Comment # 4 on bug 962694 from Fabian Vogt

(In reply to Takashi Iwai from comment #3)
> The part marked is the problem.  In this case, mpt3sas module does exist in
> 4.3 kernel but it isn't loaded because 4.3 driver doesn't support yet
> mpt2sas devices.  dracut still believes it's safe to ignore.  Oops.
> 
> That said, dracut can't handle the case where a module is folded into
> another.

Jup, the module handling (the "hostonly" part) logic is totally broken and
flawed for kernel migrations.
https://github.com/haraldh/dracut/commit/07a081f352497258862ae164d11d9e6dc2c5b959
solved some of the issues, but most still remain...
The source of this issue is that

>     # check if module is loaded
>    [[ ${host_modules["$_modenc"]} ]] && return 0

does not trigger for mpt3sas, as host_modules does not contain aliases.
I'd say that it should be replaced by

>    # check if module could be loaded.
>    # If not, it's either new or already loaded -> install it!
>    [[ $(/usr/sbin/modprobe -nv "$_mod" 2>/dev/null | wc -l) -eq 0 ]] && return 0

Adding aliases to $host_modules would probably work as well, but that would
require parsing the output of modinfo for each module. This way is IMO more
obvious and easier.

Ronan: Can you test the above? Simply replace the lines in
/usr/lib/dracut/dracut-init.sh:1029 and run mkinitrd from the 4.3.3 kernel.