[PATCH] usb: dwc3: gadget: Fix match_ep callback for NXP UUU tool

[Francesco Dolcini]

On Sun, Apr 13, 2025 at 11:14:07AM +0200, Marek Vasut wrote:
> On 4/11/25 4:21 PM, Francesco Dolcini wrote:
> > On Tue, Apr 08, 2025 at 04:51:22PM +0200, Marek Vasut wrote:
> > > On 4/8/25 11:06 AM, Francesco Dolcini wrote:
> > > > On Mon, Mar 24, 2025 at 03:36:52PM +0100, Marek Vasut wrote:
> > > > > On 3/24/25 3:16 PM, Francesco Dolcini wrote:
> > > > > > On Mon, Mar 24, 2025 at 02:53:23PM +0100, Marek Vasut wrote:
> > > > > > > On 3/24/25 1:30 PM, Francesco Dolcini wrote:
> > > > > > > > On Mon, Mar 24, 2025 at 09:26:03AM +0100, Mattijs Korpershoek wrote:
> > > > > > > > > Hi Francesco,
> > > > > > > > > 
> > > > > > > > > On lun., mars 24, 2025 at 09:03, Francesco Dolcini <francesco at dolcini.it> wrote:
> > > > > > > > > 
> > > > > > > > > > Hello Mattijs, Marek
> > > > > > > > > > 
> > > > > > > > > > On Thu, Mar 20, 2025 at 10:47:02AM +0100, Mattijs Korpershoek wrote:
> > > > > > > > > > > On mer., mars 19, 2025 at 23:07, Marek Vasut <marex at denx.de> wrote:
> > > > > > > > > > > 
> > > > > > > > > > > > The UUU tool excepts the interrupt-in endpoint to be ep1in, otherwise
> > > > > > > > > > > > it crashes. This is a result of the previous hard-coded EP setup in
> > > > > > > > > > > > drivers/usb/gadget/epautoconf.c which did special-case EP allocation
> > > > > > > > > > > > for SPL builds, and which was since converted to this callback, but
> > > > > > > > > > > > without the special-case EP allocation in SPL part.
> > > > > > > > > > > > 
> > > > > > > > > > > > This reinstates the SPL part in an isolated manner, only for NXP iMX
> > > > > > > > > > > > SoCs, only for SPL builds, and only for the ep1in interrupt-in endpoint.
> > > > > > > > > > 
> > > > > > > > > > UUU can (and in our case is) used also on non-NXP i.MX platforms.
> > > > > > > > > > What should we do?
> > > > > > > > > 
> > > > > > > > > Do reproduce the problem (UUU tool crashes) on those platforms with
> > > > > > > > > recent U-Boot versions (v2024.10+) ?
> > > > > > > > 
> > > > > > > > Not tested, my comment is purely based on the code and the commit message.
> > > > > > > > Older U-Boot versions (up to v2024.04, included) are working fine, with UUU used
> > > > > > > > with TI K3 SoCs (AM69, AM62, AM62P).
> > > > > > > Are you talking about the NXP UUU ?
> > > > > > 
> > > > > > yes, it works just fine on not-NXP SoC.
> > > > > Then please test if it still works, and if not, this patch needs to be
> > > > > expanded to cover TI ... or apply unconditionally in SPL (sigh).
> > > > 
> > > > I just found the time to check some logs from our CI in which we execute such a
> > > > workflow.  We do use UUU only from U-Boot proper, so we are not going to be
> > > > affected by this SPL specific change.
> > > > 
> > > > However I found this error in the logs, on both a TI AM62 and an i.MX8MP
> > > > 
> > > > ```
> > > > Starting download of 40524455 bytes
> > > > dwc3-generic-peripheral usb at 31000000: request 0000000099f219c0 was not queued to ep1in-bulk
> > > > ..........................................................................
> > > > ..........................................................................
> > > > ..........................................................................
> > > > ..........................................................................
> > > > .............
> > > > downloading of 40524455 bytes finished
> > > > dwc3-generic-peripheral usb at 31000000: request 0000000099f219c0 was not queued to ep1in-bulk
> > > > dwc3-generic-peripheral usb at 31000000: request 0000000099f219c0 was not queued to ep1in-bulk
> > > > Starting download of 1675 bytes
> > > > dwc3-generic-peripheral usb at 31000000: request 0000000099f219c0 was not queued to ep1in-bulk
> > > > downloading of 1675 bytes finished
> > > > dwc3-generic-peripheral usb at 31000000: request 0000000099f219c0 was not queued to ep1in-bulk
> > > > dwc3-generic-peripheral usb at 31000000: request 0000000099f219c0 was not queued to ep1in-bulk
> > > > Starting download of 82 bytes
> > > > dwc3-generic-peripheral usb at 31000000: request 0000000099f219c0 was not queued to ep1in-bulk
> > > > downloading of 82 bytes finished
> > > > dwc3-generic-peripheral usb at 31000000: request 0000000099f219c0 was not queued to ep1in-bulk
> > > > dwc3-generic-peripheral usb at 31000000: request 0000000099f219c0 was not queued to ep1in-bulk
> > > > ```
> > > > 
> > > > the download is successful however, despite those error messages. Any idea? Is
> > > > this related to this topic?
> > > I have a feeling it is the UUU which should be fixed to not depend on the
> > > ep1-in , really.
> > > 
> > > You could however quickly try and apply the change in this patch not only to
> > > SPL, but to U-Boot on your board as well, and see if those warnings
> > > disappear.
> > 
> > So, I did a quick test and the issue is still there.
> > 
> > And looking better at the error now, this is about ep1in-bulk, that is
> > not affected by this change at all.
> > 
> > 
> > diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> > index 477ecd020985..55b248505de5 100644
> > --- a/drivers/usb/dwc3/gadget.c
> > +++ b/drivers/usb/dwc3/gadget.c
> > @@ -1632,23 +1632,7 @@ usb_ep *dwc3_gadget_match_ep(struct usb_gadget *gadget,
> >          if (usb_endpoint_is_bulk_out(desc))
> >                  return dwc3_find_ep(gadget, "ep2out");
> >          if (usb_endpoint_is_int_in(desc)) {
> > -               /*
> > -                * Special workaround for NXP UUU tool in SPL.
> > -                *
> > -                * The tool excepts the interrupt-in endpoint to be ep1in,
> > -                * otherwise it crashes. This is a result of the previous
> > -                * hard-coded EP setup in drivers/usb/gadget/epautoconf.c
> > -                * which did special-case EP allocation for SPL builds,
> > -                * and which was since converted to this callback, but
> > -                * without the special-case EP allocation in SPL part.
> > -                *
> > -                * This reinstates the SPL part in an isolated manner,
> > -                * only for NXP iMX SoCs, only for SPL builds, and only
> > -                * for the ep1in interrupt-in endpoint.
> > -                */
> > -               if (IS_ENABLED(CONFIG_MACH_IMX) && IS_ENABLED(CONFIG_XPL_BUILD))
> > -                       return dwc3_find_ep(gadget, "ep1in");
> > -               return dwc3_find_ep(gadget, "ep3in");
> > +               return dwc3_find_ep(gadget, "ep1in");
> >          }
> This should really be fixed in the UUU tool, not worked around in U-Boot.
> Can you try and see if the UUU tool can be fixed and not hard-code ep1 ?

I do not have time to look into that at the moment, I created an issue
https://github.com/nxp-imx/mfgtools/issues/472, hopefully NXP will have a look.

Francesco