Hi,

On Tue, May 25, 2021 at 12:29:01PM +0100, Andre Przywara wrote:
> On Mon, 24 May 2021 13:59:46 +0200
> Maxime Ripard <maxime@cerno.tech> wrote:
> 
> Hi Maxime,
> 
> > On Wed, May 19, 2021 at 11:41:47AM +0100, Andre Przywara wrote:
> > > At least the Allwinner H616 SoC requires a weird quirk to make most
> > > USB PHYs work: Only port2 works out of the box, but all other ports
> > > need some help from this port2 to work correctly: The CLK_BUS_PHY2 and
> > > RST_USB_PHY2 clock and reset need to be enabled, and the SIDDQ bit in
> > > the PMU PHY control register needs to be cleared. For this register to
> > > be accessible, CLK_BUS_ECHI2 needs to be ungated. Don't ask ....
> > > 
> > > Instead of disguising this as some generic feature, do exactly that
> > > in our PHY init:
> > > If the quirk bit is set, and we initialise a PHY other than PHY2, ungate
> > > this one special clock, and clear the SIDDQ bit. We can pull in the
> > > other required clocks via the DT.
> > > 
> > > Signed-off-by: Andre Przywara <andre.przywara@arm.com>  
> > 
> > What is this SIDDQ bit doing exactly?
> 
> I probably know as much as you do, but as Jernej pointed out, in some
> Rockchip code it's indeed documented as some analogue PHY supply switch:
> ($ git grep -i siddq drivers/phy/rockchip)
> 
> In fact we had this pin/bit for ages, it was just hidden as BIT(1) in
> our infamous PMU_UNK1 register. Patch 10/17 drags that into the light.

Ok

> > I guess we could also expose this using a power-domain if it's relevant?
> 
> Mmmh, interesting idea. So are you thinking about registering a genpd
> provider in sun4i_usb_phy_probe(), then having a power-domains property
> in the ehci/ohci nodes, pointing to the PHY node? And if yes, should
> the provider be a subnode of the USB PHY node, with a separate
> compatible? That sounds a bit more involved, but would have the
> advantage of allowing us to specify the resets and clocks from PHY2
> there, and would look a bit cleaner than hacking them into the
> other EHCI/OHCI nodes.

I'm not sure we need a separate device node, we could just register the
phy driver as a genpd provider, and then with an arg (so with
of_genpd_add_provider_onecell?) the index of the USB controller we want
to power up.

> I would not touch the existing SoCs (even though it seems to apply to
> them as well, just not in the exact same way), but I can give it a
> try for the H616. It seems like the other SIDDQ bits (in the other
> PHYs) are still needed for operation, but the PD provide could actually
> take care of this as well.
> 
> Does that make sense or is this a bit over the top for just clearing an
> extra bit?

Using what I described above should be fairly simple, so if we can fit
in an available and relevant abstraction, yeah, I guess :)

Maxime