Re: [PATCH v5 00/75] usb: dwc3: sync code with Linux v6.16

On 2026-08-10 15:26, Anshul Dalal wrote:
> On Mon Aug 10, 2026 at 1:52 PM IST, Anders Roxell wrote:
> > On 2026-08-06 18:47, Anshul Dalal wrote:
> >> On Thu, 16 Jul 2026 15:41:30 +0200, Anders Roxell <[email protected]> wrote:
> >> > [...]

[...]

> >
> > And in dwc3_gadget_uboot_handle_interrupt(), same file:
> >
> > -		struct dwc3_event_buffer *evt;
> > -		dwc3_thread_interrupt(0, dwc);
> > +		struct dwc3_event_buffer *evt = dwc->ev_buf;
> > +		dwc3_thread_interrupt(0, evt);
> 
> Thanks for the pointers Anders but I encounter a similar failure even
> with the diff you suggested. I enabled some logging for the SPL by
> defining the DEBUG macro in the following files (dfu.c spl.c spl_ram.c
> dwc3-am62.c composite.c g_dnl.c).

Hi Anshul,

Good news. I found two more bugs on top of the two I already sent you
(the event buffer and allow_connect). With all four the device
enumerates.

> 
> Failure logs:
> 
> 	U-Boot SPL 2026.07-00860-g5a9b7a715be2 (Aug 10 2026 - 15:06:27 +0530)
> 	SYSFW ABI: 4.0 (firmware rev 0x000c '12.1.0--v12.01.00 (Clever Cat)')
> 	Set clock rates for '/a53@0', CPU: 1250MHz at Speed Grade 'V'
> 	SPL initial stack usage: 17104 bytes
> 	SPL malloc() before relocation used 0x64d0 bytes (25 KB)
> 	>>SPL: board_init_r()
> 	spl_init
> 	Trying to boot from DFU
> 	g_dnl_register: g_dnl_driver.name = usb_dnl_dfu
> 	g_dnl_bind: gadget: 0x81c44a98 cdev: 0x81c66bc0
> 	usb_add_config: adding config #1 'usb_dnload'/81c67cc0
> 	g_dnl_do_config: configuration: 0x81c67cc0 composite dev: 0x81c66bc0
> 	adding 'dfu'/81c67d34 to config 'usb_dnload'/81c67cc0
> 	cfg 1/81c67cc0 speeds: high full
> 	usb_add_config: interface 0 = dfu/81c67d34
> 	g_dnl_bind: calling usb_gadget_connect for controller 'dwc3-gadget'
> 	usb_dnl_dfu: ready
> 
> <CTRL-C> from user
> 
> 	unbind function 'dfu'/81c67d34
> 	g_dnl_unbind: calling usb_gadget_disconnect for controller 'dwc3-gadget'
> 	Legacy image
> 	u_boot_pos = ffffffff
> 	Raw boot image support not enabled, proceeding to other boot methods
> 	Error: -22
> 	SPL: Unsupported Boot Device!
> 	SPL: failed to boot from all boot devices
> 	### ERROR ### Please RESET the board ###
> 
> Succes logs (same debug symbols ontop of main branch):
> 
> 	U-Boot SPL 2026.10-rc1-00209-g0bdd5e5daf95 (Aug 10 2026 - 14:58:28 +0530)
> 	SYSFW ABI: 4.0 (firmware rev 0x000c '12.1.0--v12.01.00 (Clever Cat)')
> 	Set clock rates for '/a53@0', CPU: 1250MHz at Speed Grade 'V'
> 	SPL initial stack usage: 17104 bytes
> 	SPL malloc() before relocation used 0x64d0 bytes (25 KB)
> 	>>SPL: board_init_r()
> 	spl_init
> 	Trying to boot from DFU
> 	g_dnl_register: g_dnl_driver.name = usb_dnl_dfu
> 	g_dnl_bind: gadget: 0x81c44aa4 cdev: 0x81c4af40
> 	usb_add_config: adding config #1 'usb_dnload'/81c4c040
> 	g_dnl_do_config: configuration: 0x81c4c040 composite dev: 0x81c4af40
> 	adding 'dfu'/81c4c0b4 to config 'usb_dnload'/81c4c040
> 	cfg 1/81c4c040 speeds: high full
> 	usb_add_config: interface 0 = dfu/81c4c0b4
> 	g_dnl_bind: calling usb_gadget_connect for controller 'dwc3-gadget'
> 	usb_dnl_dfu: ready
> 	composite_suspend: suspend
> 	set_config: high speed config #1: usb_dnload
> 	non-core control reqa1.03 v0000 i0000 l6
> 	non-core control req21.01 v0000 i0000 l4096
> ...(repeated)
> 	#non-core control reqa1.03 v0000 i0000 l6
> 	non-core control reqa1.03 v0000 i0000 l6
> 	DOWNLOAD ... OK
> 	Ctrl+C to exit ...
> 
> I'm not that familiar with the DW3 driver, let me know if you need me to
> test with any further changes.
> 
> >
> > evt was used there without being set as well.
> >
> > All of this is in v6 that I plan to send out. I have no AM62p and it is
> > only compile tested, so please try it on your board so we know if it
> > solves your issue before I send out v6.
> 
> I was able to reproduce the issue on beagleplay since it uses the same
> DWC3. You can try it locally if you have that board with you by
> resetting to DFU mode from U-Boot prompt:
> 
> => mw.w 43000030 53; res

I could not use the mw.w 43000030 53; res trick on the BeagleBoard-X15
that I found. But I can run DFU directly, it is the same DWC3 gadget
path you hit.

