Re: [PATCH] platform/loongarch: laptop: Fix backlight device lifetime on unload

From: Luo Qiu

Date: Fri Nov 20 2026 - 19:45:51 EST



On 11/20/26 6:16 PM, Huacai Chen wrote:
On Fri, Nov 20, 2026 at 5:13 PM Luo Qiu <luoqiu@xxxxxxxxxxxxxxx> wrote:

On 11/20/26 4:16 PM, Huacai Chen wrote:
Hi, Qiu,

On Thu, Nov 19, 2026 at 8:22 PM Luo Qiu <luoqiu@xxxxxxxxxxxxxxx> wrote:
The driver registers a platform backlight device during module init but
discards the returned pointer. Consequently, the device is never
unregistered when the module is removed, leaving its ops callbacks
pointing into freed module text. After module removal, reading the
backlight's actual_brightness attribute can call the stale callback and
crash the kernel. Registration failures are also silently ignored.

Store the returned pointer, use it for resume, and unregister it on
module exit. Propagate registration failures to avoid silently
continuing without a backlight device.

Fixes: 6246ed09111f ("LoongArch: Add ACPI-based generic laptop driver")
Signed-off-by: Luo Qiu <luoqiu@xxxxxxxxxxxxxxx>
---
drivers/platform/loongarch/loongson-laptop.c | 19 +++++++++++++------
1 file changed, 13 insertions(+), 6 deletions(-)

diff --git a/drivers/platform/loongarch/loongson-laptop.c b/drivers/platform/loongarch/loongson-laptop.c
index 4ec56621db24..ec0742e5ed46 100644
--- a/drivers/platform/loongarch/loongson-laptop.c
+++ b/drivers/platform/loongarch/loongson-laptop.c
@@ -56,6 +56,7 @@ static acpi_handle hotkey_handle;
static struct key_entry hotkey_keycode_map[GENERIC_HOTKEY_MAP_MAX];

static bool bl_powered;
+static struct backlight_device *backlight_dev;
static int loongson_laptop_backlight_update(struct backlight_device *bd);

/* 2. ACPI Helpers and device model */
@@ -194,13 +195,12 @@ static int loongson_hotkey_resume(struct device *dev)
{
int status = 0;
struct key_entry ke;
- struct backlight_device *bd;

- bd = backlight_device_get_by_type(BACKLIGHT_PLATFORM);
- if (bd) {
- loongson_laptop_backlight_update(bd) ?
+ if (backlight_dev) {
+ loongson_laptop_backlight_update(backlight_dev) ?
pr_warn("Loongson_backlight: resume brightness failed") :
- pr_info("Loongson_backlight: resume brightness %d\n", bd->props.brightness);
+ pr_info("Loongson_backlight: resume brightness %d\n",
+ backlight_dev->props.brightness);
}

/*
@@ -409,6 +409,7 @@ static int laptop_backlight_register(void)
{
int status = 0, ret;
struct backlight_properties props;
+ struct backlight_device *bd;

memset(&props, 0, sizeof(props));

@@ -426,9 +427,12 @@ static int laptop_backlight_register(void)
props.power = BACKLIGHT_POWER_ON;
props.type = BACKLIGHT_PLATFORM;

- backlight_device_register("loongson_laptop",
+ bd = backlight_device_register("loongson_laptop",
NULL, NULL, &backlight_laptop_ops, &props);
+ if (IS_ERR(bd))
+ return PTR_ERR(bd);

+ backlight_dev = bd;

return 0;
}
@@ -622,6 +626,9 @@ static void __exit generic_acpi_laptop_exit(void)
generic_subdriver_exit(&generic_sub_drivers[i]);
}
}
+
+ backlight_device_unregister(backlight_dev);
backlight_device_register is called conditionally but
backlight_device_unregister() is called unconditionally, I think this
is wrong.

Huacai
Hi, Huacai.

Thanks for the review. The unconditional call is intentional and is safe

because backlight_device_unregister() explicitly accepts a NULL device:

void backlight_device_unregister(struct backlight_device *bd)
{
if (!bd)
return;
...
}

backlight_dev is statically initialized to NULL. It is only assigned after
backlight_device_register() succeeds and the IS_ERR() check has passed. If
registration is skipped or fails, the pointer remains NULL, so the call in
the module exit path is a no-op rather than a NULL dereference.

I can change it to `if (backlight_dev) backlight_device_unregister(...)` in
a follow-up if you prefer that style.
Yes, I think that is better. And it is also better to put this block
before the "if (generic_inputdev)" block.
Thanks for the suggestion. I would prefer to keep the backlight
unregistration after the generic_inputdev cleanup block.

loongson_hotkey_resume() accesses backlight_dev, and that resume callback
belongs to loongson_hotkey_driver, which is only unregistered by
generic_subdriver_exit(). If we unregister the backlight before that block,
the PM callback remains reachable while its backlight_dev dependency may
already have been released.

The backlight device itself is independent of generic_inputdev. The
ordering is required by the hotkey PM callback lifetime, rather than by a
device dependency between the backlight and the input device.

I think keeping the teardown order as:

    generic_inputdev/generic_subdriver_exit()
    backlight_device_unregister()

is safer.

Qiu Luo


Huacai

Thanks.

Qiu Luo
+ backlight_dev = NULL;
}

module_init(generic_acpi_laptop_init);
--
2.56.1