Hi Simon,
Thanks for the patch. Generally this patch touches two
areas - replacement of redundant code with bcm6328_led_set,
and locking reorganization. These should be split into
two separate patches. Nonetheless, I've noticed some
issues in the code, please refer below.
On 10/29/2015 08:48 PM, Simon Arlott wrote:
> When ensuring a consistent initial LED state in bcm6328_led (as they may
> be blinking instead of on/off), the LED register is set using a copy of
> bcm6328_led_set(). To avoid further errors relating to active low handling,
> call this function directly instead.
>
> As bcm6328_led_set() expects to acquire the spinlock, narrow the locking
> to only cover reading of the current state. There is no need to hold the
> spinlock between reading the current value and setting it again because
> the LED device has not yet been registered.
>
> Signed-off-by: Simon Arlott <[email protected]>
> ---
> drivers/leds/leds-bcm6328.c | 14 +++++---------
> 1 file changed, 5 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
> index c7ea5c6..db327bd 100644
> --- a/drivers/leds/leds-bcm6328.c
> +++ b/drivers/leds/leds-bcm6328.c
> @@ -264,7 +264,6 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
> unsigned long *blink_leds, unsigned long *blink_delay)
> {
> struct bcm6328_led *led;
> - unsigned long flags;
> const char *state;
> int rc;
>
> @@ -286,13 +285,12 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
> "linux,default-trigger",
> NULL);
>
> - spin_lock_irqsave(lock, flags);
> if (!of_property_read_string(nc, "default-state", &state)) {
> if (!strcmp(state, "on")) {
> led->cdev.brightness = LED_FULL;
> } else if (!strcmp(state, "keep")) {
> void __iomem *mode;
> - unsigned long val, shift;
> + unsigned long val, shift, flags;
>
> shift = bcm6328_pin2shift(led->pin);
> if (shift / 16)
> @@ -300,9 +298,12 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
> else
> mode = mem + BCM6328_REG_MODE_LO;
>
> + spin_lock_irqsave(lock, flags);
> val = bcm6328_led_read(mode) >>
> BCM6328_LED_SHIFT(shift % 16);
> val &= BCM6328_LED_MODE_MASK;
> + spin_unlock_irqrestore(lock, flags);
> +
> if ((led->active_low && val == BCM6328_LED_MODE_ON) ||
> (!led->active_low && val == BCM6328_LED_MODE_OFF))
> led->cdev.brightness = LED_FULL;
> @@ -315,12 +316,7 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
> led->cdev.brightness = LED_OFF;
> }
>
> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
> - (!led->active_low && led->cdev.brightness == LED_OFF))
> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
> - else
> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
There are some problems with active_low mode here, I didn't recognize
earlier.
I'd expect that active_low implies reverse logic, i.e.:
LED_FULL -> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
LED_OFF -> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
Let's take a look at bcm6328_led_set:
if ((led->active_low && value == LED_OFF) ||
(!led->active_low && value != LED_OFF))
bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
else
bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
which, for active_low case, boils down to:
LED_FULL -> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
LED_OFF -> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
and for !active_low case to:
LED_FULL -> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
LED_OFF -> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
so, this is the other way round.
In bcm6328_led we have:
if ((led->active_low && val == BCM6328_LED_MODE_ON) ||
(!led->active_low && val == BCM6328_LED_MODE_OFF))
led->cdev.brightness = LED_FULL;
else
led->cdev.brightness = LED_OFF;
which, for active_low case, boils down to:
BCM6328_LED_MODE_ON -> led->cdev.brightness = LED_FULL
BCM6328_LED_MODE_OFF -> led->cdev.brightness = LED_OFF
and for !active_low case to:
BCM6328_LED_MODE_ON -> led->cdev.brightness = LED_OFF
BCM6328_LED_MODE_OFF -> led->cdev.brightness = LED_FULL
again, the other way round.
All this looks like active_low really means active high.
Correct me if I'm wrong.
Alvaro, Jonas, could you also help to clarify this discrepancy?
> - spin_unlock_irqrestore(lock, flags);
> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
>
> led->cdev.brightness_set = bcm6328_led_set;
> led->cdev.blink_set = bcm6328_blink_set;
>
--
Best Regards,
Jacek Anaszewski
Hello Jacek,
BCM6328_LED_MODE_ON and BCM6328_LED_MODE_OFF values were extracted from
Broadcom's GPL code, in which they assume leds are active low by default.
I can confirm the code is correct as it is right now, since those values
match the active high / low values of the LEDs managed by GPIO instead
of by using this driver.
Regards,
Álvaro.
El 04/11/2015 a las 16:41, Jacek Anaszewski escribió:
> Hi Simon,
>
> Thanks for the patch. Generally this patch touches two
> areas - replacement of redundant code with bcm6328_led_set,
> and locking reorganization. These should be split into
> two separate patches. Nonetheless, I've noticed some
> issues in the code, please refer below.
