On Tuesday 22 March 2005 10:28 am, Lothar Wassmann wrote:
> Hi,
> 
> > There are also chiprev-specific issues on PXA.  If you're using
> > a PXA 250, especially with B steppings, you might see issues
> >
> No, we are only using PXA255, because the 250's were just too buggy.
> For the PXA255 I found only one Erratum regarding SET/CLEAR FEATURE
> requests that respond with ACK instead of STALL when an illegal
> feature is requested and some minor documentation changes.

See what I mean about Intells errata documentation being incomplete?
That hardware has quite a few more problems than that.  :)


> > Hmm, I don't remember seeing such failures.  I probably wouldn't
> > for "test9", using a fast OHCI host.  But "test10" uses some of
> > those, and so I'd expect to see such errors with it.  Can you
> > reproduce the failures with an OHCI host?
> > 
> I ran the test with a PXA270 OHCI HC and the pxa27x_udc driver from:
> http://www.free-electrons.com/pub/drivers/pxa27x-0218.patch
> but it just hangs after completing test 9 successfully.

I have no reason to believe there's a functional pxa27x UDC driver
yet.  But it'd surprise me if their OHCI was problematic; it's
way too easy to license a good working OHCI and toss it into your
silicon integration soup.


> But the driver only sends out one request at the start of every other
> frame, so the problem cannot occur.

I came across a description of some limitations of MSFT's USB stack
recently, which said it can't issue more than one URB per frame.
That might partially explain why you're seeing som of this...

 
> > Are you sure you're not just cutting the end-of-frame timings
> > a bit too close?  Maybe you shouldn't send the SETUP quite so
> > near the next frame's SOF.  (If you have control over that...)
> > I remember a similar issue with the SL811 code.
> > 
> Unlike the SL811 (which is more or less only a parallel to USB serial
> converter) the ISP1362 won't start a transfer that would overrun the
> end of a frame. You can safely queue up more data than fits into a
> frame and the controller will spread the packets across multiple
> frames without errors. There is normally no need to check the
> allocated frame time by software.

That sounds good ...


> Nevertheless I tried to move the SETUP transfer away from the
> end of the frame by checking the HcFmRemaining register before
> queueing a SETUP request. This indeed made this error disappear.

... that sounds better!

- Dave

 
> 
> Lothar Wassmann
> 


-------------------------------------------------------
This SF.net email is sponsored by Demarc:
A global provider of Threat Management Solutions.
Download our HomeAdmin security software for free today!
http://www.demarc.com/info/Sentarus/hamr30
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to