Hi Fabio, On Mon, Jun 29, 2020 at 10:26 AM Fabio Estevam <[email protected]> wrote: > > Just tested 5.4.24_2.1.0 on an imx6qp sabresd and DHCP also fails there.
I think I discovered the problem ! When I compare the sabresd devicetree on mainline with the actual sabresd schematics, the devicetree is incorrect ! Things still work, but only by accident. The sabresd has an AR8131 PHY, which generates the enet ref clock, not the imx6. So on the schematic we see that the clock output of the PHY is wired to imx6 ENET_REF_CLK, so it can be used as a clock source. And GPIO_16 is disconnected, as it should, because the imx6 is not generating the ref clk. But the devicetree is written as if the imx6 is providing the clock ! See here: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/arch/arm/boot/dts/imx6qdl-sabresd.dtsi?h=v5.7.6#n513 Also there is no override of the fec PTP clock: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/arch/arm/boot/dts/imx6qdl-sabresd.dtsi?h=v5.7.6#n202 Although Shawn's mainline patch mandates this? https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=v5.7.6&id=810c0ca879098a993e2ce0a190d24d11c17df748 This will work, but only by accident. So on a plus, when we (incorrectly) switch the bypass bit on, things stop working.
