fb_notifier_callback is called on any event fired by fb_notifier_call_chain. Events may, or may not contain some data (fb_event.data).
In case of FB_EVENT_BLANK fb_event.data contains a pointer to an integer holding the blank state. The Problem is, that in ledtrig-backlight.c - fb_notifier_callback the pointer to blank state is
dereferenced BEFORE the event-type is checked.
Obviously this leads to problems with other events than FB_EVENT_BLANK, where fb_event.data is undefined or NULL.
It seems, that this problem existed ever since the driver was added.
Like in drivers/video/backlight/backlight.c line 43 I would suggest to return immediately on events other than FB_EVENT_BLANK.
Signed-off-by: Manfred Schlaegl <[email protected]>
---
Background information:
I'm currently working on a IMX53(ARM) based hardware and Linux 3.11-rc5 and detected a problem in drivers/leds/trigger/ledtrig-backlight.c
ledtrig is used on our hardware (besides pwm-backlight) to switch off a gpio-line connected to display-supply enable-line, when the display is blanked.
On first boot after configuring ledtrig I detected an invalid memory access in ledtrig-backlight.c - fb_notifier_callback. In my special case the segfault was caused by an FB_EVENT_FB_REGISTERED,
fired with undefined fb_event.data in drivers/video/fbmem.c line 1653.
diff --git a/drivers/leds/trigger/ledtrig-backlight.c b/drivers/leds/trigger/ledtrig-backlight.c
index 3c9c88a..2538dbe 100644
--- a/drivers/leds/trigger/ledtrig-backlight.c
+++ b/drivers/leds/trigger/ledtrig-backlight.c
@@ -36,26 +36,28 @@ static int fb_notifier_callback(struct notifier_block *p,
struct bl_trig_notifier, notifier);
struct led_classdev *led = n->led;
struct fb_event *fb_event = data;
- int *blank = fb_event->data;
- int new_status = *blank ? BLANK : UNBLANK;
+ int *blank;
+ int new_status;
+
+ /* If we aren't interested in this event, skip it immediately ... */
+ if (event != FB_EVENT_BLANK)
+ return 0;
- switch (event) {
- case FB_EVENT_BLANK:
- if (new_status == n->old_status)
- break;
+ blank = fb_event->data;
+ new_status = *blank ? BLANK : UNBLANK;
- if ((n->old_status == UNBLANK) ^ n->invert) {
- n->brightness = led->brightness;
- __led_set_brightness(led, LED_OFF);
- } else {
- __led_set_brightness(led, n->brightness);
- }
+ if (new_status == n->old_status)
+ return 0;
- n->old_status = new_status;
-
- break;
+ if ((n->old_status == UNBLANK) ^ n->invert) {
+ n->brightness = led->brightness;
+ __led_set_brightness(led, LED_OFF);
+ } else {
+ __led_set_brightness(led, n->brightness);
}
+ n->old_status = new_status;
+
return 0;
}
On Tue, Aug 13, 2013 at 4:17 AM, Manfred Schlaegl
<[email protected]> wrote:
> fb_notifier_callback is called on any event fired by fb_notifier_call_chain. Events may, or may not contain some data (fb_event.data).
> In case of FB_EVENT_BLANK fb_event.data contains a pointer to an integer holding the blank state. The Problem is, that in ledtrig-backlight.c - fb_notifier_callback the pointer to blank state is
> dereferenced BEFORE the event-type is checked.
> Obviously this leads to problems with other events than FB_EVENT_BLANK, where fb_event.data is undefined or NULL.
> It seems, that this problem existed ever since the driver was added.
>
> Like in drivers/video/backlight/backlight.c line 43 I would suggest to return immediately on events other than FB_EVENT_BLANK.
>
Indeed, this is an issue. Thanks for this nice catch and fixing.
I merged this patch into my tree and it will hit 3.11 release.
Thanks,
-Bryan
> Signed-off-by: Manfred Schlaegl <[email protected]>
> ---
>
> Background information:
> I'm currently working on a IMX53(ARM) based hardware and Linux 3.11-rc5 and detected a problem in drivers/leds/trigger/ledtrig-backlight.c
> ledtrig is used on our hardware (besides pwm-backlight) to switch off a gpio-line connected to display-supply enable-line, when the display is blanked.
> On first boot after configuring ledtrig I detected an invalid memory access in ledtrig-backlight.c - fb_notifier_callback. In my special case the segfault was caused by an FB_EVENT_FB_REGISTERED,
> fired with undefined fb_event.data in drivers/video/fbmem.c line 1653.