>
> On 10/29/2015 08:48 PM, Simon Arlott wrote:
>> When ensuring a consistent initial LED state in bcm6328_led (as they may
>> be blinking instead of on/off), the LED register is set using a copy of
>> bcm6328_led_set(). To avoid further errors relating to active low
>> handling,
>> call this function directly instead.
>>
>> As bcm6328_led_set() expects to acquire the spinlock, narrow the locking
>> to only cover reading of the current state. There is no need to hold the
>> spinlock between reading the current value and setting it again because
>> the LED device has not yet been registered.
>>
>> Signed-off-by: Simon Arlott <[email protected]>
>> ---
>> drivers/leds/leds-bcm6328.c | 14 +++++---------
>> 1 file changed, 5 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
>> index c7ea5c6..db327bd 100644
>> --- a/drivers/leds/leds-bcm6328.c
>> +++ b/drivers/leds/leds-bcm6328.c
>> @@ -264,7 +264,6 @@ static int bcm6328_led(struct device *dev, struct
>> device_node *nc, u32 reg,
>> unsigned long *blink_leds, unsigned long *blink_delay)
>> {
>> struct bcm6328_led *led;
>> - unsigned long flags;
>> const char *state;
>> int rc;
>>
>> @@ -286,13 +285,12 @@ static int bcm6328_led(struct device *dev,
>> struct device_node *nc, u32 reg,
>> "linux,default-trigger",
>> NULL);
>>
>> - spin_lock_irqsave(lock, flags);
>> if (!of_property_read_string(nc, "default-state", &state)) {
>> if (!strcmp(state, "on")) {
>> led->cdev.brightness = LED_FULL;
>> } else if (!strcmp(state, "keep")) {
>> void __iomem *mode;
>> - unsigned long val, shift;
>> + unsigned long val, shift, flags;
>>
>> shift = bcm6328_pin2shift(led->pin);
>> if (shift / 16)
>> @@ -300,9 +298,12 @@ static int bcm6328_led(struct device *dev,
>> struct device_node *nc, u32 reg,
>> else
>> mode = mem + BCM6328_REG_MODE_LO;
>>
>> + spin_lock_irqsave(lock, flags);
>> val = bcm6328_led_read(mode) >>
>> BCM6328_LED_SHIFT(shift % 16);
>> val &= BCM6328_LED_MODE_MASK;
>> + spin_unlock_irqrestore(lock, flags);
>> +
>> if ((led->active_low && val == BCM6328_LED_MODE_ON) ||
>> (!led->active_low && val == BCM6328_LED_MODE_OFF))
>> led->cdev.brightness = LED_FULL;
>> @@ -315,12 +316,7 @@ static int bcm6328_led(struct device *dev,
>> struct device_node *nc, u32 reg,
>> led->cdev.brightness = LED_OFF;
>> }
>>
>> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
>> - (!led->active_low && led->cdev.brightness == LED_OFF))
>> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>> - else
>> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>
> There are some problems with active_low mode here, I didn't recognize
> earlier.
>
> I'd expect that active_low implies reverse logic, i.e.:
>
> LED_FULL -> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
> LED_OFF -> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>
> Let's take a look at bcm6328_led_set:
>
> if ((led->active_low && value == LED_OFF) ||
> (!led->active_low && value != LED_OFF))
> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
> else
> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>
> which, for active_low case, boils down to:
>
> LED_FULL -> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
> LED_OFF -> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>
> and for !active_low case to:
>
> LED_FULL -> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
> LED_OFF -> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>
> so, this is the other way round.
>
> In bcm6328_led we have:
>
> if ((led->active_low && val == BCM6328_LED_MODE_ON) ||
> (!led->active_low && val == BCM6328_LED_MODE_OFF))
> led->cdev.brightness = LED_FULL;
> else
> led->cdev.brightness = LED_OFF;
>
> which, for active_low case, boils down to:
>
> BCM6328_LED_MODE_ON -> led->cdev.brightness = LED_FULL
> BCM6328_LED_MODE_OFF -> led->cdev.brightness = LED_OFF
>
> and for !active_low case to:
>
> BCM6328_LED_MODE_ON -> led->cdev.brightness = LED_OFF
> BCM6328_LED_MODE_OFF -> led->cdev.brightness = LED_FULL
>
> again, the other way round.
>
> All this looks like active_low really means active high.
> Correct me if I'm wrong.
>
> Alvaro, Jonas, could you also help to clarify this discrepancy?
>
>
>> - spin_unlock_irqrestore(lock, flags);
>> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
>>
>> led->cdev.brightness_set = bcm6328_led_set;
>> led->cdev.blink_set = bcm6328_blink_set;
>>
>
>
Hi Alvaro,
On 11/04/2015 04:46 PM, Álvaro Fernández Rojas wrote:
> Hello Jacek,
>
> BCM6328_LED_MODE_ON and BCM6328_LED_MODE_OFF values were extracted from
> Broadcom's GPL code, in which they assume leds are active low by default.
