[lwip-users] Re: [lwip] pcb->queuelenbug not solved?

[Adam Dunkels]

Hi Andrea!

On Thursday 24 January 2002 20.53, you wrote:
> - I tried again my porting (CVS 20020121) with the last patches in pbuf.c
> and tcp_output.c.
> It is going better, but sometimes happens that pcb->queuelen become 255 or
> 254. I manage to work the stack with a simple dirty modification, that
> forced to zero pcb->queuelen when it is more than 250 (very dirty...). I'll
> test better tomorrow, but it seems to work.

Hmmm, I have been running my test code for quite some time without any 
problems. I was using the sys_timeout() function to call a function every 100 
ms that sent some data (just as you described your setup). Perhaps you could 
post some code that triggers the bug?

> - Always the same problem with Keil: please initialize to 0 p->ref and
> p->flags in pbuf_init.

Is that really needed? Those two fields are filled by the pbuf_alloc() 
function, so there should be no need to initialize them to zero. What kind of 
problems have you experienced?

> Are there any other function or piece of code (alloc, free, ...) that is
> based on the assumption that the compiler initialize to 0 every new
> variable? Let me know.

I'll go through the code and see what I'll find.

Stuff like this can be hard to find, especially since gcc automatically 
clears the bss segment, so your comments and findings are very interesting. 
By the way, does anybody know some way to make gcc fill the bss segment with 
random data for testing things like this?

> - Testing with nmap now it is working for both lwip and iup.

Great!

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