[RFT PATCH v1 5/5] usb: Handle QT_TOKEN_STATUS_XACTERR error when sending data

[Marek Vasut]

On 3/23/20 8:18 AM, Lukasz Majewski wrote:
> Hi Marek,

Hi,

>> On 3/22/20 2:00 PM, Lukasz Majewski wrote:
>>> This code adds check if QT_TOKEN_STATUS_XACTERR error occurred.
>>> When it is detected the token is reconfigured and transmission is
>>> retried.
>>>
>>> This code is the port to newest U-Boot of the fix from - "rayvt"
>>> (from [1]).
>>>
>>> Links:
>>> [1] - https://forum.doozan.com/read.php?3,35295,35295#msg-35295
>>> [2] -
>>> https://www.dropbox.com/s/nrkrd1no63viuu8/uboot-bodhi-2016.05-timeoutTD.patch?dl=0
>>>
>>>
>>> Signed-off-by: Lukasz Majewski <lukma at denx.de>
>>> [Unfortunately, the original patch [2] did not contain S-o-B from
>>> the original author - "rayvt"]
>>> ---
>>>
>>>  drivers/usb/host/ehci-hcd.c | 27 ++++++++++++++++++++++++++-
>>>  1 file changed, 26 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/usb/host/ehci-hcd.c
>>> b/drivers/usb/host/ehci-hcd.c index 0a77111f80..45eda7ad24 100644
>>> --- a/drivers/usb/host/ehci-hcd.c
>>> +++ b/drivers/usb/host/ehci-hcd.c
>>> @@ -315,6 +315,7 @@ ehci_submit_async(struct usb_device *dev,
>>> unsigned long pipe, void *buffer, int timeout;
>>>  	int ret = 0;
>>>  	struct ehci_ctrl *ctrl = ehci_get_ctrl(dev);
>>> +	int trynum;
>>>  
>>>  	debug("dev=%p, pipe=%lx, buffer=%p, length=%d, req=%p\n",
>>> dev, pipe, buffer, length, req);
>>> @@ -560,6 +561,10 @@ ehci_submit_async(struct usb_device *dev,
>>> unsigned long pipe, void *buffer,
>>> ehci_writel(&ctrl->hcor->or_usbsts, (usbsts & 0x3f)); 
>>>  	/* Enable async. schedule. */
>>> +	trynum = 1;	/* No more than 2 tries, in case of
>>> XACTERR. */ +retry_xacterr:;
>>> +	vtd = &qtd[qtd_counter - 1];
>>> +
>>>  	cmd = ehci_readl(&ctrl->hcor->or_usbcmd);
>>>  	cmd |= CMD_ASE;
>>>  	ehci_writel(&ctrl->hcor->or_usbcmd, cmd);  
>>
>> Are you sure always retrying the transfer if you get XACTERR is the
>> right thing to do, even if you get that e.g. on filesystem writes ?
> 
> Please correct me if I'm wrong, but this function - ehci_submit_async
> - doesn't work with filesystem. It just setups proper EHCI descriptor
>   and checks if it was sent with or without errors.

If you're writing into a block device, this function is used. If you get
XACTERR during the transfer and you retry, is there a chance the data
get actually written to the device ?

> When the XACTERR happens, proper flag is cleared and the transmission
> is retried.

What happens to the payload if it's a host-to-device transfer, do the
data get discarded or do they get written to the storage ?

-- 
Best regards,
Marek Vasut