> I can confirm the code is correct as it is right now, since those values
> match the active high / low values of the LEDs managed by GPIO instead
> of by using this driver.
BCM6328_LED_MODE_ON and BCM6328_LED_MODE_OFF should represent the values
that actually set the LED state according to the current logic.
Otherwise it will confuse people who will be analyzing this code.
We are interested in the logic as it is seen from this driver's
perspective and not GPIO perspective.
IMO the values should be swapped.
Best Regards,
Jacek Anaszewski
> El 04/11/2015 a las 16:41, Jacek Anaszewski escribió:
>> Hi Simon,
>>
>> Thanks for the patch. Generally this patch touches two
>> areas - replacement of redundant code with bcm6328_led_set,
>> and locking reorganization. These should be split into
>> two separate patches. Nonetheless, I've noticed some
>> issues in the code, please refer below.
>>
>> On 10/29/2015 08:48 PM, Simon Arlott wrote:
>>> When ensuring a consistent initial LED state in bcm6328_led (as they may
>>> be blinking instead of on/off), the LED register is set using a copy of
>>> bcm6328_led_set(). To avoid further errors relating to active low
>>> handling,
>>> call this function directly instead.
>>>
>>> As bcm6328_led_set() expects to acquire the spinlock, narrow the locking
>>> to only cover reading of the current state. There is no need to hold the
>>> spinlock between reading the current value and setting it again because
>>> the LED device has not yet been registered.
>>>
>>> Signed-off-by: Simon Arlott <[email protected]>
>>> ---
>>> drivers/leds/leds-bcm6328.c | 14 +++++---------
>>> 1 file changed, 5 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
>>> index c7ea5c6..db327bd 100644
>>> --- a/drivers/leds/leds-bcm6328.c
>>> +++ b/drivers/leds/leds-bcm6328.c
>>> @@ -264,7 +264,6 @@ static int bcm6328_led(struct device *dev, struct
>>> device_node *nc, u32 reg,
>>> unsigned long *blink_leds, unsigned long *blink_delay)
>>> {
>>> struct bcm6328_led *led;
>>> - unsigned long flags;
>>> const char *state;
>>> int rc;
>>>
>>> @@ -286,13 +285,12 @@ static int bcm6328_led(struct device *dev,
>>> struct device_node *nc, u32 reg,
>>> "linux,default-trigger",
>>> NULL);
>>>
>>> - spin_lock_irqsave(lock, flags);
>>> if (!of_property_read_string(nc, "default-state", &state)) {
>>> if (!strcmp(state, "on")) {
>>> led->cdev.brightness = LED_FULL;
>>> } else if (!strcmp(state, "keep")) {
>>> void __iomem *mode;
>>> - unsigned long val, shift;
>>> + unsigned long val, shift, flags;
>>>
>>> shift = bcm6328_pin2shift(led->pin);
>>> if (shift / 16)
>>> @@ -300,9 +298,12 @@ static int bcm6328_led(struct device *dev,
>>> struct device_node *nc, u32 reg,
>>> else
>>> mode = mem + BCM6328_REG_MODE_LO;
>>>
>>> + spin_lock_irqsave(lock, flags);
>>> val = bcm6328_led_read(mode) >>
>>> BCM6328_LED_SHIFT(shift % 16);
>>> val &= BCM6328_LED_MODE_MASK;
>>> + spin_unlock_irqrestore(lock, flags);
>>> +
>>> if ((led->active_low && val == BCM6328_LED_MODE_ON) ||
>>> (!led->active_low && val == BCM6328_LED_MODE_OFF))
>>> led->cdev.brightness = LED_FULL;
>>> @@ -315,12 +316,7 @@ static int bcm6328_led(struct device *dev,
>>> struct device_node *nc, u32 reg,
>>> led->cdev.brightness = LED_OFF;
>>> }
>>>
>>> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
>>> - (!led->active_low && led->cdev.brightness == LED_OFF))
>>> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>>> - else
>>> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>>
>> There are some problems with active_low mode here, I didn't recognize
>> earlier.
>>
>> I'd expect that active_low implies reverse logic, i.e.:
>>
>> LED_FULL -> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>> LED_OFF -> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>>
>> Let's take a look at bcm6328_led_set:
>>
>> if ((led->active_low && value == LED_OFF) ||
>> (!led->active_low && value != LED_OFF))
>> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>> else
>> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>>
>> which, for active_low case, boils down to:
>>
>> LED_FULL -> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>> LED_OFF -> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>>
>> and for !active_low case to:
>>
>> LED_FULL -> bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>> LED_OFF -> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>>
>> so, this is the other way round.
