Jan Menzel: > Did you check the performance for your system? I suppose with buffers in > non-cached memory you do not have any benefits of the d-cache. If you > can't get any of the options I suggested to run, you still might use > dedicated non-cached RX buffers in the driver and copy received data > into cached LWIP memory. This way the processing of all the headers will > still benefit from the d-cache even though you have the extra work of > coping the data.
My current focus is to improve overall system performance e.g. data processing using float maths, interrupt handling, SPI communication etc. This 'core' is about 5x faster if I can enable the cache. Network performance is not that important - yet. I've implemented cache management by using explicit DMA cache flush / invalidate by address for XDMAC/SPI but at the moment I do not have the time and especially knowledge in lwip to implement that for networking too. Therefore I decided to use lwip code 'as is' and only placed descriptors and buffers into non-cached area what seems to be at least not slower than before and suits my needs at the moment. Therefore I did not spent much time to study or implement all options discussed here but I really appreciate any comments in order to improve networking in the near future. To achieve this I'll have to update lwip to 2.x first. For productive system I'm afraid that I have to claim on 1.4.1 for now. _______________________________________________ lwip-users mailing list lwip-users@nongnu.org https://lists.nongnu.org/mailman/listinfo/lwip-users
