Re: [lwip-users] Timers & lwip tasks (a multithread question with LwIP)

[Jonathan Larmour]

Frédéric BERNON wrote:
> Lwip needs some "timers" for ARP (etharp_tmr), DHCP (dhcp_fine_tmr &
> dhcp_coarse_tmr), SNMP (snmp_timer), etc...
>
> My question is mainly about arp timer, because the table handled in
> etharp.c is in read/write access. In several ports, I see that arp_timer
> is initialized by the main() task during the netif_add(...) call (with
> the "init" function). But when we use the tcpip_init, which create its
> own task, isn't it possible that the arp list was "corrupt" due to a
> multithreaded access?

Not every port can be considered accurate - in fact I doubt any of them are 
in one sense :-).

I don't know which ports you've looked at, but the eCos and UNIX ports (at 
least) wait until the tcpip task is definitely running,  by using the 
callback passed as the first argument to tcpip_init. That way the ARP timer 
is definitely set up afterwards, one way or the other. This could be done 
either directly in the callback at that point, or by some subsequent part 
of the initialisation function, which would have gone to sleep after the 
call to tcpip_init until woken up by that callback.

> And if it's right, isn't it a better place to initialize these timers
> directly in tcpip_thread(...), like the ip_timer for IP_REASSEMBLY, to
> be sure that only one task will use lwip core in the same time?

The tcpip_thread code is generic, and isn't aware of whether or when the 
arp timer should be started.

Jifl
--
eCosCentric Ltd  http://www.eCosCentric.com/  The eCos and RedBoot experts
  **  Visit us at ESC Silicon Valley <http://www.embedded.com/esc/sv> **
  **  April 3-5 2007, Booth 1922, San Jose McEnery Convention Center  **
Company legal info, address and number:   http://www.ecoscentric.com/legal
------["The best things in life aren't things."]------      Opinions==mine