>>
>> In bcm6328_led we have:
>>
>> if ((led->active_low && val == BCM6328_LED_MODE_ON) ||
>> (!led->active_low && val == BCM6328_LED_MODE_OFF))
>> led->cdev.brightness = LED_FULL;
>> else
>> led->cdev.brightness = LED_OFF;
>>
>> which, for active_low case, boils down to:
>>
>> BCM6328_LED_MODE_ON -> led->cdev.brightness = LED_FULL
>> BCM6328_LED_MODE_OFF -> led->cdev.brightness = LED_OFF
>>
>> and for !active_low case to:
>>
>> BCM6328_LED_MODE_ON -> led->cdev.brightness = LED_OFF
>> BCM6328_LED_MODE_OFF -> led->cdev.brightness = LED_FULL
>>
>> again, the other way round.
>>
>> All this looks like active_low really means active high.
>> Correct me if I'm wrong.
>>
>> Alvaro, Jonas, could you also help to clarify this discrepancy?
>>
>>
>>> - spin_unlock_irqrestore(lock, flags);
>>> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
>>>
>>> led->cdev.brightness_set = bcm6328_led_set;
>>> led->cdev.blink_set = bcm6328_blink_set;
>>>
>>
>>
>
>
When ensuring a consistent initial LED state in bcm6328_led (as they may
be blinking instead of on/off), the LED register is set using an inverted
copy of bcm6328_led_set(). To avoid further errors relating to active low
handling, call this function directly instead.
Signed-off-by: Simon Arlott <[email protected]>
---
I've decided not to move the locking as it should really cover the
setting of led->cdev.brightness too (not that it matters as the device
is not yet registered).
drivers/leds/leds-bcm6328.c | 8 ++------
1 file changed, 2 insertions(+), 6 deletions(-)
diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
index c7ea5c6..95d0cf9 100644
--- a/drivers/leds/leds-bcm6328.c
+++ b/drivers/leds/leds-bcm6328.c
@@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
} else {
led->cdev.brightness = LED_OFF;
}
-
- if ((led->active_low && led->cdev.brightness == LED_FULL) ||
- (!led->active_low && led->cdev.brightness == LED_OFF))
- bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
- else
- bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
spin_unlock_irqrestore(lock, flags);
+ bcm6328_led_set(&led->cdev, led->cdev.brightness);
+
led->cdev.brightness_set = bcm6328_led_set;
led->cdev.blink_set = bcm6328_blink_set;
--
2.1.4
--
Simon Arlott
The values of BCM6328_LED_MODE_ON and BCM6328_LED_MODE_OFF were named
for active low LEDs. These should be swapped so that they are named for
the default case of active high LEDs.
Signed-off-by: Simon Arlott <[email protected]>
---
On 05/11/15 10:41, Jacek Anaszewski wrote:
> On 11/04/2015 04:46 PM, Álvaro Fernández Rojas wrote:
>> BCM6328_LED_MODE_ON and BCM6328_LED_MODE_OFF values were extracted from
>> Broadcom's GPL code, in which they assume leds are active low by default.
>> I can confirm the code is correct as it is right now, since those values
>> match the active high / low values of the LEDs managed by GPIO instead
>> of by using this driver.
>
> BCM6328_LED_MODE_ON and BCM6328_LED_MODE_OFF should represent the values
> that actually set the LED state according to the current logic.
> Otherwise it will confuse people who will be analyzing this code.
> We are interested in the logic as it is seen from this driver's
> perspective and not GPIO perspective.
>
> IMO the values should be swapped.
drivers/leds/leds-bcm6328.c | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
index 95d0cf9..0329dee 100644
--- a/drivers/leds/leds-bcm6328.c
+++ b/drivers/leds/leds-bcm6328.c
@@ -48,10 +48,10 @@
BCM6328_SERIAL_LED_SHIFT_DIR)
#define BCM6328_LED_MODE_MASK 3
-#define BCM6328_LED_MODE_OFF 0
+#define BCM6328_LED_MODE_ON 0
#define BCM6328_LED_MODE_FAST 1
#define BCM6328_LED_MODE_BLINK 2
-#define BCM6328_LED_MODE_ON 3
+#define BCM6328_LED_MODE_OFF 3
#define BCM6328_LED_SHIFT(X) ((X) << 1)
/**
@@ -126,9 +126,9 @@ static void bcm6328_led_set(struct led_classdev *led_cdev,
*(led->blink_leds) &= ~BIT(led->pin);
if ((led->active_low && value == LED_OFF) ||
(!led->active_low && value != LED_OFF))
- bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
- else
bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
+ else
+ bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
spin_unlock_irqrestore(led->lock, flags);
}
@@ -303,8 +303,8 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
val = bcm6328_led_read(mode) >>
BCM6328_LED_SHIFT(shift % 16);
val &= BCM6328_LED_MODE_MASK;
- if ((led->active_low && val == BCM6328_LED_MODE_ON) ||
- (!led->active_low && val == BCM6328_LED_MODE_OFF))
+ if ((led->active_low && val == BCM6328_LED_MODE_OFF) ||
+ (!led->active_low && val == BCM6328_LED_MODE_ON))
led->cdev.brightness = LED_FULL;
else
led->cdev.brightness = LED_OFF;
--
2.1.4
--
Simon Arlott
Hi Simon,
On 11/15/2015 02:32 PM, Simon Arlott wrote:
> When ensuring a consistent initial LED state in bcm6328_led (as they may
> be blinking instead of on/off), the LED register is set using an inverted
> copy of bcm6328_led_set(). To avoid further errors relating to active low
> handling, call this function directly instead.
