Skip to contentSkip to conversation

pwm_bl: Fix overflow condition

PATCH In review Derek Basehore 2 participants 4 messages Aug 28, 2017 – Aug 30, 2017

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pwm_bl: Fix overflow condition

This fixes and overflow condition that happens with a high value of brightness-levels-scale by using a 64-bit variable. The issue would prevent a range of higher brightness levels from being set.

---
 drivers/video/backlight/pwm_bl.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/video/backlight/pwm_bl.c +5 -2
index 76311ec5e400..e7ffd2108acf 100644
@@ -88,14 +88,17 @@ static void pwm_backlight_power_off(struct pwm_bl_data *pb)
 static int compute_duty_cycle(struct pwm_bl_data *pb, int brightness)
 {
 	unsigned int lth = pb->lth_brightness;
-	int duty_cycle;
+	s64 duty_cycle;
 
 	if (pb->levels)
 		duty_cycle = pb->levels[brightness];
 	else
 		duty_cycle = brightness;
 
-	return (duty_cycle * (pb->period - lth) / pb->scale) + lth;
+	duty_cycle *= pb->period - lth;
+	do_div(duty_cycle, pb->scale);
+
+	return duty_cycle + lth;
 }
 
 static int pwm_backlight_update_status(struct backlight_device *bl)
Signature · 1 line
2.14.1.342.g6490525c54-goog
#1 lore raw
Quoted 21 lines · Derek Basehore

This fixes and overflow condition that happens with a high value of brightness-levels-scale by using a 64-bit variable. The issue would prevent a range of higher brightness levels from being set.

---
 drivers/video/backlight/pwm_bl.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/video/backlight/pwm_bl.c +1 -1
index 76311ec5e400..e7ffd2108acf 100644
@@ -88,14 +88,17 @@ static void pwm_backlight_power_off(struct pwm_bl_data *pb)
 static int compute_duty_cycle(struct pwm_bl_data *pb, int brightness)
 {
 	unsigned int lth = pb->lth_brightness;
-	int duty_cycle;
+	s64 duty_cycle;
 
 	if (pb->levels)

drivers/video/backlight/pwm_bl.c static void pwm_backlight_power_off(struct pwm_bl_data *pb) · line 102

Quoted 10 lines
drivers/video/backlight/pwm_bl.c +4 -1
 		duty_cycle = pb->levels[brightness];
 	else
 		duty_cycle = brightness;
 
-	return (duty_cycle * (pb->period - lth) / pb->scale) + lth;
+	duty_cycle *= pb->period - lth;
+	do_div(duty_cycle, pb->scale);
+
+	return duty_cycle + lth;
 }

I don't think your commit message accurately describes the change here. The overflow that you're preventing might happen with a large value of pb->period (or rather, in combination with a large value of duty_cycle) but it's unrelated to pb->scale.

Also, the semantics of do_div() are that it takes an unsigned dividend, so your duty_cycle should be a u64.

Thierry

#2 lore raw
Quoted 26 lines · Thierry Reding
Quoted 25 lines · Derek Basehore

This fixes and overflow condition that happens with a high value of brightness-levels-scale by using a 64-bit variable. The issue would prevent a range of higher brightness levels from being set.

---
 drivers/video/backlight/pwm_bl.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/video/backlight/pwm_bl.c +1 -1
index 76311ec5e400..e7ffd2108acf 100644
@@ -88,14 +88,17 @@ static void pwm_backlight_power_off(struct pwm_bl_data *pb)
 static int compute_duty_cycle(struct pwm_bl_data *pb, int brightness)
 {
      unsigned int lth = pb->lth_brightness;
-     int duty_cycle;
+     s64 duty_cycle;
 
      if (pb->levels)
              duty_cycle = pb->levels[brightness];
      else
              duty_cycle = brightness;
 
Quoted 11 lines

drivers/video/backlight/pwm_bl.c static void pwm_backlight_power_off(struct pwm_bl_data *pb) · line 102

Quoted 6 lines
drivers/video/backlight/pwm_bl.c +4 -1
-     return (duty_cycle * (pb->period - lth) / pb->scale) + lth;
+     duty_cycle *= pb->period - lth;
+     do_div(duty_cycle, pb->scale);
+
+     return duty_cycle + lth;
 }

I don't think your commit message accurately describes the change here. The overflow that you're preventing might happen with a large value of pb->period (or rather, in combination with a large value of duty_cycle) but it's unrelated to pb->scale.

I'm referring to the of property brightness-levels-scale. If there aren't levels defined in a DTS, duty_cycle can be up to this value. I'll change the CM to describe what's happening based on the variable names from the function instead.

Quoted 3 lines

Also, the semantics of do_div() are that it takes an unsigned dividend, so your duty_cycle should be a u64.

I'll change it.

Quoted 2 lines

Thierry

#3 lore raw
Quoted 40 lines · dbasehore .
Quoted 26 lines · Thierry Reding
Quoted 25 lines · Derek Basehore

This fixes and overflow condition that happens with a high value of brightness-levels-scale by using a 64-bit variable. The issue would prevent a range of higher brightness levels from being set.

---
 drivers/video/backlight/pwm_bl.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/video/backlight/pwm_bl.c +1 -1
index 76311ec5e400..e7ffd2108acf 100644
@@ -88,14 +88,17 @@ static void pwm_backlight_power_off(struct pwm_bl_data *pb)
 static int compute_duty_cycle(struct pwm_bl_data *pb, int brightness)
 {
      unsigned int lth = pb->lth_brightness;
-     int duty_cycle;
+     s64 duty_cycle;
 
      if (pb->levels)
              duty_cycle = pb->levels[brightness];
      else
              duty_cycle = brightness;
 
Quoted 11 lines

drivers/video/backlight/pwm_bl.c static void pwm_backlight_power_off(struct pwm_bl_data *pb) · line 102

Quoted 6 lines
drivers/video/backlight/pwm_bl.c +4 -1
-     return (duty_cycle * (pb->period - lth) / pb->scale) + lth;
+     duty_cycle *= pb->period - lth;
+     do_div(duty_cycle, pb->scale);
+
+     return duty_cycle + lth;
 }

I don't think your commit message accurately describes the change here. The overflow that you're preventing might happen with a large value of pb->period (or rather, in combination with a large value of duty_cycle) but it's unrelated to pb->scale.

Quoted 1 line

I'm referring to the of property brightness-levels-scale.

I don't see a brightness-levels-scale property in either the DT bindings or the pwm-backlight driver. Do you have any pointers?

Thierry