Re: [PATCH] extcon: Release locking when sending the notification of connector state

[H. Nikolaus Schaller]

> Am 14.06.2018 um 12:39 schrieb H. Nikolaus Schaller <hns@xxxxxxxxxxxxx>:
> 
> Hi Roger and Chanwoo,
> 
>> Am 14.06.2018 um 12:18 schrieb Chanwoo Choi <cw00.choi@xxxxxxxxxxx>:
>> 
>> + H. Nikolaus Schaller <hns@xxxxxxxxxxxxx>
>> 
>> On 2018ë 06ì 14ì 13:14, Chanwoo Choi wrote:
>>> Previously, extcon used the spinlock before calling the notifier_call_chain
>>> to prevent the scheduled out of task and to prevent the notification delay.
>>> When spinlock is locked for sending the notification, deadlock issue
>>> occured on the side of extcon consumer device. To fix this issue,
>>> extcon consumer device should always use the work. it is always not
>>> reasonable to use work.
>>> 
>>> To fix this issue on extcon consumer device, release locking when sending
>>> the notification of connector state.
>>> 
>>> Fixes: ab11af049f88 ("extcon: Add the synchronization extcon APIs to support the notification")
>>> Cc: stable@xxxxxxxxxxxxxxx
>>> Cc: Roger Quadros <rogerq@xxxxxx>
>>> Cc: Kishon Vijay Abraham I <kishon@xxxxxx>
>>> Signed-off-by: Chanwoo Choi <cw00.choi@xxxxxxxxxxx>
>>> ---
>>> drivers/extcon/extcon.c | 3 ++-
>>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>> 
>>> diff --git a/drivers/extcon/extcon.c b/drivers/extcon/extcon.c
>>> index 8bff5fd18185..f75b08a45d4e 100644
>>> --- a/drivers/extcon/extcon.c
>>> +++ b/drivers/extcon/extcon.c
>>> @@ -433,8 +433,8 @@ int extcon_sync(struct extcon_dev *edev, unsigned int id)
>>> 		return index;
>>> 
>>> 	spin_lock_irqsave(&edev->lock, flags);
>>> -
>>> 	state = !!(edev->state & BIT(index));
>>> +	spin_unlock_irqrestore(&edev->lock, flags);
>>> 
>>> 	/*
>>> 	 * Call functions in a raw notifier chain for the specific one
>>> @@ -448,6 +448,7 @@ int extcon_sync(struct extcon_dev *edev, unsigned int id)
>>> 	 */
>>> 	raw_notifier_call_chain(&edev->nh_all, state, edev);
>>> 
>>> +	spin_lock_irqsave(&edev->lock, flags);
>>> 	/* This could be in interrupt handler */
>>> 	prop_buf = (char *)get_zeroed_page(GFP_ATOMIC);
>>> 	if (!prop_buf) {
>>> 
> 
> I have tested on the Pyra handheld prototype and now it works. Plugging in an OTG cable
> enables/disables OTG power as expected and there are no kernel oops any more.

I did take some minutes to check and it now also works again on the OMAP5EVM.

BR,
Nikolaus