Re: [PATCH v6 6/8] clk: sunxi-ng: a733: Add mod clocks support

[Vinicius Pedrosa <vinicius.eduardo.pedrosa@gmail.com>]

Hi Andre,

On 11/2/26 16:32, Andre Przywara wrote:
[...]
> > +static const struct clk_parent_data usb01_ref_parents[] = {
> > +	{ .hw = &sys_24M_clk.hw },
> > +	{ .fw_name = "hosc" },
> > +};
> > +static SUNXI_CCU_MUX_DATA(usb01_ref_clk, "usb01-ref", usb01_ref_parents, 0x1340, 24, 3, 0);
> > +
> > +static SUNXI_CCU_MUX_DATA(usb2_u2_ref_clk, "usb2-u2-ref", usb01_ref_parents, 0x1348, 24, 3, 0);
>
> Don't these two clocks miss the gate at bit 31? So it should be
> SUNXI_CCU_MUX_DATA_WITH_GATE instead?

Yes, both have the gate at bit 31 in the manual, and the vendor CCU
models 0x1340 with it too. usb2-u2-ref resets to off, which is the one
I hit on my board. usb01-ref resets to on, so nothing breaks today.

With the gate added, clk_disable_unused would switch usb01-ref off
unless something claims it, and in my local DT nothing does (the
USB0/USB1 PHYs take osc24M like the A523 ones). On my Cubie A7Z,
clearing bit 31 of 0x1340 at runtime stops the USB1 Wi-Fi traffic, and
after rebinding EHCI1/OHCI1 the device no longer enumerates (error -62),
4 out of 4, while it enumerates again with the bit set. So at least USB1
will need CLK_USB01_REF kept enabled once the gate is added, likely
through a consumer in the USB controller or PHY, as suggested by the
vendor DT. I only tested USB1, by writing the register, not the gated
CCU itself.

Thanks also for your guidance throughout these discussions.

Thanks,
Vinicius