Re: [lwip-users] Bug in pbuf.c regarding PBUF_POOL

[Jonathan Larmour]

Pettinato, Jim wrote:
> Leon et al, I agree completely... If you have a pre-emptive OS, the
> interface need only signal it needs serviced; a high-priority task can
> handle feeding packets into the stack. Without such a prioritized
> platform though, performance would be severely degraded if handled any
> other way than directly in the ISR.
>
> I also agree that I wouldn't push a packet all the way up the stack
> directly in the ISR... 

Just in case, no-one so far has been suggesting passing the packet into the 
stack from an ISR, but I think the rest of your mail shows you do know that :).

I'm not sure a high priority task is sufficient, if that task can block for 
an indeterminate time waiting for a lower priority task (semaphores suffer 
from priority inversion). So it really makes no difference whether it's 
called from a high priority thread or a bottom half - we don't want to 
block either way round.

If PBUF_POOL pbufs were only allocated from a single context (and different 
bottom halfs would be sufficient for this, for example), this could all be 
avoided; but they aren't. It's not clear to me why many of these calls are 
from PBUF_POOL, rather than PBUF_RAM.

Incidentally, having to have a bunch of threads around to handle these 
sorts of issues is a big waste due to the extra thread control blocks and 
thread stacks - precious memory in the context in which lwIP is intended to 
be used. Personally, I'm changing lwIP not to rely on such a model for 
ethernet at least - for me it's a simple change in fact.

Jifl
--
eCosCentric    http://www.eCosCentric.com/    The eCos and RedBoot experts
------["The best things in life aren't things."]------      Opinions==mine