Re: [lwip-users] port for lwip

[address@hidden]

ambarisha b wrote:
> I am working on a OS independent port for lwip. I have a couple of 
> issues with this.
Which version of lwIP are you using? I'm assuming 1.4.0 for the rest of 
this mail.
>  The function declarations for semaphores and mailboxes are different 
> in sys_arch.txt and from lwip/sys.h. For example:
>
> sys_sem_t sys_sem_new(u8_t count);
> err_t sys_sem_new(sys_sem_t *sem, u8_t count);
>
> Why is this difference? Am I missing something here?
No, it's us that have been missing something: updating the docs. Theses 
functions have been changed to take a pointer so that sys_sem_t can be 
directly mapped to a preallocated OS-struct. Before this change 
sys_sem_new() could only return a pointer to such a structure that had 
to be allocated somewhere else. This should make OS ports both smaller 
and easier. I'll update the docs in CVS.
> Another is that sys_mbox_post takes the pointer to the message. But 
> the size of the message isn't passed. If the size of the message is 
> passed in the buffer it self, it hasn't been documented. So, how does 
> this work?
An mbox always contains 'void*' messages. The function that reads 
messages from the mbox knows which type is allowed for this mbox and 
casts to the correct type. If multiple messages are allowed in an mbox, 
we pass a struct that contains type information. Since only pointers are 
passed through the mbox, allocation/deallocation is taken care of 
outside the mbox implementation.
> A couple of functions require that time spent waiting, be returned in 
> milliseconds. How accurate should this be? What are the practical 
> limits that you would suggest?
First, you never should return '0' if you waited, even if you spent far 
less than a millisecond waiting. As to accuracy, that's difficult to 
say. It should be as accurate as you can get it to, but just returning 
OS ticks here (which are normally in a range of 1 to 20ms) should 
probably be good enough...

Simon