Re: [lwip-users] Out of memory errors under load

[JM]

I'm not calling lwIP from main() loop; the main loop is empty. I've verified there's one entry point only, and actually right now the timeouts interrupt handler is commented out. So the only way into lwIP is with ethernet RX interrupt.

I've done more investigation and discovered something. I put a trap after calling ethernet_input() in the ethernet Rx function to catch if p->ref is ever not zero. When I ping or load the webpage manually, it never breaks. When I use the load tester, it hits that breakpoint because p->ref = 1. However, ethernet_input() also never returns anything but ERR_OK.

Looking into ethernet_input(), p->ref is sometimes 1 and p->tot_len = 63 after ip_input(p, netif) (in my situation anyway) BUT p->tot_len was 107 before ip_input(p, netif); was called! This seems odd to me as I thought the pbuf is completely consumed and should be freed after  ip_input(p, netif). Is this correct?

On Monday, March 26, 2018, 3:25:54 PM CDT, Sergio R. Caprile <address@hidden> wrote:

The "ping" does not call the application, the GET does.

I would argument your memory is getting freed slower than it is alloced,

but there have been numerous bug fixes and way so many enhancements in

the web server since 1.4.1 that you should really try git head.

Strange things can happen if you don't respect the threading rules. Are

you calling lwIP just from the main loop or do you use interrupts for rx

and the main loop for sys_check_timeouts() ? (don't!, either all main or

all interrupts, single thread, single context)

_______________________________________________

lwip-users mailing list

address@hidden

https://lists.nongnu.org/mailman/listinfo/lwip-users