detection_required(), causing a
NULL pointer dereference.

Verify that wled->wled_auto_detection_required is non-NULL before
invoking it in wled_ovp_irq_handler().

Fixes: f16899a6a7d4 ("backlight: qcom-wled: Add callback functions")
Closes: https://sashiko.dev/#/bug/linux-d3eba287-2af9-43fb-851f-78ab3df0f35b
Cc: stable@vger.kernel.org
Signed-off-by: Lee Jones <lee@kernel.org>
---
Cc: David Heidelberg <david@ixit.cz>
Cc: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

 drivers/video/backlight/qcom-wled.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/video/backlight/qcom-wled.c b/drivers/video/backlight/qcom-wled.c
index 8d4d7fe5e84d..06300d403f8f 100644
--- a/drivers/video/backlight/qcom-wled.c
+++ b/drivers/video/backlight/qcom-wled.c
@@ -917,7 +917,8 @@ static irqreturn_t wled_ovp_irq_handler(int irq, void *_wled)
 			int_sts, fault_sts);
 
 	if (fault_sts & WLED3_CTRL_REG_OVP_FAULT_BIT) {
-		if (wled->wled_auto_detection_required(wled)) {
+		if (wled->wled_auto_detection_required &&
+		    wled->wled_auto_detection_required(wled)) {
 			mutex_lock(&wled->lock);
 			wled_auto_string_detection(wled);
 			mutex_unlock(&wled->lock);
-- 
2.56.0.385.gac671c819a-goog

Re: [PATCH 5/6] backlight: qcom-wled: Check for auto-detection callback in wled_ovp_irq_handler()

[Konrad Dybcio]

On 11/4/26 7:39 PM, Lee Jones wrote:
> The wled_auto_detection_required callback is only populated for WLED4
> and WLED5 hardware, remaining NULL on WLED3 (where automatic string
> detection is not supported).
> 
> However, wled_configure_ovp_irq() still registers wled_ovp_irq_handler()
> on WLED3 when an "ovp" interrupt is present in the Device Tree.  If an
> OVP fault interrupt fires on WLED3, wled_ovp_irq_handler()
> unconditionally invokes wled->wled_auto_detection_required(), causing a
> NULL pointer dereference.
> 
> Verify that wled->wled_auto_detection_required is non-NULL before
> invoking it in wled_ovp_irq_handler().
> 
> Fixes: f16899a6a7d4 ("backlight: qcom-wled: Add callback functions")
> Closes: https://sashiko.dev/#/bug/linux-d3eba287-2af9-43fb-851f-78ab3df0f35b
> Cc: stable@vger.kernel.org
> Signed-off-by: Lee Jones <lee@kernel.org>
> ---

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

Konrad

[PATCH 6/6] backlight: qcom-wled: Avoid AB-BA deadlock in wled_module_enable()

[Lee Jones]

When disabling the WLED module (for example from wled_update_status() or
wled_short_irq_handler()), the caller holds wled->lock and invokes
wled_module_enable(wled, false), which calls disable_irq(wled->ovp_irq)
and blocks waiting for any executing OVP threaded interrupt handler to
complete.

If wled_ovp_irq_handler() is concurrently executing, it attempts to
acquire wled->lock before running wled_auto_string_detection(),
resulting in an AB-BA deadlock between wled->lock and IRQ completion.

Use disable_irq_nosync() instead of disable_irq() in
wled_module_enable() so that the OVP interrupt is masked without
synchronously waiting for the threaded handler while holding wled->lock.
Because wled_ovp_irq_handler() may already be waiting on wled->lock when
the module is disabled, verify under wled->lock that the module is still
enabled and not disabled by a short before running
wled_auto_string_detection().

Fixes: 8663c188beea ("backlight: qcom-wled: Add auto string detection logic")
Closes: https://sashiko.dev/#/bug/linux-3afa151c-7b8f-432c-8f85-8c43a83b964a
Cc: stable@vger.kernel.org
Signed-off-by: Lee Jones <lee@kernel.org>
---
Cc: David Heidelberg <david@ixit.cz>
Cc: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

 drivers/video/backlight/qcom-wled.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/video/backlight/qcom-wled.c b/drivers/video/backlight/qcom-wled.c
index 06300d403f8f..6c0f631cc975 100644
--- a/drivers/video/backlight/qcom-wled.c
+++ b/drivers/video/backlight/qcom-wled.c
@@ -367,7 +367,7 @@ static int wled_module_enable(struct wled *wled, int val)
 			schedule_delayed_work(&wled->ovp_work, HZ / 100);
 		} else {
 			if (!cancel_delayed_work_sync(&wled->ovp_work))
-				disable_irq(wled->ovp_irq);
+				disable_irq_nosync(wled->ovp_irq);
 		}
 	}
 
@@ -919,8 +919,15 @@ static irqreturn_t wled_ovp_irq_handler(int irq, void *_wled)
 	if (fault_sts & WLED3_CTRL_REG_OVP_FAULT_BIT) {
 		if (wled->wled_auto_detection_required &&
 		    wled->wled_auto_detection_required(wled)) {
+			u32 val;
+
 			mutex_lock(&wled->lock);
-			wled_auto_string_detection(wled);
+			if (!wled->disabled_by_short) {
+				rc = regmap_read(wled->regmap, wled->ctrl_addr +
+						 WLED3_CTRL_REG_MOD_EN, &val);
+				if (!rc && (val & WLED3_CTRL_REG_MOD_EN_MASK))
+					wled_auto_string_detection(wled);
+			}
 			mutex_unlock(&wled->lock);
 		}
 	}
-- 
2.56.0.385.gac671c819a-goog

Re: [PATCH 6/6] backlight: qcom-wled: Avoid AB-BA deadlock in wled_module_enable()

[Konrad Dybcio]

On 11/4/26 7:39 PM, Lee Jones wrote:
> When disabling the WLED module (for example from wled_update_status() or
> wled_short_irq_handler()), the caller holds wled->lock and invokes
> wled_module_enable(wled, false), which calls disable_irq(wled->ovp_irq)
> and blocks waiting for any executing OVP threaded interrupt handler to
> complete.
> 
> If wled_ovp_irq_handler() is concurrently executing, it attempts to
> acquire wled->lock before running wled_auto_string_detection(),
> resulting in an AB-BA deadlock between wled->lock and IRQ completion.
> 
> Use disable_irq_nosync() instead of disable_irq() in
> wled_module_enable() so that the OVP interrupt is masked without
> synchronously waiting for the threaded handler while holding wled->lock.
> Because wled_ovp_irq_handler() may already be waiting on wled->lock when
> the module is disabled, verify under wled->lock that the module is still
> enabled and not disabled by a short before running
> wled_auto_string_detection().
> 
> Fixes: 8663c188beea ("backlight: qcom-wled: Add auto string detection logic")
> Closes: https://sashiko.dev/#/bug/linux-3afa151c-7b8f-432c-8f85-8c43a83b964a
> Cc: stable@vger.kernel.org
> Signed-off-by: Lee Jones <lee@kernel.org>
> ---

Acked-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

Konrad