Hi Ulisses,
On Tue, Feb 7, 2012 at 12:20 PM, Ulisses Furquim <ulisses@xxxxxxxxxxxxxx> wrote:
> Hi Andre,
>
> On Tue, Feb 7, 2012 at 12:46 PM, Andre Guedes
> <andre.guedes@xxxxxxxxxxxxx> wrote:
>> Hi Ulisses,
>>
>> On Tue, Feb 7, 2012 at 10:26 AM, Ulisses Furquim <ulisses@xxxxxxxxxxxxxx> wrote:
>>> Hi Andre,
>>>
>>> On Tue, Feb 7, 2012 at 10:20 AM, Andre Guedes
>>> <andre.guedes@xxxxxxxxxxxxx> wrote:
>>>> Hi Lizardo,
>>>>
>>>> On Tue, Feb 7, 2012 at 7:48 AM, Anderson Lizardo
>>>> <anderson.lizardo@xxxxxxxxxxxxx> wrote:
>>>>> Hi Andre,
>>>>>
>>>>> On Tue, Feb 7, 2012 at 12:29 AM, Andre Guedes <aguedespe@xxxxxxxxx> wrote:
>>>>>> diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c
>>>>>> index 6808069..3933ccd 100644
>>>>>> --- a/net/bluetooth/hci_event.c
>>>>>> +++ b/net/bluetooth/hci_event.c
>>>>>> @@ -3255,12 +3255,10 @@ static inline void hci_le_adv_report_evt(struct hci_dev *hdev,
>>>>>> void *ptr = &skb->data[1];
>>>>>> s8 rssi;
>>>>>>
>>>>>> - hci_dev_lock(hdev);
>>>>>> -
>>>>>
>>>>> So there is no need to lock hdev between the hci_add_adv_entry() and
>>>>> mgmt_device_found() calls? This looks different from what is done for
>>>>> BR/EDR for the inquiry cache.
>>>>
>>>> Yes, mgmt_device_found() does not require locking hdev->lock.
>>>
>>> We could then move the lock and unlock calls to inside the loop. But
>>> as we might have more than one call to hci_add_adv_entry() it'd be
>>> good to lock and unlock only once, no? Any problems I don't see?
>>
>> Yes, that's right. For this particular case, it may be better to lock
>> hdev outside while() and call the thread-unsafe version here.
>>
>> This way, it may be better we just drop patches 02/03 and 03/03.
>>
>>>>>> while (num_reports--) {
>>>>>> struct hci_ev_le_advertising_info *ev = ptr;
>>>>>>
>>>>>> - __hci_add_adv_entry(hdev, ev);
>>>>>> + hci_add_adv_entry(hdev, ev);
>>>>>>
>>>>>> rssi = ev->data[ev->length];
>>>>>> mgmt_device_found(hdev, &ev->bdaddr, LE_LINK, ev->bdaddr_type,
>>>>>> @@ -3268,8 +3266,6 @@ static inline void hci_le_adv_report_evt(struct hci_dev *hdev,
>>>>>>
>>>>>> ptr += sizeof(*ev) + ev->length + 1;
>>>>>> }
>>>>>> -
>>>>>> - hci_dev_unlock(hdev);
>>>>>> }
>>>>>>
>>>>>> static inline void hci_le_ltk_request_evt(struct hci_dev *hdev,
>>>>>> --
>>>>>> 1.7.9
>>>
>>> While I don't see anything wrong with your changes I don't think we
>>> really need it. All the other functions that need to be called with
>>> hdev->lock held don't have "__" prefix so it'll be different than the
>>> others. And you added 3 new locked functions but your last patch only
>>> uses 2 of them and only in 2 places. Unless I'm missing something here
>>> we don't really need this refactoring at all. Do you have any other
>>> reason to do that? Are you gonna use those functions in other
>>> patchset?
>>
>> Yes, some other functions don't have the prefix "__" and that fact
>> makes a bit painful and error-prone since we always have to dig in
>> the "call chain" to know if we need to hold hdev->lock or not.
>> Prefixing a function with "__" is just a standard way to indicate
>> that.
>
> I understand that. I just don't know if Marcel will want to change
> them all to have "__" prefixes, though. Having only one subset with
> this prefix can make things even more confusing, don't you agree?
Yes, this is pretty much an RFC series, I just realized I missed
changing the --subject-prefix.
> And have you been working with these functions or is this just a
> cleanup you thought it'd be good to do?
Just a cleanup.
Andre
--
To unsubscribe from this list: send the line "unsubscribe linux-bluetooth" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
