Comment # 16 on bug 1184924 from Ivan Ivanov

(In reply to Fabian Vogt from comment #15)
> > [    0.000000] Call trace:
> > [    0.000000]  kthread_should_stop+0x2c/0x60
> > [    0.000000]  add_bootloader_randomness+0x2c/0x38
> > [    0.000000]  efi_config_parse_tables+0x120/0x250
> > [    0.000000]  efi_init+0x138/0x1e0
> > [    0.000000]  setup_arch+0x394/0x778
> > [    0.000000]  start_kernel+0x90/0x568
> 
> This is called only in the case that crng_init != 0.
> 
> This is indeed the case, because earlier it prints:
> > [    0.000000] random: fast init done
> 
> My theory is that in the kexec case, there is additional
> entropy available through the rng-seed property in the device
> tree passed by the previous kernel, so the pool is already
> initialized when grabbing entropy through EFI.
> 
> This issue does not appear on Tumbleweed (FWICT), question is
> which commit(s) fixed it.

If not mistaken we don't have CONFIG_RANDOM_TRUST_BOOTLOADER enabled
in Tumbleweed and that is way we don't see the issue there.

Otherwise, AFAICT, as you righty analysed, issue is that 
add_hwgenerator_randomness() is supposed to be called from
dedicated kernel thread and not from main init code.