>
> Signed-off-by: Simon Arlott <[email protected]>
> ---
> I've decided not to move the locking as it should really cover the
> setting of led->cdev.brightness too (not that it matters as the device
> is not yet registered).
>
> drivers/leds/leds-bcm6328.c | 8 ++------
> 1 file changed, 2 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
> index c7ea5c6..95d0cf9 100644
> --- a/drivers/leds/leds-bcm6328.c
> +++ b/drivers/leds/leds-bcm6328.c
> @@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
> } else {
> led->cdev.brightness = LED_OFF;
> }
> -
> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
> - (!led->active_low && led->cdev.brightness == LED_OFF))
> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
> - else
> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
> spin_unlock_irqrestore(lock, flags);
>
> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
> +
You're not protecting bcm6328_led_set with spin_lock here.
Please describe why this is not needed in the commit message.
> led->cdev.brightness_set = bcm6328_led_set;
> led->cdev.blink_set = bcm6328_blink_set;
>
>
--
Best Regards,
Jacek Anaszewski
When ensuring a consistent initial LED state in bcm6328_led (as they may
be blinking instead of on/off), the LED register is set using an inverted
copy of bcm6328_led_set(). To avoid further errors relating to active low
handling, call this function directly instead.
As bcm6328_led_set() acquires the same spinlock again when updating the
register, it is called after unlocking.
Signed-off-by: Simon Arlott <[email protected]>
---
drivers/leds/leds-bcm6328.c | 8 ++------
1 file changed, 2 insertions(+), 6 deletions(-)
diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
index c7ea5c6..95d0cf9 100644
--- a/drivers/leds/leds-bcm6328.c
+++ b/drivers/leds/leds-bcm6328.c
@@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
} else {
led->cdev.brightness = LED_OFF;
}
-
- if ((led->active_low && led->cdev.brightness == LED_FULL) ||
- (!led->active_low && led->cdev.brightness == LED_OFF))
- bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
- else
- bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
spin_unlock_irqrestore(lock, flags);
+ bcm6328_led_set(&led->cdev, led->cdev.brightness);
+
led->cdev.brightness_set = bcm6328_led_set;
led->cdev.blink_set = bcm6328_blink_set;
--
2.1.4
--
Simon Arlott
Still wrong, you are setting the led value after unlocking the spinlock.
El 16/11/2015 a las 21:24, Simon Arlott escribió:
> When ensuring a consistent initial LED state in bcm6328_led (as they may
> be blinking instead of on/off), the LED register is set using an inverted
> copy of bcm6328_led_set(). To avoid further errors relating to active low
> handling, call this function directly instead.
>
> As bcm6328_led_set() acquires the same spinlock again when updating the
> register, it is called after unlocking.
>
> Signed-off-by: Simon Arlott <[email protected]>
> ---
> drivers/leds/leds-bcm6328.c | 8 ++------
> 1 file changed, 2 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
> index c7ea5c6..95d0cf9 100644
> --- a/drivers/leds/leds-bcm6328.c
> +++ b/drivers/leds/leds-bcm6328.c
> @@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
> } else {
> led->cdev.brightness = LED_OFF;
> }
> -
> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
> - (!led->active_low && led->cdev.brightness == LED_OFF))
> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
> - else
> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
> spin_unlock_irqrestore(lock, flags);
>
> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
> +
> led->cdev.brightness_set = bcm6328_led_set;
> led->cdev.blink_set = bcm6328_blink_set;
>
On 16/11/15 21:33, Álvaro Fernández Rojas wrote:
> Still wrong, you are setting the led value after unlocking the spinlock.
I have to unlock it because bcm6328_led_set() locks that spinlock.
> El 16/11/2015 a las 21:24, Simon Arlott escribió:
>> When ensuring a consistent initial LED state in bcm6328_led (as they may
>> be blinking instead of on/off), the LED register is set using an inverted
>> copy of bcm6328_led_set(). To avoid further errors relating to active low
>> handling, call this function directly instead.
>>
>> As bcm6328_led_set() acquires the same spinlock again when updating the
>> register, it is called after unlocking.
