On Wed, Nov 17, 2021 at 05:45:58PM +0100, Michael Walle wrote:
Am 2021-11-16 22:14, schrieb Tom Rini:
On Mon, Nov 15, 2021 at 11:45:51PM +0100, Michael Walle wrote:
Nowadays, u-boot (when CONFIG_NET_RANDOM_ETHADDR is set) will set enetaddr to a random value if not set and then pass the randomly generated MAC address to linux.
First, for clarity I'm not nak'ing this. I kind of would like to see a slight reword as I think some things aren't 100% correct, even if the "save random MAC to ethaddr environment variable" change goes in. For example, it's quite long standing that (dev|pdata)->enetaddr populates "mac-address" and "local-mac-address" and it seems in some older cases we only set the "local-mac-address" property.
fdt_fixup_memory() in common/fdt_support.c does a env_get(mac).
Whoops, yes, I misrecalled this.
I'm not even sure, if there is a connection between what is fixed up in the kernel DT and the corresponding device in u-boot if CONFIG_DM_SEQ_ALIAS is not set.
And then yes, the logic here can be further fragile at times with multiple interfaces.
This is bad for the following reasons: (1) it makes it impossible for linux to detect this error (2) linux won't trigger any fallback mechanism for the case where it didn't find any valid MAC address
This feels in some ways to be a limitation of the binding: https://www.kernel.org/doc/Documentation/devicetree/bindings/net/ethernet-co...
Agreed. But it doesn't render my argument invalid ;)
"Bug" or "feature", but yes, OK.
And it reads like we really must be populating "mac-address" with that random one and while providing a blank "local-mac-address" would be a way to say we don't know the true device one, it seems that wouldn't be used / noticed?
I guess it will just use the one populated in mac-address, i.e. the random one. Btw, the binding says "last used" what if u-boot never actually used that ethernet controller, should it then be empty or populated with the random mac address.
My gut feeling here is that we have an example of a binding that has been formalized around what exists in the wild, and was not strictly intentional and clear at all times. What we do today in the case of ethernet devices we haven't used is populate both, just like the one we have used, based on the environment. And popping over to current Linux kernel git, the handful of boards that hard-code a non-zero mac-address or local-mac-address in the dts just makes things even less-clear on what to do. Which probably leaves us with "populate mac-address with whatever you can".
(3) a saveenv will store this randomly generated MAC address in the environment
Probably, the user will also be unaware that something is wrong. He will just get different MAC addresses on each reboot, asking himself why this is the case.
As this board usually have a serial port, the user can just fix this by setting the MAC address manually in the environment. Also disable the netconsole just in case, because it cannot be guaranteed that it will work in any case. After all, this was just a convenience option, because the bootloader - right now - doesn't have the ability to read the MAC address, which is stored in the OTP. But it is far more important to have a clear view of whats wrong with a board and that means we can no longer use this Kconfig option.
Signed-off-by: Michael Walle michael@walle.cc
configs/kontron_sl28_defconfig | 2 -- 1 file changed, 2 deletions(-)
diff --git a/configs/kontron_sl28_defconfig b/configs/kontron_sl28_defconfig index af907175f1..7fb6bdbe82 100644 --- a/configs/kontron_sl28_defconfig +++ b/configs/kontron_sl28_defconfig @@ -59,8 +59,6 @@ CONFIG_OF_LIST="" CONFIG_ENV_OVERWRITE=y CONFIG_ENV_IS_IN_SPI_FLASH=y CONFIG_SYS_REDUNDAND_ENVIRONMENT=y -CONFIG_NET_RANDOM_ETHADDR=y -CONFIG_NETCONSOLE=y CONFIG_SPL_DM_SEQ_ALIAS=y CONFIG_SCSI_AHCI=y CONFIG_SATA_CEVA=y
Now, an alternate solution here would be to enable these two options still and write an ft_board_setup() with: ... if ethaddr is in the locally administered pool then fdt_delprop(... "mac-address"); fdt_delprop(... "local-mac-address");
Which would also trigger if a user sets a "locally administered" mac himself. Even if unlikely, I'm opposed to such magic and unexpected behavior. Sooner or later it will bite you.
A fair point, yes.
And that should cause the kernel to fall through the cases to find out where to get the real MAC from. I'm not super happy with this at first, but I also don't see anything clever in the binding we can do.
If I'll need this one again, then I'll just add the random mac generation in board_eth_init(), I think.
OK.