Re: [lwip-users] pbuf pool size / mss size in low memoryenvironmentand routing to slow link

[David Empson]

"Simon Goldschmidt" <address@hidden> wrote:
> "Marco Jakobs" address@hidden wrote:
> > I'm just wondering how the PPP of LWIP is handling the pbuf's, maybe
> > someone can drop a line about this ...
>
> I don't know that by heart but I think I remember the PPP code (as it is 
> ported from sources that don't know pbufs) needs the input packets in
> contigouous memory and copies from PBUF_POOL to PBUF_RAM if the input 
> packet is a linked list of pbufs (i.e. p->len != p->tot_len).

I can't see any evidence of that.

The code appears to be correctly dealing with pbufs that are smaller than 
the MTU.

On the transmit side, the IP packet supplied by LWIP is processed by 
pppifOutput(). It steps through each pbuf in the supplied chain (relying on 
pb->next = NULL to detect the end of the packet). The IP packet is copied 
into a new chain of pbufs (of type PBUF_RAW), adding the PPP framing and 
control character escaping. The expanded PPP frame could be up to twice as 
big as the MTU in the worst case. The transmit code then steps through the 
pbufs holding the PPP frame and writes each of them in turn.

The receive side is reading one byte at a time and state machine allocates 
new pbufs as required to hold the data, allocating a new pbuf each time it 
has accumulated PBUF_POOL_BUFSIZE bytes. pbuf_cat() is used to attach each 
pbuf to the existing chain.