>>
>> Signed-off-by: Simon Arlott <[email protected]>
>> ---
>> drivers/leds/leds-bcm6328.c | 8 ++------
>> 1 file changed, 2 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
>> index c7ea5c6..95d0cf9 100644
>> --- a/drivers/leds/leds-bcm6328.c
>> +++ b/drivers/leds/leds-bcm6328.c
>> @@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
>> } else {
>> led->cdev.brightness = LED_OFF;
>> }
>> -
>> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
>> - (!led->active_low && led->cdev.brightness == LED_OFF))
>> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>> - else
>> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>> spin_unlock_irqrestore(lock, flags);
>>
>> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
>> +
>> led->cdev.brightness_set = bcm6328_led_set;
>> led->cdev.blink_set = bcm6328_blink_set;
>>
>
--
Simon Arlott
Hi Alvaro, Simon,
On 11/17/2015 08:42 AM, Simon Arlott wrote:
> On 16/11/15 21:33, Álvaro Fernández Rojas wrote:
>> Still wrong, you are setting the led value after unlocking the spinlock.
>
> I have to unlock it because bcm6328_led_set() locks that spinlock.
Commit message from the first version of the patch justified that
properly. It should be preserved in the final patch:
As bcm6328_led_set() expects to acquire the spinlock, narrow the locking
to only cover reading of the current state. There is no need to hold the
spinlock between reading the current value and setting it again because
the LED device has not yet been registered.
>> El 16/11/2015 a las 21:24, Simon Arlott escribió:
>>> When ensuring a consistent initial LED state in bcm6328_led (as they may
>>> be blinking instead of on/off), the LED register is set using an inverted
>>> copy of bcm6328_led_set(). To avoid further errors relating to active low
>>> handling, call this function directly instead.
>>>
>>> As bcm6328_led_set() acquires the same spinlock again when updating the
>>> register, it is called after unlocking.
>>>
>>> Signed-off-by: Simon Arlott <[email protected]>
>>> ---
>>> drivers/leds/leds-bcm6328.c | 8 ++------
>>> 1 file changed, 2 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
>>> index c7ea5c6..95d0cf9 100644
>>> --- a/drivers/leds/leds-bcm6328.c
>>> +++ b/drivers/leds/leds-bcm6328.c
>>> @@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
>>> } else {
>>> led->cdev.brightness = LED_OFF;
>>> }
>>> -
>>> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
>>> - (!led->active_low && led->cdev.brightness == LED_OFF))
>>> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>>> - else
>>> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>>> spin_unlock_irqrestore(lock, flags);
>>>
>>> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
>>> +
>>> led->cdev.brightness_set = bcm6328_led_set;
>>> led->cdev.blink_set = bcm6328_blink_set;
>>>
>>
>
>
--
Best Regards,
Jacek Anaszewski
On 11/17/2015 09:06 AM, Jacek Anaszewski wrote:
> Hi Alvaro, Simon,
>
> On 11/17/2015 08:42 AM, Simon Arlott wrote:
>> On 16/11/15 21:33, Álvaro Fernández Rojas wrote:
>>> Still wrong, you are setting the led value after unlocking the spinlock.
>>
>> I have to unlock it because bcm6328_led_set() locks that spinlock.
>
> Commit message from the first version of the patch justified that
> properly. It should be preserved in the final patch:
>
> As bcm6328_led_set() expects to acquire the spinlock, narrow the locking
> to only cover reading of the current state. There is no need to hold the
> spinlock between reading the current value and setting it again because
> the LED device has not yet been registered.
Hmm, if so, then spin_lock in bcm6328_led isn't needed at all, as it
is guaranteed that no concurrent process will be executing this
function.
>>> El 16/11/2015 a las 21:24, Simon Arlott escribió:
>>>> When ensuring a consistent initial LED state in bcm6328_led (as they
>>>> may
>>>> be blinking instead of on/off), the LED register is set using an
>>>> inverted
>>>> copy of bcm6328_led_set(). To avoid further errors relating to
>>>> active low
>>>> handling, call this function directly instead.
>>>>
>>>> As bcm6328_led_set() acquires the same spinlock again when updating the
>>>> register, it is called after unlocking.
>>>>
>>>> Signed-off-by: Simon Arlott <[email protected]>
>>>> ---
>>>> drivers/leds/leds-bcm6328.c | 8 ++------
>>>> 1 file changed, 2 insertions(+), 6 deletions(-)
>>>>
>>>> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
>>>> index c7ea5c6..95d0cf9 100644
>>>> --- a/drivers/leds/leds-bcm6328.c
>>>> +++ b/drivers/leds/leds-bcm6328.c
>>>> @@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev,
>>>> struct device_node *nc, u32 reg,
>>>> } else {
>>>> led->cdev.brightness = LED_OFF;
>>>> }
>>>> -
>>>> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
>>>> - (!led->active_low && led->cdev.brightness == LED_OFF))
>>>> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
>>>> - else
>>>> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
>>>> spin_unlock_irqrestore(lock, flags);
>>>>
>>>> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
>>>> +
>>>> led->cdev.brightness_set = bcm6328_led_set;
>>>> led->cdev.blink_set = bcm6328_blink_set;
>>>>
>>>
>>
>>
>
>
--
Best Regards,
Jacek Anaszewski
When ensuring a consistent initial LED state in bcm6328_led (as they may
be blinking instead of on/off), the LED register is set using an inverted
copy of bcm6328_led_set(). To avoid further errors relating to active low
handling, call this function directly instead.
