Add devicetree `serial-number` as fallback for `product_serial` from DMI tables

[septatrix]

(I am not sure if this better belongs here or node-exporter, feel free to move the issue to the other repo.)

On most ARM devices there are no DMI tables and when using things like u-boot emulating UEFI they are still incomplete. Thus /sys/class/dmi/id/product_serial is not available/empty. Alternatively, however, /sys/firmware/devicetree/base/serial-number is usually available on those devices and also contains a serial number. In case the former does not exists it should fall back to the latter.

[discordianfish]

I don't know about this nor what 'devicetree' is exactly, but instead of falling back we should probably explicilty expose devicetree entries, so the caller of procfs can decide how to reach e.g serial numbers

[septatrix]

They are a mechanism for Linux to obtain data about the hardware and are widely used on e.g. ARM in the absence of UEFI.

The exposed serial numbers semantically represent the same thing as those in the DMI tables for UEFI machines. As such other projects like systemd also treat them equivalent

I would also be fine with exposing them separately in this lower level module. In that case however I would also create an issue at the node-exporter repo to treat them equivalently there

[discordianfish]

I would also be fine with exposing them separately in this lower level module. In that case however I would also create an issue at the node-exporter repo to treat them equivalently there

That sounds reasonable

[septatrix]

The docs say that one shall follow the symlink under /proc/device-tree but the result will live under /sys. How would this reflect in the module? It seems like all components get filesystem paths passed in. Should it receive only procfs or both?

https://github.com/torvalds/linux/blob/bc8198dc7ebc492ec3e9fa1617dcdfbe98e73b17/Documentation/ABI/testing/sysfs-firmware-ofw#L14-L17

[discordianfish]

I think in this case follow the symlink in /proc/device-tree