Peng_Fan: hi, Dominique here. I agree making all soc_device_match callers handle defer might be difficult, but I don't see much other way about it

We can't force another driver to load or ensure nvme is init'd before soc

Asmadeus: agree, but I am a bit concerned about soc_bus, it is a virtual bus, should we assume that only one soc_device registered during system running?

(That might actually be a problem, what if a nvme driver requires soc revision?)

Ah. but actually all i.MX has UID in ocotp

If we need get info from ocotp to fill soc device, we need nvmem ready.

And the reason to convert it to module driver is to support kexec and android GKI.

Yeah, I had stumbled on kexec failing as well; this was nice to see a fix. Unfortunately it means I can't boot with caam anymore so it will be difficult to sell :)

understand. We have raised the issue inside i.MX kernel team, probobaly we will have a patch out to community next monday.

Thanks

As for soc_bus I don't know, gut feeling is that there will be other things failing if there is a mismatch (different socs handled by the same linux providing devices handled by the same driver for example), so I feel it would be safe to assume homogenity at least... but honestly no idea