Re: Orange Pi 5 weirdness

[email protected] (Harold Gutch) writes:

>On Sun, Aug 25, 2024 at 10:33:47AM -0400, Mouse wrote:
>> > eqos0: autoconfiguration error: Wrong MAC address. Clear the multicast bit.
>> > eqos0: Ethernet address c4:ab:14:c6:33:37
>> 
>> Well, there's a bug somewhere.  Either the MAC it's printing isn't the
>> MAC it's testing or its test for multicastness is broken, because that
>> MAC does not have its multicast bit set (c4 has its 01 bit clear).

>... or option 3, the output is wrong, what it actually means is
>"Clearing the multicast bit",
>https://nxr.netbsd.org/xref/src/sys/dev/ic/dwc_eqos.c#1266

>But unless eqos is doing something special I doubt that this is the
>expected behavior.  c4:ab:14 (or c5:ab:14) doesn't seem to be a valid
>MAC OUI.


It is bits that EFI reads from the OTP area of the SOC. Linux (without EFI)
seems to use different bits.

Here is the same SoC (RK3588S) in a different SBC (and the OTP bits are
combined differently):

[     1.000004] eqos0 at acpi0 (MAC1, RKCP6543-61): mem 0xfe1c0000-0xfe1cffff irq 266,265
[     1.000004] eqos0: DesignWare EQOS ver 0x51 (0x30)
[     1.000004] eqos0: hw features 1a1173f7 111e01e8 11041041 0c370031
[     1.000004] eqos0: using 32-bit DMA
[     1.000004] eqos0: Ethernet address d6:de:d3:53:de:db
[     1.000004] rgephy0 at eqos0 phy 0: RTL8211F 1000BASE-T media interface
[     1.000004] rgephy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 1000baseT-FDX, auto

The other interfaces (rge) on that SBC don't come with a MAC address at all
and I set them with ifconfig.