>
>
> diff --git a/drivers/leds/trigger/ledtrig-backlight.c b/drivers/leds/trigger/ledtrig-backlight.c
> index 3c9c88a..2538dbe 100644
> --- a/drivers/leds/trigger/ledtrig-backlight.c
> +++ b/drivers/leds/trigger/ledtrig-backlight.c
> @@ -36,26 +36,28 @@ static int fb_notifier_callback(struct notifier_block *p,
> struct bl_trig_notifier, notifier);
> struct led_classdev *led = n->led;
> struct fb_event *fb_event = data;
> - int *blank = fb_event->data;
> - int new_status = *blank ? BLANK : UNBLANK;
> + int *blank;
> + int new_status;
> +
> + /* If we aren't interested in this event, skip it immediately ... */
> + if (event != FB_EVENT_BLANK)
> + return 0;
>
> - switch (event) {
> - case FB_EVENT_BLANK:
> - if (new_status == n->old_status)
> - break;
> + blank = fb_event->data;
> + new_status = *blank ? BLANK : UNBLANK;
>
> - if ((n->old_status == UNBLANK) ^ n->invert) {
> - n->brightness = led->brightness;
> - __led_set_brightness(led, LED_OFF);
> - } else {
> - __led_set_brightness(led, n->brightness);
> - }
> + if (new_status == n->old_status)
> + return 0;
>
> - n->old_status = new_status;
> -
> - break;
> + if ((n->old_status == UNBLANK) ^ n->invert) {
> + n->brightness = led->brightness;
> + __led_set_brightness(led, LED_OFF);
> + } else {
> + __led_set_brightness(led, n->brightness);
> }
>
> + n->old_status = new_status;
> +
> return 0;
> }
>
>
>
>
On 2013-08-27 02:51, Bryan Wu wrote:
> On Tue, Aug 13, 2013 at 4:17 AM, Manfred Schlaegl
> <[email protected]> wrote:
>> fb_notifier_callback is called on any event fired by fb_notifier_call_chain. Events may, or may not contain some data (fb_event.data).
>> In case of FB_EVENT_BLANK fb_event.data contains a pointer to an integer holding the blank state. The Problem is, that in ledtrig-backlight.c - fb_notifier_callback the pointer to blank state is
>> dereferenced BEFORE the event-type is checked.
>> Obviously this leads to problems with other events than FB_EVENT_BLANK, where fb_event.data is undefined or NULL.
>> It seems, that this problem existed ever since the driver was added.
>>
>> Like in drivers/video/backlight/backlight.c line 43 I would suggest to return immediately on events other than FB_EVENT_BLANK.
>>
>
> Indeed, this is an issue. Thanks for this nice catch and fixing.
>
> I merged this patch into my tree and it will hit 3.11 release.
>
> Thanks,
> -Bryan
That would be great.
Thanks too.
-manfred
>
>> Signed-off-by: Manfred Schlaegl <[email protected]>
>> ---
>>
>> Background information:
>> I'm currently working on a IMX53(ARM) based hardware and Linux 3.11-rc5 and detected a problem in drivers/leds/trigger/ledtrig-backlight.c
>> ledtrig is used on our hardware (besides pwm-backlight) to switch off a gpio-line connected to display-supply enable-line, when the display is blanked.
>> On first boot after configuring ledtrig I detected an invalid memory access in ledtrig-backlight.c - fb_notifier_callback. In my special case the segfault was caused by an FB_EVENT_FB_REGISTERED,
>> fired with undefined fb_event.data in drivers/video/fbmem.c line 1653.
>>
>>
>> diff --git a/drivers/leds/trigger/ledtrig-backlight.c b/drivers/leds/trigger/ledtrig-backlight.c
>> index 3c9c88a..2538dbe 100644
>> --- a/drivers/leds/trigger/ledtrig-backlight.c
>> +++ b/drivers/leds/trigger/ledtrig-backlight.c
>> @@ -36,26 +36,28 @@ static int fb_notifier_callback(struct notifier_block *p,
>> struct bl_trig_notifier, notifier);
>> struct led_classdev *led = n->led;
>> struct fb_event *fb_event = data;
>> - int *blank = fb_event->data;
>> - int new_status = *blank ? BLANK : UNBLANK;
>> + int *blank;
>> + int new_status;
>> +
>> + /* If we aren't interested in this event, skip it immediately ... */
>> + if (event != FB_EVENT_BLANK)
>> + return 0;
>>
>> - switch (event) {
>> - case FB_EVENT_BLANK:
>> - if (new_status == n->old_status)
>> - break;
>> + blank = fb_event->data;
>> + new_status = *blank ? BLANK : UNBLANK;
>>
>> - if ((n->old_status == UNBLANK) ^ n->invert) {
>> - n->brightness = led->brightness;
>> - __led_set_brightness(led, LED_OFF);
>> - } else {
>> - __led_set_brightness(led, n->brightness);
>> - }
>> + if (new_status == n->old_status)
>> + return 0;
>>
>> - n->old_status = new_status;
>> -
>> - break;
>> + if ((n->old_status == UNBLANK) ^ n->invert) {
>> + n->brightness = led->brightness;
>> + __led_set_brightness(led, LED_OFF);
>> + } else {
>> + __led_set_brightness(led, n->brightness);
>> }
>>
>> + n->old_status = new_status;
>> +
>> return 0;
>> }
>>
>>
>>
>>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-leds" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>