[PATCH] usb: ehci: Fix "EHCI timed out on TD - token=XXXX" error on ehci-hcd

[Marek Vasut]

On 3/2/20 8:54 PM, Tom Rini wrote:
> On Mon, Mar 02, 2020 at 08:19:22PM +0100, Marek Vasut wrote:
>> On 3/2/20 2:21 PM, Lukasz Majewski wrote:
>>> Hi,
>>
>> Hi,
>>
>>>> This patch aims to improve robustness of 'usb' command operation on
>>>> the ehci-hcd IP block as it ports to contemporary U-Boot the patch
>>>> described and provided in [1] (originally applicable to U-Boot
>>>> 2016.05).
>>>>
>>>> According to the fix author - "rayvt" (from [1]):
>>>>
>>>> TD stands for "Queue Element Transfer Descriptor (qTD)", which is a
>>>> micro-code instruction for the EHCI USB chip.
>>>> The "token" field is detailed information, control, and status of the
>>>> TD's data transfer. In particular, the rightmost byte is the status
>>>> field. 0x80 bit means the TD is active and the data transfer has not
>>>> yet completed. 0x08 bit means there was some sort of data transfer
>>>> error (XactErr).
>>> 	 ^^^^^^^ - this error is NOT handled now in mainline u-boot. Do
>>> 	 you have any experience with it?
>>
>> No, but then I also didn't run into those yet.
>>
>>>> If the drive gets a transfer error, it refuses to do any other I/O
>>>> transfer until the error is properly cleared and reset. U-boot did
>>>> not do this, so every subsequent disk read would get a timeout error
>>>> because the drive would not send any data. The fix is to properly
>>>> clear and reset the USB disk when it gets a transfer error.
>>
>> So basically, what this says is, "we crashed the drive, because the
>> drive suffered counter overflow, thus we fix it by resetting the drive".
>> No, the fix is to not crash the drive in the first place.
>>
>>>> Every read operation starts at the maximum block size. When the USB
>>>> pendrive is not able to correctly serve this data read request, the
>>>> dynamic reduction of IO size is performed. Up to six tries (with
>>>> smaller IO block each time) are attempted.
>>>>
>>>> A related problem is that some drives are slow to come up. Linux
>>>> handles this by issuing a spinup command and allowing more time for
>>>> the drive to respond. The same idea is applied in this fix.
>>>>
>>>> On TPC70 (i.MX6Q) once per ~10 times (without this patch):
>>>>
>>>> Bus usb at 2184200: USB EHCI 1.00
>>>> scanning bus usb at 2184200 for devices... 2 USB Device(s) found
>>>>        scanning usb for storage devices... 1 Storage Device(s) found
>>>> EHCI timed out on TD - token=0x1f8c80
>>>
>>> This is how the error gets evident. The detailed explanation is in link
>>> [1].
>>
>> Is there one specific post in that forum, or do I need to read through
>> the whole multi-page thread ?
>>
>> I would expect that if you run -- I assume 'usb reset' (the command is
>> missing above) -- then the bus gets power-cycled, hence the USB device
>> also gets power-cycled.
> 
> FWIW, the problem I have does not get fixed via "usb reset".  Ethernet
> never shows up again.  I do see right now that I disabled network tests
> on the platform a long while ago for unknown reasons, but did comment to
> myself elsewhere that network was fairly flaky in U-Boot.

There certainly is something with USB ethernet in U-Boot. I noticed the
flakiness myself, but didn't have time to investigate. Patches welcome I
guess ?