LwIP corrupts memory - How to configure Ethernet and LwIP Addresses correctly?

[Sandro_K]

Hello,
I use a STM32F777 with FreeRTOS and LwIP. I also run a display control with TouchGFX and partial framebuffer via SPI. I have configured the Ethernet and LwIP as shown in the following link: https://community.st.com/t5/stm32-mcus/implementing-a-lwip-and-freertos-v1-udp-echo-server-on-the/ta-p/780261 .

The Ethernet works so far after booting. However, there are always artifacts on the display, as LwIP seems to write to the areas of the frame buffer.

However, I have created the Linkerscript configuration as in the tutorial from the link and also added a LwIP heap section. Unfortunately it still does not work stable.
What do I have to adjust so that it works without errors?

Here is my configuration:

[STackPointer64]

Hello @Sandro_K,

Based on your description, I recommend the following approach to help isolate and resolve the issue:

Start by testing the Display Control and framebuffer functionality without enabling Ethernet. This will help determine if the artifacts are related to the display or framebuffer itself, independent of the Ethernet and LwIP operations. If the display works correctly without Ethernet, it suggests that the problem arises when Ethernet is active.

In that case, please try to reproduce the problem, and if possible, provide a minimal example of your project illustrating this issue. This will help in diagnosing whether there is a memory overlap, improper synchronization, or other conflicts between Ethernet/LwIP and the framebuffer. If the project is private, feel free to send it to me privately.

Best regards,

 

[Sandro_K]

Hello STackPointer64,

thanks for your reply.

The display control works without any problems. If the Ethernet cable is not plugged in, the application runs for hours without any problems.
I have a web server on the STM that cyclically reloads website content via GET requests every 500ms. This of course generates traffic. The problem only occurs when the website is connected to the STM and the data is reloaded cyclic.

I will prepare a minimal configuration of the project and make it available here.

[MDNev]

Hi,

I seem to have the exact problem you're encountering.  If the task which does GET requests is a higher priority than LwIP's "tcpip_thread", I will eventually get stack corruption.  If the GET request task is a lower priority, I do not get the corruption.  This is a brittle solution!  The Wireshark capture shows a sequence of about 6 or 7 TCP retransmissions right before the memory corruption is detected.  I don't remember if it is the STM or web server issuing the retransmission.  My settings appear nearly identical to yours.  Were you able to determine the root cause of the corruption?  I setup a 128K MEM_HEAP area in MPU region 1 and 1k area in MPU region 2 for the RX/TX descriptors.

[MDNev]

I spoke too soon.  I just had the same corruption with the task priorities set as I mentioned.  I think it was related to TCP_WND_UPDATE_THRESHOLD.  I changed that to 1460 bytes from 536 bytes.  Waiting to see the results.  I still get the occasional packet loss, but it hasn't crashed yet.