Hi Marcel,

On 04/01/2012 16:29, Marcel Holtmann wrote:
> Hi Guillaume,
>
>>>>>>>> and what about the case when the SIM card is present, but PIN locked?
>>>>>>>>
>>>>>>>>> According to the result, it might be interesting to send ATI when the
>>>>>>>>> constructor plugin is probe by oFono.
>>>>>>>>> Thus with +GCAP info we can decide which driver to use.
>>>>>>>> Is sending +GCAP after ATI really a standard? Have we tried anything
>>>>>>>> else besides Huawei or ZTE?
>>>>>>> I tried with more dongles from different vendors, as attached table.
>>>>>>> The scenarios include:
>>>>>>> With valid sim card, sim card PIN locked, no sim card, sim card locked.
>>>>>>> N(ROM) in table indicates the SIM in ROM already.
>>>>>>> ATI command can always return GCAP content in all tests.
>>>>>> and what about other manufactures other than Huawei, ZTE and SpeedUp?
>>>>>> What about Sierra, Ericsson etc.?
>>>>> Just checked Dell 5530 with Ericsson module,
>>>>> With SIM card or not, at+gcap can return +GCAP:+CGSM, +DS
>>>>> But the ATI only returns: D5530
>>>> I think it is clear that we need to do our homework here and properly
>>>> document the different manufacturers. Someone sending patches for our
>>>> doc/ directory?
>>> There're many vendors of 3G dongle..
>>> Huawei, ZTE (they share 70%+ of global market), Longcheer, Haier, Sentar, Viton, D-link, SCV, BandRich, Strongrising.. (more than 30 vendors in China)
>>> Sierra, Sony-Ericsson, Option, Novatel, Alcatel, Samsung, LG, AnyData, C-motech, Micromax...
>>> We can try with them step by step, but can we work out the 2 biggest firstly?
>>> Looks ATI command can work for both Huawei and ZTE dongles.
>>>
>> I agree here, the work to be done over all manufacturers will be
>> fastidious and might require a lot of dongles that we don't have currently.
>> Maybe we could do as Ying An proposed as we are sure ATI works for
>> Huawei and ZTE (at least the ones we have).
>> However, conerning ZTE I haven't seen any CDMA dongle for the moment.
>>
>>>>>>>> Also you do realize that the GAtChat object and thus the file descriptor
>>>>>>>> is owned by the modem plugin. The plugin itself is the only one that
>>>>>>>> should do any kind of IO.
>>>>>>>>
>>>>>>>> So if we require to run ATI first to identify if we are GSM or CDMA,
>>>>>>>> then this is a per modem manufacture specific detail. And we rather add
>>>>>>>> a helper function like we did for CPIN polling that makes this easier.
>>>>>>>>
>>>>>>> In current code the 'driver' is hardcoded by comparing with vendor_list[].
>>>>>>> So if it possible to break the step into several:
>>>>>>> vendor_list[] in udevng just cares about vendor - by comparing vendor ID
>>>> only,
>>>>>>> and add all possible drivers according to that vendor - (for example add
>>>>>>> WCDMA, CDMA2k, TDSCDMA, LTE ...drivers if Huawei dongle is plugged
>>>> in),
>>>>>>> and the probe interface in each driver does real probe work as to issue
>>>>>>> ATI command to ensure only correct driver will be loaded?
>>>>>> As I said before, the only time IO can be started is when the ->enable()
>>>>>> callback of the modem plugin is called. Not a second earlier.
>>>>> But if done after enable() called, from semantic aspect the correct driver has
>>>> been
>>>>> chosen. Indeed the probe() interface in each driver is not doing something to
>>>> probe,
>>>>> then can the work be done in probe()? As set CFUN=1 then doing some dongle
>>>> vendor
>>>>> specific work as query model or network mode by ATI, AT+GCAP command,
>>>> etc..? After
>>>>> that disable dongle when quit probe()?
>>>> The probe() callback is for accepting the driver and allocating required
>>>> local data structures. It is not for IO. And as you can see it has no
>>>> callback handling like enable() with set_powered().
>>>>
>>>> As I said before, no AT commands before enable() has been called. That
>>>> is how it is suppose to be. We are not changing this.
>> First, ATI command is working without sending AT+CFUN=1, we could keep
>> CFUN=1 into enable() as we do some vendor/modem type specific job there.
>>
>> Then vendor plugin can be chosen using udevng using Vendor ID, however
>> driver type (CDMA/GSM) can't lie on the Product ID. So it will be hard
>> to chose the right vendor plugin with right type.
>> And if we can't send AT command before enable() time we will face to bag
>> end e.g. :
>> For huawei plugin we send GSM specific AT command (AT^RFSWITCH) during
>> the enable() time.
>> We are also querying the sim state using polling mechanism that might
>> fail for CDMA modems that is not using SIM.
>> What would you suggest here?
> as I said before, no AT commands before ->enable() callback from the
> core.
>
> The callback ->probe() is for accepting the modem driver binding and
> allocating modem specific data memory. The callback ->remove() is for
> cleanup.
>
> The callbacks ->enable(), ->disable() and ->set_online() are the only IO
> entry points for every modem driver. And we need to keep it like this.

Ok, so I suggest to do the ATI at the very beginning of ->enable() callback.
Then depending on the ATI answer:
     - tag the huawei modem data with GSM / CDMA type.
     - send the GSM / CDMA specific AT commands followed by AT+CFUN=1.

We should not longer make the difference between "huawei" and 
"huaweicdma" into udevng.
It also means there is one unified plugin.

> The callback ->pre_sim(), ->post_sim() and ->post_online() are the entry
> points for selecting atom drivers. These are not allowed to do IO
> directly either.

Here, using the tagged type, the plugin can decide which driver to use 
to create atom.
Does it sound ok to you?

Kind regards,
Guillaume