Re: [lwip-users] arp question

[Leon Woestenberg]

Hello Jim,

forgot to answer the part on concurrency protection.

Jim Gibbons wrote:

> I agree that it is a simple matter to get etharp_output and the timer 
> to run out of the tcpip task.  I think it is substantially harder to 
> get etharp_ip_input and etharp_arp_input to run in that context.  Most 
> of us who implement netif's in multitasking environments end up with 
> an ethernet receive thread from which the etharp input functions must 
> be called.  Other brave souls do this from an interrupt context.  In 
> the former case, a mutex would provide the protection required, while 
> in the latter case SYS_LIGHTWEIGHT_PROT methods might be used to 
> protect the timer and output.
There are too many multitasking environments around to cope for. We hope 
the integrating developer has knowledge about
the multithreading issues involved for her environment. We came across a 
bunch when we merged lwIP into uCOS, such
as mutex-recursion, priority inheritance, nested interrupts.

Personally, I think the synchronization primitives in the outer API 
wrapper of lwIP are inadequately generic, but I also think
there is no such thing as a generic locking abstraction that works for 
everyone. POSIX would be great but is probably an
overkill in the context of lwIP.

Regards,

Leon.