Re: [PATCH V2] usb: ehci-mx6: Enable OTG detection on imx8mm and imx8mn

[Michael Walle <michael@walle.cc>]

Am 2022-01-11 17:52, schrieb Adam Ford:
> On Tue, Jan 11, 2022 at 9:31 AM Michael Walle <michael@walle.cc> wrote:
>> 
>> Hi,
>> 
>> Am 2022-01-11 15:20, schrieb Adam Ford:
>> > On Tue, Jan 4, 2022 at 2:32 AM Michael Walle <michael@walle.cc> wrote:
>> >> > The imx8mm and imx8mn appear compatible with imx7d-usb
>> >> > flags in the OTG driver.  If the dr_mode is defined as
>> >> > host or peripheral, the device appears to operate correctly,
>> >> > however the auto host/peripheral detection results in an error.
>> >> >
>> >> > The solution isn't just adding checks for imx8mm and imx8mn to
>> >> > the check for imx7, because the USB clock needs to be running
>> >> > to read from the USBNC_PHY_STATUS_OFFSET register or it will hang.
>> >> >
>> >> > The init_type in both priv and plat data are the same, so it doesn't
>> >> > make sense to configure the data in the plat data and copy the
>> >> > data to priv when priv can be configured directly.  Instead, rename
>> >> > ehci_usb_of_to_plat to ehci_usb_dr_mode and call it from the
>> >> > probe functions after the clocks are enabled, but before the data
>> >> > is required.
>> >> >
>> >> > With that added, the additional checks for imx8mm and imx8mn will
>> >> > allow reading the register to automatically determine the state
>> >> > (host or device) of the OTG controller.
>> >> >
>> >> > Signed-off-by: Adam Ford <aford173@gmail.com>
>> >> > ---
>> >> > V2:  Rename ehci_usb_of_to_plat to ehci_usb_dr_mode and call it
>> >> >      from the probe after the clocks are enabled, but before
>> >> >      the data is needed.
>> >> >
>> >> > diff --git a/drivers/usb/host/ehci-mx6.c b/drivers/usb/host/ehci-mx6.c
>> >> > index 1bd6147c76..f2a34b7f06 100644
>> >> > --- a/drivers/usb/host/ehci-mx6.c
>> >> > +++ b/drivers/usb/host/ehci-mx6.c
>> >>
>> >> ..
>> >>
>> >> > @@ -639,10 +639,8 @@ static int mx6_parse_dt_addrs(struct udevice *dev)
>> >> >
>> >> >  static int ehci_usb_probe(struct udevice *dev)
>> >> >  {
>> >> > -     struct usb_plat *plat = dev_get_plat(dev);
>> >> >       struct usb_ehci *ehci = dev_read_addr_ptr(dev);
>> >> >       struct ehci_mx6_priv_data *priv = dev_get_priv(dev);
>> >> > -     enum usb_init_type type = plat->init_type;
>> >> >       struct ehci_hccr *hccr;
>> >> >       struct ehci_hcor *hcor;
>> >> >       int ret;
>> >> > @@ -660,8 +658,6 @@ static int ehci_usb_probe(struct udevice *dev)
>> >> >               return ret;
>> >> >
>> >> >       priv->ehci = ehci;
>> >> > -     priv->init_type = type;
>> >>
>> >
>> > Michael,
>> >
>> >> I'm not sure this is correct. There is also this:
>> >> https://elixir.bootlin.com/u-boot/v2022.01-rc4/source/drivers/usb/host/usb-uclass.c#L407
>> >>
>> >
>> > Further down in the code, you should see:
>> >     priv->phy_type = usb_get_phy_mode(dev_ofnode(dev));
>> 
>> But that is just fetching the mode from the device tree, the
>> usb_setup_ehci_gadget() referenced above, change the mode during
>> runtime by writing the plat->init_type and reprobing the device.
>> 
>> >> Which won't work anymore. usb_setup_ehci_gadget() is called from
>> >> usb_gadget_register_driver() in ci_udc.c. The latter is the one used
>> >> on the imx, right? But I might be wrong. I'm still trying to figure
>> >> out how this all works together, because I also try to get OTG
>> >> running on the dwc3 driver. It looks like the ci_udc.c is special
>> >> here, and I wonder how a transition to UCLASS_USB_GADGET_GENERIC
>> >> might look like for this driver.
>> >
>> > This driver really isn't changing anything, it's just an order of
>> > operations.
>> 
>> It doesn't use the plat->init_type at all anymore, no?
> 
> From what I could tell, the only thing that plat->init_type did was to
> set priv->init_type.
> The priv structure appears to do most of the work.

but plat->init_type seems to change during runtime (by
usb_setup_ehci_gadget()) and with this patch applied, it doesn't
do that anymore.

>> >  Previously there was a separate that was being called to
>> > determine the init_type as being either a peripheral or host.  If it
>> > wasn't explicitly set in the device tree, the helper function would
>> > query a register, however, on the imx8mm and imx8mn platforms, the
>> > clocks were not running which resulted in a hang.  What I've done is
>> > simply move the helper function into the probe and have it run after
>> > the clocks start.  In theory the register values will be the same as
>> > they would be with the help function still in place.
>> >
>> > Both Tim Harvey and I have tested on an imx8mm and Fabio tested it on
>> > an imx7.  For me, the peripheral mode worked when the ID pin of the
>> > USB-OTG was not grounded.  When it was grounded, the system corrected
>> > identified it as a host and I was able to mount a USB drive.
>> 
>> Again, I'm missing something here, but the only user of
>> usb_setup_ehci_gadget() is usb/gadget/ci_udc.c.
> 
> I can't speak to what those functions do.   What are you suggesting
> that I do differently?

I'd start by seeing if/when usb_setup_ehci_gadget() is called and
see if plat->init_type changes. If its not called, is it dead code
then?
Sorry I just noticed this, while reviewing how this particular driver
works, because I still like to find a working OTG driver in u-boot.
I don't have any imx boards.

-michael