How I tested on x15:

  env set dfu_alt_info "${dfu_alt_info_ram}"
  dfu 0 ram 0

On the host dfu-util finds it:

  Found DFU: [0451:d022] ... name="kernel"/"fdt"/"ramdisk"

The two new bugs:

3. udc-core never sets dwc->async_callbacks. So the gadget setup() is
   never called and every GET_DESCRIPTOR returns -EINVAL.

4. ep0. The trb and the ep0 buffers are normal cached memory. In u-boot
   dma_alloc_coherent is not coherent, so nothing keeps them in sync with
   the controller dma. The resync dropped the manual flush and invalidate,
   so the controller reads a stale trb and stale buffers.

The patch below is only 3 and 4. Keep the first two you already have,
this goes on top. It is against v5. I built it and tested it on x15,
it works.

Can you test it on BeaglePlay with your reset trick? These fixes will
be part of the coming usb_dwc3_sync_v6 branch.

Cheers,
Anders


diff --git a/drivers/usb/dwc3/ep0.c b/drivers/usb/dwc3/ep0.c
index 0a4f69cbc28a..2ba05fc3f573 100644
--- a/drivers/usb/dwc3/ep0.c
+++ b/drivers/usb/dwc3/ep0.c
@@ -54,6 +54,9 @@ static void dwc3_ep0_prepare_one_trb(struct dwc3_ep *dep,
 	else
 		trb->ctrl |= (DWC3_TRB_CTRL_IOC
 				| DWC3_TRB_CTRL_LST);
+
+	dwc3_flush_cache((uintptr_t)buf_dma, len);
+	dwc3_flush_cache((uintptr_t)trb, sizeof(*trb));
 }
 
 static int dwc3_ep0_start_trans(struct dwc3_ep *dep)
@@ -811,6 +814,8 @@ static void dwc3_ep0_inspect_setup(struct dwc3 *dwc,
 	int ret = -EINVAL;
 	u32 len;
 
+	dwc3_invalidate_cache((uintptr_t)ctrl, sizeof(*ctrl));
+
 	if (!dwc->gadget_driver || !dwc->softconnect || !dwc->connected)
 		goto out;
 
@@ -856,6 +861,8 @@ static void dwc3_ep0_complete_data(struct dwc3 *dwc,
 	dwc->ep0_next_event = DWC3_EP0_NRDY_STATUS;
 	trb = dwc->ep0_trb;
 
+	dwc3_invalidate_cache((uintptr_t)trb, sizeof(*trb) * 2);
+
 	r = next_request(&ep0->pending_list);
 	if (!r)
 		return;
@@ -873,6 +880,9 @@ static void dwc3_ep0_complete_data(struct dwc3 *dwc,
 
 	length = trb->size & DWC3_TRB_SIZE_MASK;
 	transferred = ur->length - length;
+
+	if (ur->buf && !r->direction)
+		dwc3_invalidate_cache((uintptr_t)ur->dma, ur->length);
 	ur->actual += transferred;
 
 	if ((IS_ALIGNED(ur->length, ep0->endpoint.maxpacket) &&
diff --git a/drivers/usb/gadget/udc/udc-core.c b/drivers/usb/gadget/udc/udc-core.c
index 6355931f9ff1..f03ff937dfde 100644
--- a/drivers/usb/gadget/udc/udc-core.c
+++ b/drivers/usb/gadget/udc/udc-core.c
@@ -941,6 +941,8 @@ static inline void usb_gadget_udc_set_speed(struct usb_udc *udc,
 		gadget->ops->udc_set_speed(gadget, s);
 }
 
+static inline void usb_gadget_enable_async_callbacks(struct usb_udc *udc);
+
 static int udc_bind_to_driver(struct usb_udc *udc, struct usb_gadget_driver *driver)
 {
         int ret;
@@ -960,6 +962,7 @@ static int udc_bind_to_driver(struct usb_udc *udc, struct usb_gadget_driver *dri
                 driver->unbind(udc->gadget);
                 goto err1;
         }
+        usb_gadget_enable_async_callbacks(udc);
         udc->allow_connect = true;
         usb_gadget_connect(udc->gadget);
 
@@ -1156,6 +1159,7 @@ static void usb_gadget_remove_driver(struct usb_udc *udc)
 			udc->driver->function);
 
 	udc->allow_connect = false;
+	usb_gadget_disable_async_callbacks(udc);
 	usb_gadget_disconnect(udc->gadget);
 	udc->driver->disconnect(udc->gadget);
 	udc->driver->unbind(udc->gadget);