Mark Kettenis wrote: > > As I understand the driver, cpsw0 will always have a zero address if > > the "ti,cpsw" device tree node has either no child nodes at all or > > none named "local-mac-address". > > > > If a "local-mac-address" child node exists then that address is used. > > Small correction, you mean a "local-mac-address" property instead of a > node.
Ah yes, thanks! > > (if_cpsw.c:364 cpsw_attach() calling cpsw_get_port_config() ff.) > > > > Neither the snapshot dtb nor Patrick's dtb contain "local-mac-address" so > > is U-Boot modifying only the older dtb (why!?) or what is going on here? > > Yes U-Boot will update the FDT with "local-mac-address" properties. > U-Boot typically uses "ethernetN" properties under /aliases to decide > where to place the "local-mac-address" properties. So if those > aliases aren't there you end up without "local-mac-address" > properties. Aha! Patrick Wildt wrote: > > > Working on the cpsw driver I was silently wondering where the nonzero > > > MAC on my beaglebone black comes from and I still don't know the answer. > > > > Either local-mac-address or from hardware registers. For cpsw(4) it's > > local-mac-address. U-Boot configures those on bootup. May cpsw(4) access the control module registers? If yes I'd rather make the driver read the fused address(es) instead of local-mac-address, to completely avoid any problems. At the very least do so when local-mac-address is zero or not found. It seems unneccessary to depend on another software (U-Boot) when the information is available in the hardware. > > Usually there needs to be a /aliases/ethernet0 or so pointing to > > the correct node for U-Boot to set it correctly. It's possible > > that this changes as well. I will have a look. > > And that is exactly what happened, this should be reverted as well: > > - ethernet0 = &cpsw_emac0; > - ethernet1 = &cpsw_emac1; > + ethernet0 = &cpsw_port1; > + ethernet1 = &cpsw_port2; Good find! > Can you give this a shot? Sure: U-Boot 2022.10 + new dtb-6.0.8 boots both 7.2 and Nov 1 snapshot kernels on BeagleBone Black with nonzero cpsw0 address and ping -f works fine. //Peter
