Re: [PATCH v2] extcon: usb-gpio: Don't miss event during suspend/resume

From: Roger Quadros
Date: Mon Apr 11 2016 - 10:02:46 EST


On 11/04/16 16:17, Chanwoo Choi wrote:
> Hi Roger,
>
> On Mon, Apr 11, 2016 at 8:39 PM, Roger Quadros <rogerq@xxxxxx> wrote:
>> Chanwoo,
>>
>> On 11/04/16 14:12, Chanwoo Choi wrote:
>>> On 2016ë 04ì 11ì 17:37, Grygorii Strashko wrote:
>>>> On 04/11/2016 03:31 AM, Chanwoo Choi wrote:
>>>>> Hi Roger,
>>>>>
>>>>> On 2016ë 04ì 08ì 16:34, Roger Quadros wrote:
>>>>>> Pin state might have changed during suspend/resume while
>>>>>> our interrupts were disabled and if device doesn't support wakeup.
>>>>>>
>>>>>> Scan for change during resume for such case.
>>>>>>
>>>>>> Signed-off-by: Roger Quadros <rogerq@xxxxxx>
>>>>>> ---
>>>>>> v2:
>>>>>> - only check for state change during resume if device wakeup is not supported
>>>>>>
>>>>>> drivers/extcon/extcon-usb-gpio.c | 2 ++
>>>>>> 1 file changed, 2 insertions(+)
>>>>>>
>>>>>> diff --git a/drivers/extcon/extcon-usb-gpio.c b/drivers/extcon/extcon-usb-gpio.c
>>>>>> index bc61d11..118f8ab 100644
>>>>>> --- a/drivers/extcon/extcon-usb-gpio.c
>>>>>> +++ b/drivers/extcon/extcon-usb-gpio.c
>>>>>> @@ -185,6 +185,8 @@ static int usb_extcon_resume(struct device *dev)
>>>>>> int ret = 0;
>>>>>>
>>>>>> enable_irq(info->id_irq);
>>>>>> + if (!device_may_wakeup(dev))
>>>>>> + usb_extcon_detect_cable(&info->wq_detcable.work);
>>>>>
>>>>> The device_may_wakeup() check the following two states:
>>>>> - dev->power.can_wakeup - device_init_wakeup() function set the this field.
>>>>> - dev->power.wakeup - device_wakeup_enable() function set the this field.
>>>>>
>>>>> The probe function of extcon-usb-gpio.c always call the 'device_init_wakeup(dev,true).
>>>>> But, anywhere in extcon-usb-gpio.c don't handle the device_wakeup_enable() for dev->power.wakeup.
>>>>
>>>>
>>>> device_init_wakeup()
>>>> |-> device_wakeup_enable()
>>>>
>>>>>
>>>>> In the extcon-usb-gpio.c, device_may_wakeup(dev) return always 'false'.
>>>>> If you use the only device_may_wakeup(),
>>>>> device_may_wakeup() is not able to check whether interrupt is wakeup source or not.
>>>>>
>>>>
>>>> This check is correct and it also will take into account wake up settings changes
>>>> which can be made through sysfs: /sys/.../devX/power/wakeup
>>>>
>>>
>>> To Grygorii,
>>>
>>> You're right. I was mistaken. Again, I analyzed the sequence about wakeup.
>>> Thanks for your reply.
>>>
>>> 1. Register device as wakeup_source.
>>> device_init_wakeup(dev, true) on probe()
>>> device_wakeup_enable(dev)
>>> device_source_register(const char *name)
>>> struct wakeup_source *ws;
>>> ws = wakeup_source_create(name)
>>> if (ws)
>>> wakeup_source_add(ws);
>>> ...
>>> list_add_rcu(&ws->entry, &wakeup_sources);
>>> ...
>>> return ws;
>>>
>>>
>>> 2. Register the interrupt as wake_irq
>>> dev_pm_set_wake_irq(struct device *dev, int irq) on probe()
>>> struct wake_irq *wirq;
>>> wirq->dev = dev;
>>> wirq->irq = irq;
>>> dev_pm_attach_wake_irq(dev, irq, wirq);
>>> device_wakeup_attach_irq(*dev, *wakeirq)
>>> struct wakeup_source *ws;
>>> ws = dev->power.wakeup;
>>> ws->wakeirq = wakeirq;
>>>
>>>
>>> 3. Enable irq wake if device is already registed to wakeup_sources.
>>> dpm_suspend_noirq()
>>> device_wakeup_arm_wake_irqs()
>>> list_for_each_entry_rcu(ws, &wakeup_sources, entry) {
>>> if (ws->wakeirq)
>>> dev_pm_arm_wake_irq(sw->wakeirq);
>>> if (device_may_wakeup(wirq->dev))
>>> enable_irq_wake(wirq->irq);
>>>
>>>
>>> To Roger,
>>>
>>> How about using the queue_delayed_work() instead of direct call function?
>>> Because the spent time of wakeup from suspend state should be fast.
>>> So, I think that you better to use the queue_delayed_work().
>>>
>>> diff --git a/drivers/extcon/extcon-usb-gpio.c b/drivers/extcon/extcon-usb-gpio.c
>>> index 118f8ab3be73..f6cbdfe31519 100644
>>> --- a/drivers/extcon/extcon-usb-gpio.c
>>> +++ b/drivers/extcon/extcon-usb-gpio.c
>>> @@ -186,7 +186,9 @@ static int usb_extcon_resume(struct device *dev)
>>>
>>> enable_irq(info->id_irq);
>>> if (!device_may_wakeup(dev))
>>> - usb_extcon_detect_cable(&info->wq_detcable.work);
>>> + queue_delayed_work(system_power_efficient_wq,
>>> + &info->wq_detcable,
>>> + info->debounce_jiffies);
>>
>> Why not to just use queue_work() instead of queue_delayed_work()
>> as don't need to debounce the input?
>
> The use of queue_work() is good.

Unfortunately we can't as info->wq_detcable type is 'struct delayed_work' but
queue_work() expects 'struct work_struct'

I'll just call queue_delayed_work() with 0 delay :)

cheers,
-roger