Last week I managed to get LwIP 2.1.3 running on an STM32F767. There are a bunch of issues with the ST sample code for the F7, actually for anything other than the H7s.
The setup of the MPU in the examples is just plain wrong. The proper start address should be 0x2007C000 for 16k and 0x2007C000 for 512 bytes, get rid of the setup for 4G at 0x0. The descriptors should be placed at 0x2007C000 and 0x2007C0A0. The RX buffer should be placed at 0x2007C200.
The RX buffer needs to be between 4 and 8 structs long, according to the comments, it is 12 and will not fit in any 16kB region (since it is 19kB in the examples). By reducing the size to 8, it comes out to 12kB and will fit into the non-cached, non-buffered region set up with the MPU.
The top of stack should be relocated BELOW 0x2007C000 or else stack growth will smash into the buffers and descriptors. They are descriptors, not decryptors.
LwIP is...LwIP. The documentation is hard to read because the code was hard to write. There are few to no examples, well there are the unit tests, and the same 2 ST examples (HTTP and echo). If you have problems with LwIP, the various forums will give you a timeout before an answer. Perhaps nobody, even in the developer group, actually uses LwIP. The only sample code you'll find is reposts of the older distributions. The LwIP people seem to not like RTOS', so unless you're banging chunks of silicon together, you're on your own.
So, it does work, but it's going to be a rough ride.
