[BUG]odroid-c2 does not hotplug usb-devices

[Martin Blumenstingl]

Hi Anand,

On Sun, Dec 20, 2020 at 2:46 PM Anand Moon <linux.amoon at gmail.com> wrote:
[...]
> > On Sat, Dec 19, 2020 at 8:53 PM Anand Moon <linux.amoon at gmail.com> wrote:
> > [...]
> > > I was also looking into this issue so I made some changes in the
> > > phy driver to resolve the issue. Plz share your thoughts on the changes below.
> > first I have some questions :-)
> > 1. do you see the same problem that Otto sees? this means: a) USB
> > hotplug works as long as at least one device is plugged in at boot b)
> > (if I understand Otto correctly then) it breaks once all USB devices
> > have been removed
>
> On C1/C2 issue is when single usb hdd is connected to board it will
> not get detected.
> unless you connect another usb keyboard or wireless dongle to the board.
>
> *USB hot plug only works if two ore more usb devices are connected.*
I am not sure if that's really the same issue as described by Otto

[...]
> > > amoon at ThinkPad-T440s:~/mainline/linux-aml-5.y-devel$ git diff
> > > diff --git a/arch/arm64/boot/dts/amlogic/meson-gxbb.dtsi
> > > b/arch/arm64/boot/dts/amlogic/meson-gxbb.dtsi
> > > index 7c029f552a23..363dd2ac17e6 100644
> > > --- a/arch/arm64/boot/dts/amlogic/meson-gxbb.dtsi
> > > +++ b/arch/arm64/boot/dts/amlogic/meson-gxbb.dtsi
> > > @@ -20,6 +20,7 @@ usb0_phy: phy at c0000000 {
> > >                         #phy-cells = <0>;
> > >                         reg = <0x0 0xc0000000 0x0 0x20>;
> > >                         resets = <&reset RESET_USB_OTG>;
> > > +                       reset-names = "phy-reset";
> > >                         clocks = <&clkc CLKID_USB>, <&clkc CLKID_USB0>;
> > >                         clock-names = "usb_general", "usb";
> > >                         status = "disabled";
> > > @@ -30,6 +31,7 @@ usb1_phy: phy at c0000020 {
> > >                         #phy-cells = <0>;
> > >                         reg = <0x0 0xc0000020 0x0 0x20>;
> > >                         resets = <&reset RESET_USB_OTG>;
> > > +                       reset-names = "phy-reset";
> > >                         clocks = <&clkc CLKID_USB>, <&clkc CLKID_USB1>;
> > >                         clock-names = "usb_general", "usb";
> > >                         status = "disabled";
> > I don't see why the above two changes are needed
> > see my comment about of_reset_control_get_shared below
> >
>
> Yep I borrowed this logic from rockchip usb phy controller.
> but I missed the action on reset.
>
> err = devm_add_action_or_reset(base->dev, rockchip_usb_phy_action,  rk_phy);
> if (err)
> return err;
(assuming that your comment relates to the of_reset_control_get_shared
call below)
I checked rockchip_usb_phy_action and it's not calling reset_control_put either
this looks strange to me and I would say that there's a memory leak.

[...]
> > >  static const struct phy_ops phy_meson8b_usb2_ops = {
> > >         .power_on       = phy_meson8b_usb2_power_on,
> > >         .power_off      = phy_meson8b_usb2_power_off,
> > > +       .reset          = phy_meson8b_usb2_reset,
> > >         .owner          = THIS_MODULE,
> > >  };
> > I tested this on my Odroid-C1: phy_meson8b_usb2_reset is never called
> > after checking the dwc2 code this is expected: only in one very
> > specific case the dwc2 driver calls phy_reset
> > can you please find out how phy_meson8b_usb2_reset is called in your kernel?
> >
>
> Yep, Its not getting called on my board, l might be missing some thing.
are you sure that your patch fixes USB hotplug for you?

the new reset callback is a no-op because it's not called
the code below (of_reset_control_get_shared) only fetches the same
reset line that we are already using
so I am not sure what changed behavior you are seeing - can you please
explain this in more detail?

> > > @@ -271,6 +277,10 @@ static int phy_meson8b_usb2_probe(struct
> > > platform_device *pdev)
> > >                 return -EINVAL;
> > >         }
> > >
> > > +       priv->reset = of_reset_control_get_shared(pdev->dev.of_node,
> > > "phy-reset");
> > this causes a memory-leak upon driver removal
> > also a few lines above we are already getting the reset line, so why
> > is this needed?
>
> I had a feeling that USB_OTG reset will be shared between both the phy
> controller.
indeed, the USB_OTG reset is shared

> Odroid C2 reset controller have additional RESET_USB_DDR_[0,3] reset,
> but Odroid C1 dose not have this feature.
have you found where the patched Amlogic kernel/u-boot use these reset lines?
if they need to be managed by us then we probably have a bug, because
we're not currently using these anywhere

> Next time I will try to do more testing on both the devices C1/C2.
> before sending some code for review or testing.
I am happy about everyone who helps fixing the USB hotplug issue!
it's a tricky issue where I don't even know if the USB PHY, dwc2
controller or something else is the culprit
so any help is appreciated :)


Best regards,
Martin