HardFault with lwip+freeRTOS and STM32f745

[twagner]

Posted on April 18, 2016 at 11:02

Hello,

I'm using the lwip stack in combination with FreeRTOS for a http server (Netconn API) and modbus TCP (FreeModbus). I'm using the lwip stack and FreeRTOS from the ''STM32Cube_FW_F7_V1.3.0''. Evriting works great for 10-20 min but after that time the processor crashes with a HardFault. The HardFault happens in the function tcp_output at the line ''if (TCP_SEQ_LT(ntohl(seg->tcphdr->seqno), ntohl(useg->tcphdr->seqno))) {''. The processor wants to load ''seg->tcphdr->seqno'' and ''useg->tcphdr->seqno'' but one pointer is not valid and the processor crashes at the address ''0800d50a'' with a HardFault and the flag unaligned memory access is set. I activated the lwip_stats and get 60 errors at the memp[6]. In the lwipopts.h the SYS_LIGHTWEIGHT_PROT ist set and I teset different configuration in the lwipopts.h but noting helped. Where can the problem come from? Here my disassembly: 1010              if (TCP_SEQ_LT(ntohl(seg->tcphdr->seqno), ntohl(useg->tcphdr->seqno))) { 0800d4fc:   ldr r3, [r4, &sharp12] 0800d4fe:   ldr r0, [r3, &sharp4] 0800d500:   bl 0x80097d4 <lwip_ntohl> 0800d504:   mov r6, r0 0800d506:   ldr.w r3, [r8, &sharp12] 0800d50a:   ldr r0, [r3, &sharp4] 0800d50c:   bl 0x80097d4 <lwip_ntohl> 0800d510:   subs r0, r6, r0 0800d512:   cmp r0, &sharp0 0800d514:   bge.n 0x800d546 <tcp_output+526> 1012                struct tcp_seg **cur_seg = &(pcb->unacked); #lwip #hardfault #stm32f745

[Tesla DeLorean]

Posted on April 18, 2016 at 11:18

It looks like useg-tcphdr is not aligned. You could sanity check that it is 4-byte aligned, and memcpy out seqno, but this might just move the problem down stream.

[twagner]

Posted on April 18, 2016 at 12:06

I activated the MEMP_DEBUG and get the message:

memp_malloc: out of memory in pool TCP_PCB I increased the MEMP_NUM_TCP_PCB (for test purposes to 1000

http://www.dict.cc/englisch-deutsch/purposes.html

) but the meassage comes only later.

[Tesla DeLorean]

Posted on April 18, 2016 at 13:51

A bit scary that it doesn't recognize calloc/malloc return NULLs

So perhaps you have a resource leak, or are getting heap fragmentation. The former you'd want to chase down by understanding who allocated the resource, and then that all exit/return paths handle the release properly. For the latter, a fixed size pool method, where different size allocations a pull from a size appropriate pools, so small but longer term allocations don't bisect large blocks of otherwise free memory.

[sayagyimyo]

Posted on August 16, 2016 at 05:54 hi sir

I am myo from sg. I am trying to run Modbus tcp server on stm32 f207zg.

do you have source code for your Modbus tcp project.

can you share with me.

my mail is

mailto:sayagyimyo@gmail.com

thanks in advance.