As bcm6328_led_set() expects to acquire the spinlock, call it after
unlocking. There is no need to hold the spinlock between reading the
current value and setting it again because the LED device has not yet
been registered.
Signed-off-by: Simon Arlott <[email protected]>
---
On 17/11/15 08:15, Jacek Anaszewski wrote:
> On 11/17/2015 09:06 AM, Jacek Anaszewski wrote:
>> On 11/17/2015 08:42 AM, Simon Arlott wrote:
>>> On 16/11/15 21:33, Álvaro Fernández Rojas wrote:
>>>> Still wrong, you are setting the led value after unlocking the spinlock.
>>>
>>> I have to unlock it because bcm6328_led_set() locks that spinlock.
>>
>> Commit message from the first version of the patch justified that
>> properly. It should be preserved in the final patch:
>>
>> As bcm6328_led_set() expects to acquire the spinlock, narrow the locking
>> to only cover reading of the current state. There is no need to hold the
>> spinlock between reading the current value and setting it again because
>> the LED device has not yet been registered.
>
> Hmm, if so, then spin_lock in bcm6328_led isn't needed at all, as it
> is guaranteed that no concurrent process will be executing this
> function.
No, it's still required because it has to protect the read/modify/write
for all the other LED devices that use the same register.
drivers/leds/leds-bcm6328.c | 8 ++------
1 file changed, 2 insertions(+), 6 deletions(-)
diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
index c7ea5c6..95d0cf9 100644
--- a/drivers/leds/leds-bcm6328.c
+++ b/drivers/leds/leds-bcm6328.c
@@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
} else {
led->cdev.brightness = LED_OFF;
}
-
- if ((led->active_low && led->cdev.brightness == LED_FULL) ||
- (!led->active_low && led->cdev.brightness == LED_OFF))
- bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
- else
- bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
spin_unlock_irqrestore(lock, flags);
+ bcm6328_led_set(&led->cdev, led->cdev.brightness);
+
led->cdev.brightness_set = bcm6328_led_set;
led->cdev.blink_set = bcm6328_blink_set;
--
2.1.4
--
Simon Arlott
On 11/22/2015 09:40 PM, Simon Arlott wrote:
> When ensuring a consistent initial LED state in bcm6328_led (as they may
> be blinking instead of on/off), the LED register is set using an inverted
> copy of bcm6328_led_set(). To avoid further errors relating to active low
> handling, call this function directly instead.
>
> As bcm6328_led_set() expects to acquire the spinlock, call it after
> unlocking. There is no need to hold the spinlock between reading the
> current value and setting it again because the LED device has not yet
> been registered.
>
> Signed-off-by: Simon Arlott <[email protected]>
> ---
> On 17/11/15 08:15, Jacek Anaszewski wrote:
>> On 11/17/2015 09:06 AM, Jacek Anaszewski wrote:
>>> On 11/17/2015 08:42 AM, Simon Arlott wrote:
>>>> On 16/11/15 21:33, Álvaro Fernández Rojas wrote:
>>>>> Still wrong, you are setting the led value after unlocking the spinlock.
>>>>
>>>> I have to unlock it because bcm6328_led_set() locks that spinlock.
>>>
>>> Commit message from the first version of the patch justified that
>>> properly. It should be preserved in the final patch:
>>>
>>> As bcm6328_led_set() expects to acquire the spinlock, narrow the locking
>>> to only cover reading of the current state. There is no need to hold the
>>> spinlock between reading the current value and setting it again because
>>> the LED device has not yet been registered.
>>
>> Hmm, if so, then spin_lock in bcm6328_led isn't needed at all, as it
>> is guaranteed that no concurrent process will be executing this
>> function.
>
> No, it's still required because it has to protect the read/modify/write
> for all the other LED devices that use the same register.
Right, other already registered LED class devices can interfere.
> drivers/leds/leds-bcm6328.c | 8 ++------
> 1 file changed, 2 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
> index c7ea5c6..95d0cf9 100644
> --- a/drivers/leds/leds-bcm6328.c
> +++ b/drivers/leds/leds-bcm6328.c
> @@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
> } else {
> led->cdev.brightness = LED_OFF;
> }
> -
> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
> - (!led->active_low && led->cdev.brightness == LED_OFF))
> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
> - else
> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
> spin_unlock_irqrestore(lock, flags);
>
> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
> +
> led->cdev.brightness_set = bcm6328_led_set;
> led->cdev.blink_set = bcm6328_blink_set;
>
>
--
Best Regards,
Jacek Anaszewski
On 11/16/2015 09:24 PM, Simon Arlott wrote:
> When ensuring a consistent initial LED state in bcm6328_led (as they may
> be blinking instead of on/off), the LED register is set using an inverted
> copy of bcm6328_led_set(). To avoid further errors relating to active low
> handling, call this function directly instead.
