Re: dwc3-qcom: tx-fifo-resize regression on Poco F1 (sdm845) with v5.15-rc1

[Wesley Cheng]

Hi Amit,

On 9/21/2021 3:18 AM, Amit Pundir wrote:
> On Tue, 21 Sept 2021 at 13:00, Wesley Cheng <wcheng@xxxxxxxxxxxxxx> wrote:
>>
>> Hi Amit,
>>
>> On 9/21/2021 12:04 AM, Amit Pundir wrote:
>>> Hi Wesley,
>>>
>>> On Tue, 21 Sept 2021 at 02:44, Wesley Cheng <wcheng@xxxxxxxxxxxxxx> wrote:
>>>>
>>>> Hi Amit,
>>>>
>>>> On 9/20/2021 1:45 PM, Amit Pundir wrote:
>>>>> Hi Wesley, All,
>>>>>
>>>>> I see a reboot loop on Xiaomi Pocophone F1 (sdm845) with TX FIFO
>>>>> resize patches which landed in v5.15-rc1. Upstream commit cefdd52fa045
>>>>> "usb: dwc3: dwc3-qcom: Enable tx-fifo-resize property by default" to
>>>>> be specific, which switched on this feature by default.
>>>>>
>>>>> At times the phone crashes into the fastboot mode after the reboot
>>>>> loop, but mostly end up booting to UI after a while. This is what it
>>>>> looks like https://people.linaro.org/~amit.pundir/beryllium-userdebug/PXL_20210920_162749483.mp4.
>>>>>
>>>>
>>>> I believe Android will attempt a number of bootup sequences and if it
>>>> fails, it falls back to fastboot mode.  Are there any available logs you
>>>> might be able to collect to see where the issue is?
>>>
>>> It is a stock phone with no UART access, so I can't get early crash
>>> logs unless I'm booted up to adb shell. I can try getting some info
>>> using pstore-ramoops but warm reset support for sdm845 was not
>>> upstreamed when I tried using that the last time.
>>>
>>
>> I see, can we maybe avoid the actual resizing by commenting out the
>> following writel() calls, but let the fifo resize logic calculate the EPs?
>>
>> void dwc3_gadget_clear_tx_fifos(struct dwc3 *dwc)
>> {
>> ...
>>                 /* Don't change TXFRAMNUM on usb31 version */
>>                 size = DWC3_IP_IS(DWC3) ? 0 :
>>                         dwc3_readl(dwc->regs, DWC3_GTXFIFOSIZ(num >> 1)) &
>>                                    DWC31_GTXFIFOSIZ_TXFRAMNUM;
>>                 /* Comment the dwc3_writel() */
>>                 //dwc3_writel(dwc->regs, DWC3_GTXFIFOSIZ(num >> 1), size);
>>
>> and
>>
>> static int dwc3_gadget_resize_tx_fifos(struct dwc3_ep *dep)
>> {
>> ...
>>         /* Comment the dwc3_writel() */
>>         //dwc3_writel(dwc->regs, DWC3_GTXFIFOSIZ(dep->number >> 1), fifo_size);
>>         dwc->num_ep_resized++;
>>
>> Those 2 writel() would be the one that actually programs the TXFIFO
>> register.  I hope when commented out, no resize should actually happen
>> anymore.
>>
>> With this, hopefully we can get some logs from the device at least :)
> 
> I can boot fine with above 2 writel() removed but I don't see EP
> counts being printed in the log anywhere and the only relevant
> message in dmesg I see is this:
> 
>     Duplicate name in dwc3@a600000, renamed to "tx-fifo-resize#1"
> 
> which is interesting because I don't see tx-fifo-resize property being
> set by any of the upstream sdm845 or relevant dtsi files.
> 
> 
Thanks for testing.  What is the current composition the device is
exposing to the host?

Can we get a ftrace log w/ the DWC3 events enabled, as well as the
kernel log? (if not at boot up then maybe a plug in/out)  This will at
least help us identify when the DWC3 ep enable is being called (which
triggers the resize routine)

Thanks
Wesley Cheng

-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project