** update: This problem is introduced in kernel 5.12.14. Reverting the meson-memcpy changes made in that patch makes the problem go away. I don’t know yet why a change to the handling of SD/MMC I/O would cause wifi packet corruption.
Hi. I may have found an odd and potentially subtle bug introduced in the recent update of fenix to use kernel 5.14-rc5
VIM3 v1.2, 4GB/32GB model.
Ubuntu-20.04 image downloaded from Khadas firmware page.
Linux-5.14-rc5 built with latest git pull of fenix in docker container, installed with dpkg -i
Noticed random DNS lookup lag and found packets being received by my DNS server from the VIM3 with bad checksums.
Packets are being seemingly randomly corrupted such that the last byte is intact but the 3 bytes before the last are overwritten with zeros. They look fine monitoring with Wireshark on the VIM3, but are corrupted when received on the remote end.
Testing with ping -s found that this only happens with packets from the default size up to -s 575, above that size there is no corruption.
This does not happen with wired ethernet, only with wifi, and does not happen with kernel 5.12 (built from fenix-1.0.7-release).
ping -f to a local host on 5.14-rc5 results in ~33% packet loss, on 5.12 it’s ~0.1%.
ping -f -s 576 on 5.14-rc5 yields a reasonable packet loss figure (<1%), in line with 5.12.
Tested with a couple different APs, on both 2.4 and 5 GHz bands, and observed no difference.