[ECOS] ecos + lwip

[Simon Kallweit]

Mandeep Sandhu wrote:
>>>> <snip>
>>>>
>>>> cyg_lwip_sequential_init();
>>>>
>>>> cyg_lwip_thread_new( "My thread", m_main, (void *)"My thread",
>>>> my_stack, 0x1000, 5 );
>>>> ...
>>>> ...
>>>> <snip>
>>>>
>>>> I'll enable more debugs to see where it's getting stuck.
>>> After changing it to 2, it's able to find the interface (et0) with all
>>> the right settings.
>> Well, that does not really make sense. cyg_lwip_sequential_init() will not
>> return until the lwip stack is initialized, and initializing ethernet device
>> drivers is part of that. This *must* work even when the TCPIP thread has
>> lower priority, as long as you start your application thread *after* the
>> execution of cyg_lwip_sequential_init(). By starting *after* I mean that you
>> also have to make sure that cyg_lwip_sequential_init() and starting your
>> applications thread is done in the same thread!
>>
>> I hope I didn't miss the obvious here, but that's how it should work.
> 
> Well, I checked the code again. The netif init is done as part of
> tcpip_init_done().
> This I see is being called in tcpip_thread(). This function is being
> executed in the
> tcpip thread's context.
> 
> I guess this is why the priority of the threads matters.
> 
> Would do you think?

There is a semaphore created in cyg_lwip_sequential_init() which will 
block until all initialization (including ethernet devices) in the tcpip 
thread is done. Therefore it's priority does not matter. 
cyg_lwip_sequential_init() should not return before initialization is 
completed.

I just reviewed the code and found that the semaphore is initialized 
with 1, which is clearly a bug, it should be initialized with 0 so that 
the following call to cyg_semaphore_wait() is actually blocking the 
thread until initialization is done.

My mistake, sorry! Hope it works as intended now. New release:

http://download.westlicht.ch/lwip-20090903.tar.gz

Simon

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss