[lwip-users] Re: [lwip] memp_malloc

[Adam Dunkels]

On Thursday 07 March 2002 10.23, you wrote:
> > Why does the code use memp_malloc at some places and memp_mallocp at
> > others . I think if the code has to be run in mutithreading environment
> > then memp_mallocp must always be used since this function is gaurded by
> > semaphores .
>
> ...unless you know that resource will only be accessed by one thread.
> Having had a look at the code (I don't know if this is right, but it looks
> this way to me) the only places that use memp_malloc are in api_lib.c and
> they are all requesting memory of the same type (MEMP_API_MSG).  This type
> is not requested from anywhere else, and the code in api_lib.c is always
> executed by the same thread (I think!), so by not protecting it with
> semaphores you should get a significant increase in performance.  (API
> msgs are still a big bottleneck, but it'd be worse otherwise)
>
> Could all be wrong though.  Adam will be along shortly to correct me...

You are right (as usual :-). But, as Mumtaz points out, code run in different 
threads must be protected by semaphores (as the mallocp/freep functions 
does). Since the api_lib code can be executed by different threads in 
parallell, they *should* in fact be protected.

Bug in lwIP.... new release coming out shortly :-)

/adam
-- 
Adam Dunkels <address@hidden>
http://www.sics.se/~adam
[This message was sent through the lwip discussion list.]