> > > > Hi Andrey, > > > > > It simple a) don't apply to master tree; b) not working - MAX_RX_LENGTH > > > constant not > > > touched, GSW_REG_GMACCR not tweaked for accepting jumbo frames. > > > > a) Sorry, I missed your answer and thanks for the review. Yes, it was > > not applying to master. It was based on another patch deeper in my > > tree. I fixed that a long time ago but I postponed the fix as I didn't > > see any interest from ML (I was wrong). I was focusing on upstream > > patches (realtek dsa switches merged into 5.18) and postponed the > > ramips fixes for when they would be needed. > > > > b) MAX_RX_LENGTH seems to be used only for keeping the buffer at least > > over (MAX_RX_LENGTH - FE_RX_ETH_HLEN) and it does not care when mtu is > > actually even bigger than (MAX_RX_LENGTH - FE_RX_ETH_HLEN). See: > > MAX_RX_LENGTH affects buffers for DMA transfers.
Is it because when you unconditionally enable jumbo frames in switch, you need to prepare the ethernet driver to receive a larger frame, even when mtu is 1500 or smaller? > > > c) MAX_RX_JUMBO and MAX_RX_PKT_LEN in GSW_REG_GMACCR are switch regs > > while I touched only ethernet frame engine regs. > > There are two parts - FE and switch. And both need to be properly programmed. > > Attached patch tested on real MT7620 with internal switch and it is > able to operate with a 2048-bytes frame (included single VLAN tag). Are you ignoring CONFIG_SOC_MT7621 support for this driver? (it is indeed upstream for recent kernels). The patch worked as expected! Thanks a lot! Do you plan to submit it? Tested-by: Luiz Angelo Daros de Luca <luizl...@gmail.com> _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel