[PATCH 09/10] libbsd.txt: Describe current state of WLAN.

[Chris Johns]

On 09/11/2017 19:17, Christian Mauderer wrote:
> Am 09.11.2017 um 09:03 schrieb Chris Johns:
>> On 09/11/2017 17:41, Christian Mauderer wrote:
>>>
>>> currently there is still no rc.conf support. I took a look at it and
>>> found out that I can't do it in the scope of the project. Therefore I
>>> only added a note regarding this in the "known restrictions".
>>
>> Hmmm.
> 
> Yes, I know. Not really a solution. But please keep in mind that WiFi on
> RTEMS is a work in progress. It can't be expected that from "no WiFi" to
> "WiFi completely integrated" is only one step.
> 

Understood. A ticket is a good to capture any details you have and to
acknowledge this is outstanding.

>>
>>> In FreeBSD, the creation of such devices would be handled by rc.conf
>>> lines like
>>>     wlans_rtwn0="wlan0"
>>>     ifconfig_wlan0="DHCP"
>>>
>>> It wouldn't be a great problem to parse such lines in rc.conf and call a
>>> matching ifconfig command. 
>>
>> The rc.conf API we have should be able to handle something like that or it
>> should be close.
> 
> I would have expected that too. Therefore: That one shouldn't be a big
> problem.
> 

Agreed.

>>
>>> But there is a problem due to the nature of
>>> the rtwn0 device: It's an USB device. Due to that, it isn't there right
>>> from the start. Depending on quite a number of conditions it will appear
>>> sooner or later during the startup or - in the worst case - a user could
>>> plug it after a few minutes of run time.
>>
>> Understood, enumeration takes time.
>>
>>>
>>> Please correct me if I'm wrong but as far as I know, the rc.conf support
>>> in RTEMS currently is only executed once during start without
>>> remembering any configuration for later use. The source can be either a
>>> file or a string.
>>
>> Is rc.conf run more than once in FreeBSD? I know the `service` command uses it
>> to know if a service is enabled, ie onestart etc. Is there something else?
>>
> 
> I'm no FreeBSD user and I only have a rough knowledge about how things
> are done there. From my Linux-point-of-view I would expect that there is
> something similar to udev that does some actions as soon as a new device
> is discovered. I just somehow expected that in case of a network device
> this action on FreeBSD would be to look in the rc.conf whether there is
> something to do with this device. But I can be wrong too.
> 

I suggest an old PC with a USB port and a cheap USB Wifi adaptor FreeBSD
supports. Install FreeBSD which is easy and then hack the init scripts adding
`set -x` where you think it is needed. I have FreeBSD on a 8G USB stick for this
type of testing.

I would have a look but I have misplaced my mini USB wifi adaptor.

>>> With that in mind, I would expect that roughly the following would be
>>> necessary for that implementation:
>>>
>>> - To handle dynamically added WiFi and network devices through rc.conf,
>>> either the complete rc.conf string or at least the information from the
>>> wlans_xxx and ifconfig_xxx lines has to be kept somewhere for later use.
>>>
>>> - Some kind of handler or daemon for detecting new USB devices would be
>>> necessary. As far as I know, there is currently no such thing. The only
>>> other application with dynamic device detection that I know of is the
>>> media listener. But I think that is on block device level and not on the
>>> USB level.
>>
>> I am confused about the way FreeBSD does this. I did not think rc.conf is rerun
>> at all. As I said it is parsed by lots of things. We should follow the same
>> model if that is feasible. If you could determine what FreeBSD does it would go
>> a long way to knowing which path is the best path to follow.
> 
> Like said above, I would expect rc.conf to get re-parsed for new network
> devices. But I'm not sure about that.
> 
> We can't follow that model exactly. As far as I know, we can have at
> least two sources for our rc.conf: Either a string or a file. In case of
> the string, we would have to copy that string so we can parse it later
> again. For big configs, that might wouldn't be the best idea.

I cannot say until I understand what FreeBSD does. This needs to investigated.

> 
> A better solution could be to prepare some lists with configurations for
> devices. These lists could then be used on demand.
> 

I do not follow what this means.

>>
>>> I would have liked to add that integration to have a clean solution.
>>
>> I agree on needing a clean solution but it is not clear yet what that is.
>>
>>> But
>>> like I said, there isn't enough free budget left in that project to do
>>> it. I really hope that we can find some time and budget for it in the
>>> future.
>>
>> I respect and thank you for being open about this. I view getting a suitable
>> solution for rc.conf is important for the project because this stuff is
>> complicated to get working. I have raised rc.conf before and I had hoped it was
>> understood we need to support it. It is crazy to require users to have some
>> parts in rc.conf and other parts in complex fragile custom code to make this
>> stack work. This has been a valid criticism of RTEMS for years and I would like
>> to see us address it.
> 
> Agreed. I would prefer the configuration in rc.conf too. 

Excellent.

> But again: WiFi in RTEMS is (from my point of view) still a work in progress.

Understood.

> 
>>
>> Could you please determine what rc.conf work is needed and document it in a
>> ticket? I feel you are best to do this since you are across the code in detail.
>>
> 
> Good point. I'll create a ticket.
> 

Thanks
Chris