>
> As bcm6328_led_set() acquires the same spinlock again when updating the
> register, it is called after unlocking.
>
> Signed-off-by: Simon Arlott <[email protected]>
> ---
> drivers/leds/leds-bcm6328.c | 8 ++------
> 1 file changed, 2 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
> index c7ea5c6..95d0cf9 100644
> --- a/drivers/leds/leds-bcm6328.c
> +++ b/drivers/leds/leds-bcm6328.c
> @@ -314,14 +314,10 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
> } else {
> led->cdev.brightness = LED_OFF;
> }
> -
> - if ((led->active_low && led->cdev.brightness == LED_FULL) ||
> - (!led->active_low && led->cdev.brightness == LED_OFF))
> - bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
> - else
> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
> spin_unlock_irqrestore(lock, flags);
>
> + bcm6328_led_set(&led->cdev, led->cdev.brightness);
> +
> led->cdev.brightness_set = bcm6328_led_set;
> led->cdev.blink_set = bcm6328_blink_set;
>
>
Applied, thanks.
--
Best Regards,
Jacek Anaszewski
On 11/15/2015 02:34 PM, Simon Arlott wrote:
> The values of BCM6328_LED_MODE_ON and BCM6328_LED_MODE_OFF were named
> for active low LEDs. These should be swapped so that they are named for
> the default case of active high LEDs.
>
> Signed-off-by: Simon Arlott <[email protected]>
> ---
> On 05/11/15 10:41, Jacek Anaszewski wrote:
>> On 11/04/2015 04:46 PM, Álvaro Fernández Rojas wrote:
>>> BCM6328_LED_MODE_ON and BCM6328_LED_MODE_OFF values were extracted from
>>> Broadcom's GPL code, in which they assume leds are active low by default.
>>> I can confirm the code is correct as it is right now, since those values
>>> match the active high / low values of the LEDs managed by GPIO instead
>>> of by using this driver.
>>
>> BCM6328_LED_MODE_ON and BCM6328_LED_MODE_OFF should represent the values
>> that actually set the LED state according to the current logic.
>> Otherwise it will confuse people who will be analyzing this code.
>> We are interested in the logic as it is seen from this driver's
>> perspective and not GPIO perspective.
>>
>> IMO the values should be swapped.
>
> drivers/leds/leds-bcm6328.c | 12 ++++++------
> 1 file changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/leds/leds-bcm6328.c b/drivers/leds/leds-bcm6328.c
> index 95d0cf9..0329dee 100644
> --- a/drivers/leds/leds-bcm6328.c
> +++ b/drivers/leds/leds-bcm6328.c
> @@ -48,10 +48,10 @@
> BCM6328_SERIAL_LED_SHIFT_DIR)
>
> #define BCM6328_LED_MODE_MASK 3
> -#define BCM6328_LED_MODE_OFF 0
> +#define BCM6328_LED_MODE_ON 0
> #define BCM6328_LED_MODE_FAST 1
> #define BCM6328_LED_MODE_BLINK 2
> -#define BCM6328_LED_MODE_ON 3
> +#define BCM6328_LED_MODE_OFF 3
> #define BCM6328_LED_SHIFT(X) ((X) << 1)
>
> /**
> @@ -126,9 +126,9 @@ static void bcm6328_led_set(struct led_classdev *led_cdev,
> *(led->blink_leds) &= ~BIT(led->pin);
> if ((led->active_low && value == LED_OFF) ||
> (!led->active_low && value != LED_OFF))
> - bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
> - else
> bcm6328_led_mode(led, BCM6328_LED_MODE_ON);
> + else
> + bcm6328_led_mode(led, BCM6328_LED_MODE_OFF);
> spin_unlock_irqrestore(led->lock, flags);
> }
>
> @@ -303,8 +303,8 @@ static int bcm6328_led(struct device *dev, struct device_node *nc, u32 reg,
> val = bcm6328_led_read(mode) >>
> BCM6328_LED_SHIFT(shift % 16);
> val &= BCM6328_LED_MODE_MASK;
> - if ((led->active_low && val == BCM6328_LED_MODE_ON) ||
> - (!led->active_low && val == BCM6328_LED_MODE_OFF))
> + if ((led->active_low && val == BCM6328_LED_MODE_OFF) ||
> + (!led->active_low && val == BCM6328_LED_MODE_ON))
> led->cdev.brightness = LED_FULL;
> else
> led->cdev.brightness = LED_OFF;
>
Applied, thanks.
--
Best Regards,
Jacek Anaszewski