Hi Krzysztof,
On Tue, Nov 22, 2022 at 11:23 AM Krzysztof Kozlowski
<krzysztof.kozlowski@xxxxxxxxxx> wrote:
> On 22/11/2022 10:01, Geert Uytterhoeven wrote:
> >>>>> The h2mode bit (and probably a few other controls we haven't figured out
> >>>>> yet) in the sysctrl must be set before any of the USB devices is active.
> >>>>> Hence it's safest for the sysctrl to do this before any of the USB drivers
> >>>>> probes.
> >>>>
> >>>> Again, this does not differ from many, many of other devices. All of
> >>>> them must set something in system controller block, before they start
> >>>> operating (or at specific time). It's exactly the same everywhere.
> >>>
> >>> The issue here is that there are two _different drivers_ (USB host
> >>> and device). When both are modular, and the driver that depends on the
> >>> sysctrl setting is loaded second, you have a problem: the sysctrl change
> >>> must not be done when the first driver is already using the hardware.
> >>>
> >>> Hence the sysctrl driver should take care of it itself during early
> >>> initialization (it's the main clock controller, so it's a dependency
> >>> for all other I/O device drivers).
> >>
> >> I assumed you have there bit for the first device (which can switch
> >> between USB host and USB device) to choose appropriate mode. The
> >> bindings also expressed this - "the USBs are". Never said anything about
> >> dependency between these USBs.
> >>
> >> Are you saying that the mode for first device cannot be changed once the
> >> second device (which is only host) is started? IOW, the mode setup must
> >> happen before any of these devices are started?
> >
> > Exactly.
> >
> >> Anyway with sysctrl approach you will have dependency and you cannot
> >> rely on clock provider-consumer relationship to order that dependency.
> >> What if you make all clocks on and do not take any clocks in USB device?
> >
> > Enabling the clocks does not have anything to do with this ordering.
>
> That was the argument from Herve, that ordering is guaranteed by clocks.
>
> > Clock consumers that are part of the clock domain are probed after
> > clock providers. If the clock is missing, that would be an incorrect
> > description in DTS.
>
> If not clocks, what else is guaranteeing the ordering? You did not
> express it in DT.
clocks and power-domains
And if not clocks and power-domains... Oops, we didn't express in DT
that the SoC needs to be powered at all ;-)
Can we please stop this pointless discussion?
Thanks!
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
