Re: [PATCH qmi lte v3 5/6] qmi: stop listening to packet service notifications

[Denis Kenzior <denkenz@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2103 bytes --]

Hi Jonas,

On 04/17/2017 02:45 PM, Jonas Bonn wrote:
> On 04/17/2017 07:19 PM, Denis Kenzior wrote:
>> Hi Jonas,
>>
>> On 04/15/2017 10:34 AM, Jonas Bonn wrote:
>>> The packet service status notification tells whether there is _any_
>>> active and enabled context.  Using this to decide whether to
>>> release any given context makes no sense.
>>
>> The gobi driver only supports a single context and afaik only a single
>> network interface is available.  Do you have hardware with support for
>> multiple active contexts?
>
> On an LTE network you can start a "dedicated" bearer.  As far as I
> understand it, the way this is done is to call "start network"
> specifying a profile ID other than the default.  In the ofono context, I
> presume this would be done by creating a new context and activating it:
> it just needs the same APN as the default bearer, and normally one would
> specify other QoS settings.  Given that, then yes, I presume my hardware
> support multiple active contexts; if the Gobi driver only supports one
> context it's presumably because it doesn't support LTE.

Okay, so now you're opening a whole new can of worms.  Dedicated bearers 
are similar to secondary PDP contexts in 2G/3G.  Their handling is even 
more special than primary PDP contexts / default bearers.

I wouldn't presume that your hardware supports this.  You need to check. 
  How many physical network interfaces are exposed?  If its just one 
like the gobi has, then you have your answer.

>
>>
>> Otherwise I think it would be safe to assume that in a single-context
>> case, this code is doing the right thing.
>
> I think, for LTE, that the netreg atom gives you everything you need and
> that this notification is superflous.  I'm thinking now that for non-LTE
> this code might be relevant:  can the context detach spontaneously
> without unregistering from the network?  If yes, then we probably need
> this, still.
>

A context can be forcefully deactivated by the network operator at any 
time.  So yes, this is actually needed.

Regards,
-Denis