On 11/09/2020 18:22, Marek Behún wrote:
On Fri, 11 Sep 2020 17:52:26 +0200
Andre Heider <a.hei...@gmail.com> wrote:
Hi Marek,
On 11/09/2020 13:55, Marek Behún wrote:
On Wed, 9 Sep 2020 00:38:31 +0200 Pali Rohár <p...@kernel.org>
wrote:
On Tuesday 08 September 2020 08:52:56 Tom Rini wrote:
Note that when CONFIG_NET_RANDOM_ETHADDR is set, we only use a
random MAC address when we haven't found one either on the
hardware or environment.
I know.
It also prints a warning that you are using a random MAC,
so if it's documented on how to recover the real MAC a user should
see that warning and fix it.
In case you did backup of MAC address or you have MAC address
printed on sticker, you can restore it. If you loaded distribution
U-Boot which erase MAC address and you have not did any backup,
then your MAC address is forever lost.
On Turris MOX we write the MAC address into OTP of the SOC during
manufacturing.
It is possible to write code that burns the MAC address into OTP, I
consider this a better option than enabling random MAC address.
Maybe we can enable random MAC address, and if MAC address is not
found in environment nor OTP, issue a warning, something like
"WARNING: MAC address lost, please burn the MAC address of your
device to OTP with command xyz"
What do you think?
if there's a mac stored in otp during manufacturing, that's of course
the best solution. There's no need for CONFIG_NET_RANDOM_ETHADDR
then. But globalscale does not do that.
Doing it afterwards, so u-boot claiming some otp space for itself,
and instructing the user how to write to it sounds too
dangerous/error-prone.
For me CONFIG_NET_RANDOM_ETHADDR is a knob that should be enabled if
there's no mac address stored in a sane way (where saving it just to
an u-boot env during manufacturing doesn't count as sane, especially
if the vendor moves the spi env offset around in a firmware update).
So I think CONFIG_NET_RANDOM_ETHADDR is enough.
I understand Pali's concerns, though.
The thing is that if we enable CONFIG_NET_RANDOM_ETHADDR by default,
many users that have managed to wipe their env won't care about that
they are using randomly generated MAC address and won't solve the issue,
which is again the spirit of correctly configure networks.
If we do not enable CONFIG_NET_RANDOM_ETHADDR, the worst that can
happen is that the device won't boot from network. This will force the
users to solve the issue, which is not that hard
setenv ethaddr aa:bb:cc:dd:ee:ff (address from the sticker)
saveenv
If the users boots from SD/eMMC/SATA/USB, Linux won't have problem with
network, since it will generate random MAC address anyway.
Good point, so let's assume the user doesn't have a mac stored.
Without CONFIG_NET_RANDOM_ETHADDR:
* u-boot refuses to do anything network related
* Linux generates a random mac, network works
With CONFIG_NET_RANDOM_ETHADDR:
* u-boot generates a random mac, network works
* Linux uses the same random mac, passed on from u-boot throught the
device-tree, network works
It's still random, but at least it's consistent ;)
The worst case scenario does not look that bad to me if we don't enable
CONFIG_NET_RANDOM_ETHADDR by default.
I think the option CONFIG_NET_RANDOM_ETHADDR should be used only by
developers when they are developing on devices that have not yet burned
the MAC addresses, or on some embedded devices that don't have space
for saving a MAC address (no storage for env or anything else, do such
devices exist?).
It looks like it's used for more than that though:
$ git ls-files configs|wc -l
1362
$ git grep CONFIG_NET_RANDOM_ETHADDR configs|wc -l
209
Some popular sbc's are part of that list.
But it shouldn't be used as a default setting for a device that has
storage where it can store the address.
But it would be nice if e.g. a board could set specific env vars as
indestructible/unwipeable/precious/whatever, which instructs `env
default -a` to keep those (which is common after updating the
bootloader). Maybe that's an idea worth pursuing?
Yes, that would be nice :)
Thanks,
Andre
Marek
