[U-Boot] [PATCH] usb: fix usb start problem with SMSC USB hub and Toshiba USB stick

[Anatolij Gustschin]

Hi Remy,

On Wed, 1 Sep 2010 10:49:22 +0200
Remy Bohmer <linux at bohmer.net> wrote:
...
> 2010/8/31 Anatolij Gustschin <agust at denx.de>:
> > Checking the status field of the qTD token in the current code
> > do not take into acount cases where endpoint stall (halted) bit
> > is set together with some other status bits. As a result clearing
> > stall on an endpoint won't be done if other status bits were set.
> 
> Reading this description and this code:
> 
> >        /* special handling of STALL in DATA phase */
> > -       if ((result < 0) && (us->pusb_dev->status &
> > USB_ST_STALLED)) {
> > +       if ((result < 0) &&
> > +           (us->pusb_dev->status & (USB_ST_STALLED |
> > USB_ST_CRC_ERR))) { USB_STOR_PRINTF("DATA:stall\n");
> 
> Description and code do not seem to match. The old implementation
> explicitly checked if the STALLED bit was set, and if so the endpoint
> would be cleared. Now, it seems you are clearing the endpoint _also_
> when the CRC_ERR bit is set.

Yes. This was intentional but unfortunately not directly mentioned
in the patch description. The reason is the following:

in all cases where transaction error (XactErr) was reported with
additionally STALLED bit set, the handling of the STALL condition
successfully recovered from the error case and the device was
recognized.

in cases where only transaction error was reported (STALLED bit
not set) the recovery from the error case didn't succeed and
"Device NOT ready" (Request Sense returned 06 28 00) status was
finally reported. I didn't find the information in the EHCI spec
how the recovery from the XactErr should be done properly. Do
you know how to handle it?

> There are a lot more reasons, at least on other host controllers that
> set the status USB_ST_CRC_ERR, does this change not break the other
> code? (A simple grep will show you the situations where it is
> returned.)

This change will probably break other code. I'll drop it here and
additionally set USB_ST_STALLED flag in the ehci driver instead.

> > E.g. 'usb start' often fails with Toshiba USB stick 0x930/0x6545
> > connected to the SMSC USB 2.0 hub 0x424/0x2514. Debugging the
> > issue showed that while bulk IN transfers with length of 13 or
> > 18 the status field of the qTD token sometimes indicates trans-
> > action error (XactErr) and sometimes additionally endpoint halted
> > state. In the latter case resetting the USB device in error
> > recovery code fails as no clear stall request on the endpoint
> > will be done.
> 
> OK
> 
> > However this fix is not enough to solve 'usb start' problem
> > with hub/stick combination mentioned above. Running with lot of
> > debug code in ehci_submit_async() I've never seen the problem
> > with usb stick recognition. After removing this debug code the
> > similar problem sometimes showed up again. Therefore the patch
> > also adds delay in ehci_submit_async() for above-mentioned
> > hub/stick combination. Even without this delay the fix is an
> 
> Why only for these USB sticks? Are these sticks special among the
> thousands of different sticks out there?

I do not have the right equipment to analyze this problem properly.
This is the special USB stick-hub combination. The same USB stick
is always recognized properly if it is connected directly. Other
USB sticks also work with this USB hub.

> Or could it be more reliable to always do this delay for all USB
> sticks? Or even better, what are we missing here that the delay is
> needed at all?

Doing this delay for other USB sticks will negatively affect data
transfer speed.

> > diff --git a/drivers/usb/host/ehci-hcd.c
> > b/drivers/usb/host/ehci-hcd.c index 37d056e..7463a75 100644
> > --- a/drivers/usb/host/ehci-hcd.c
> > +++ b/drivers/usb/host/ehci-hcd.c
> 
> Is this problem only valid for the EHCI code? I can imagine that the
> other host controllers (like UHCI and OHCI and so on) code suffer the
> same problem and need similar fixes...
> (At least I know that I have here a couple of USB sticks that show
> similar problems with OHCI ;-) )

Maybe, maybe not. The problem shows up on SheevaPlug and mpc5121ads
boards, both using USB EHCI. I do not have the possibility to test
it on other host controllers currently.

Best regards,
Anatolij