imx8mm/imx8mn hang on usb stop / ehci_shutdown

[Adam Ford]

On Thu, Apr 27, 2023 at 12:14 PM Tim Harvey <tharvey at gateworks.com> wrote:
>
> On Wed, Mar 15, 2023 at 3:01 PM Tim Harvey <tharvey at gateworks.com> wrote:
> >
> > Greetings,
> >
> > I'm seeing a hang on imx8mm-venice and imx8mn-venice boards which have
> > a USB host controller in host mode when ehci_shutdown() is called
> > (which is called for 'usb stop' as well as when booting a kernel once
> > 'usb start' has been issued).
> >
> > This appears to be caused by the ehci_shutdown() function,
> > specifically the write to the or_portsc register to set EHCI_PS_SUSP:
> >                 for (i = 0; i < max_ports; i++) {
> >                         reg = ehci_readl(&ctrl->hcor->or_portsc[i]);
> >                         reg |= EHCI_PS_SUSP;
> >                         ehci_writel(&ctrl->hcor->or_portsc[i], reg);
> >                 }
> >
> > Does anyone else out there with an imx8mm/imx8mn board with usbotg1 or
> > usbotg2 configured with dr_mode="host" see this as well?
> >
> > I'm not clear where the equivalent code would be in the Linux kernel
> > to compare with. I do know that commenting out the ehcI_write above
> > keeps ehci_shutdown from hanging but does cause the handshake to fail
> > on the next cmd that disables CMD_RUN and thus prints "EHCI failed to
> > shut down host controller.".
> >
> > Any ideas?
> >
> > Best Regards,
> >
> > Tim
>
> I haven't heard any response here so I did some more digging. This
> hang should be affecting all imx8mm/imx8mn devices that have USB host
> controllers configured for host mode (dr_mode = "host") and would
> appear for anyone that ever does a 'usb stop' (or does a 'usb start'
> followed by a kernel boot for example of loading kernel from USB) in
> U-Boot.
>
> The issue is caused by the pgc_otg{1,2} and/or pgc_hsiomix power
> domains getting disabled for the 'usb_hub' device (which is present
> for the USB host even if you have no hub) prior to ehci_shutdown being
> called.
>
> The following patch resolves this but I'm thinking there should be a
> better way to control this from ehci-mx6.c?

There was some re-organization of the power domains in Linux in the
device tree [1].  Do you know if those propagated to U-Boot?

adam

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/arch/arm64/boot/dts/freescale/imx8mm.dtsi?h=v6.3&id=4585c79ff477f9517b7f384a4fce351417e8fa36

>
> diff --git a/common/usb_hub.c b/common/usb_hub.c
> index 85c0822d8b7b..03237deaa0be 100644
> --- a/common/usb_hub.c
> +++ b/common/usb_hub.c
> @@ -948,7 +948,7 @@ U_BOOT_DRIVER(usb_generic_hub) = {
>         .name   = "usb_hub",
>         .id     = UCLASS_USB_HUB,
>         .of_match = usb_hub_ids,
> -       .flags  = DM_FLAG_ALLOC_PRIV_DMA,
> +       .flags  = DM_FLAG_ALLOC_PRIV_DMA | DM_FLAG_LEAVE_PD_ON,
>  };
>
>
> Best Regards,
>
> Tim