Re: [PATCH v11 00/30] usb: dwc3: part 2, make the resynced code work

[Michal Simek <michal.simek@amd.com>]

On 11/10/26 16:18, Marek Vasut wrote:
> On 11/7/26 8:21 AM, Anders Roxell wrote:
> 
> Hello everyone,
> 
>> This is part two of two. It applies on top of part one [1], which resets
>> the DWC3 driver and imports it up to Linux v6.16. Part one alone breaks
>> the build for every board that enables DWC3. This part adds back the
>> U-Boot glue, the XHCI and UDC updates and the build fixes, so the tree
>> builds and the gadget works again. It builds on Jerome's work [2].
>>
>> The two parts have to go in together. Part one did not change, so I do
>> not resend it. The v10 patches [1] still apply on ab4dc5aacf95.
>>
>> The resynced epautoconf matches endpoints on their caps field instead of
>> on their name. Five UDC drivers never filled that in, so each one gets a
>> patch next to its other fixup: at91_udc, atmel_usba_udc, ci_udc, musb-new
>> and mtu3. With those and dwc2, every UDC driver in the tree sets caps.
>>
>> Build tested one board per driver: every DWC3 glue layer in the tree,
>> dwc3 on riscv (spacemit_k1), RK3588 and RK3399, dwc2 with and without
>> DM_USB_GADGET, at91_udc, atmel_usba_udc, ci_udc with SPL SDP, musb-new
>> on sunxi and in the am335x_evm SPL, and mtu3. 24 boards, no warnings.
>>
>> Ilias found a few bisect gaps, so I built eight of the boards at every
>> patch. Gadget configs build again from patch 20, the other UDC drivers
>> from their own fixup patch.
>>
>> Tested on hardware with v10, per vendor:
>>
>> - AMD/Xilinx: Love tested and sent a Tested-by.
>> - TI: DFU and ums on BeagleBoard-X15 (me), and on AM62p and BeaglePlay
>>    (Anshul).
>> - NXP: ci_udc with ums on an i.MX93 board (me).
>> - Amlogic: gadget and host on Khadas VIM3 (Mattijs).
>> - Rockchip: host on RK3399 Puma, same as master (Quentin). RK3588
>>    Tiger had two regressions, see below.
>>
>> Tested on hardware with v11:
>>
>> - NXP: ci_udc with ums on i.MX93, chainloaded from the vendor U-Boot.
>>    Three ums rounds, the host read 64 MiB each time. It needs usbotg1 in
>>    peripheral mode.
>> - TI: dwc3 ums on BeagleBoard-X15, the host enumerates it at high
>>    speed. That is where I found the event buffer bug in patch 15.
>> - Rockchip: host on RK3399 Puma and RK3588 Tiger (Quentin). The two
>>    Tiger regressions from v10 are gone, and he sent a Tested-by:
>>
>>    Tested-by: Quentin Schulz <quentin.schulz@cherry.de> # RK3399 Puma 
>> usbdrd_dwc3_1
>>    Tested-by: Quentin Schulz <quentin.schulz@cherry.de> # RK3588 Tiger 
>> usb_host1_xhci + usb_host2_xhci
>> - Rockchip: host and gadget on RK3576, a Flipper One and an RK3576 EVB1
>>    (Alexey). No regressions against master, and he sent a Tested-by:
>>
>>    Tested-by: Alexey Charkov <alchark@flipper.net> # Rockchip RK3576
>>
>> The rest is build tested only.
> I would like to move this series forward, so let me get input on the status. 
> Below are the vendors which use DWC3 core that I could find in tree, can I get a 
> go/no-go on this series from each of you ? I hope the CC is complete, if not, 
> please add the missing people.
> 
> allwinner
> altera
> amlogic
> apple
> broadcom
> exynos
> freescale
> microchip
> qcom
> rockchip
> socionext
> st
> ti
> xilinx
Go from AMD/Xilinx side.

Thanks,
Michal