I'm having problems with an LWIP TCP client on STM327508DK. The communications fail after a few messages.

[JColl.2]

Hi all,

I am working with a stm32f7508dk. My code is running XIP on 1-bank QSPI.

Here is the MEMORY definitions in my linker file:

   MEMORY

   {

     RAM   (xrw)   : ORIGIN = 0x20000000,  LENGTH = 320K

     /* FLASH   (rx)   : ORIGIN = 0x8000000,  LENGTH = 64K */

     FLASH (rx)     : ORIGIN = 0x90000000, LENGTH = 16384K

   }

Here is the setup in main()

    MPU_Config();

    SCB_InvalidateICache();

//    SCB->CCR |= (1 <<18);    /* Enable branch prediction */

//    __DSB();

    SCB_InvalidateICache();

    SCB_EnableICache();

    SCB_InvalidateDCache();

    SCB_EnableDCache();

I've implemented USB CDC, and LWIP Tcp_echo_client and Tcp_echo_server as Tcp_client and Tcp_server respectively. These are implemented as raw API, no OS.

Tcp_server opens a listen socket and receives messages only rarely -- used to configure the system. It seems to work properly. I see the failure described below whether I've received any messages on the listening port or not.

The system receives data from USB which triggers a message to be sent from the Tcp_client to a database at a specified address. The data-reception and message trigger occurs about once every 3-5 seconds.

Based on various memory configurations, the messaging will fail after as few as 30 messages, or as many as 300 messages, but it always fails.

Note, each message has about 3 frames of data. SYN/ACK ... data .. FIN/ACK look proper using wireshark to analyze on the receiving computer.

Related settings in lwipopts.h:

#define TCP_STATS                    1

#define TCP_MSS                       536

#define MEM_SIZE                       0

#define MEM_LIBC_MALLOC                1

#define MEM_USE_POOLS                  0

#define MEMP_MEM_MALLOC                0

#define TCP_DEBUG                      LWIP_DBG_ON

monitoring TCP debug messages, the messages start as:

   tcp_recved: received 242 bytes, wnd 2144 (0).

   tcp_recved: received 187 bytes, wnd 2144 (0).

   tcp_recved: received 45 bytes, wnd 2144 (0).

   tcp_close: closing in State: ESTABLISHED

   TCP connection closed: FIN_WAIT_1 27017 -> 52433.

   tcp_pcb_purge

   tcp_slowtmr: no active pcbs

   tcp_slowtmr: no active pcbs

   tcp_connect to port 27017

after about 4 messages I see:

   tcp_alloc: killing off oldest TIME-WAIT connection

   tcp_kill_timewait: killing oldest TIME-WAIT PCB 0x20027ca0 (28)

   tcp_connect to port 27017

   tcp_recved: received 242 bytes, wnd 2144 (0).

   tcp_recved: received 187 bytes, wnd 2144 (0).

   tcp_recved: received 45 bytes, wnd 2144 (0).

   tcp_close: closing in State: ESTABLISHED

   TCP connection closed: FIN_WAIT_1 27017 -> 52437.

   tcp_pcb_purge

   tcp_slowtmr: no active pcbs

   tcp_slowtmr: no active pcbs

   tcp_slowtmr: no active pcbs

which, as I understand, is acceptable -- killing and reusing conversations that are finished but not yet timed out.

After a while, I start seeing additional entries announcing 'processing active pcb' and 'polling application' which seem to get more prevalent over time.

   tcp_alloc: killing off oldest TIME-WAIT connection

   tcp_kill_timewait: killing oldest TIME-WAIT PCB 0x20027c04 (19)

   tcp_connect to port 27017

   tcp_recved: received 242 bytes, wnd 2144 (0).

   tcp_recved: received 187 bytes, wnd 2144 (0).

   tcp_slowtmr: processing active pcb

   tcp_slowtmr: polling application

   tcp_recved: received 45 bytes, wnd 2144 (0).

   tcp_close: closing in State: ESTABLISHED

   TCP connection closed: FIN_WAIT_1 27017 -> 52574.

   tcp_pcb_purge

   tcp_slowtmr: no active pcbs

   tcp_slowtmr: no active pcbs

More:

   tcp_connect to port 27017

   tcp_slowtmr: processing active pcb

   tcp_slowtmr: polling application

   tcp_slowtmr: processing active pcb

   tcp_slowtmr: polling application

   tcp_recved: received 242 bytes, wnd 2144 (0).

   tcp_recved: received 187 bytes, wnd 2144 (0).

   tcp_slowtmr: processing active pcb

   tcp_slowtmr: polling application

   tcp_recved: received 45 bytes, wnd 2144 (0).

   tcp_close: closing in State: ESTABLISHED

   TCP connection closed: FIN_WAIT_1 27017 -> 52644.

   tcp_pcb_purge

   tcp_slowtmr: no active pcbs

At some point I start seeing:

   tcp_fasttmr: delayed ACK

Shortly after, an endless loop of the following:

   tcp_alloc: killing off oldest TIME-WAIT connection

   tcp_alloc: killing off oldest LAST-ACK connection

   tcp_alloc: killing off oldest CLOSING connection

   tcp_alloc: killing oldest connection with prio lower than 64

   tcp_alloc: killing off oldest TIME-WAIT connection

   tcp_alloc: killing off oldest LAST-ACK connection

   tcp_alloc: killing off oldest CLOSING connection

   tcp_alloc: killing oldest connection with prio lower than 64

Using MEM_LIBC_MALLOC 1 allowed for many more messages.

I've seen comments pointing toward problems with dcache, DTCM, pool configuration.

Can anyone point me toward a guide on setting up LWIP specifically for the STM32F750x8?

Thanks,

Jim C

[Walid ZRELLI]

Hello @JColl.2​  ,

Memory allocation might be the problem here because you are opening TCP connections relatively frequently.

You could possibly set up a single TCP connection instead of a per transfer connection and see what the result is.

I hope this helps.

BeST Regards,

Walid