On Mon, Mar 25, 2002 at 01:46:57PM +0100, Georg Acher wrote:
> On Mon, Mar 25, 2002 at 12:21:48PM +0000, Nick Craig-Wood wrote:
> > The data stream gets corrupted in a very specific way - there are no
> > bytes added or removed to the serial output but there is one byte
> > overwritten with null (0x00) with 0,1 or 2 more corrupted bytes after
> > it.
> 
> Ouch... that looks like a problem for non-cacheline-sized transfers. Does it
> also happen with 48bytes?

I satisfied myself that the problem transfer lengths were 32..62 bytes
as passed to write().  63 bytes transfer seemed to be OK as did 1..31
bytes.

The driver adds 1 byte of length to the start of the packet.  The max
transfer size is 63 bytes.

  OK  2..32 bytes
  Bad 33..63 bytes
  OK  64 Bytes

> > However (and this is what makes a simple bug report not so simple) the
> > unpatched keyspan driver works absolutely perfectly on my desktop
> > machine.  On my laptop it doesn't give this error either but it spends
> > so much time messing around in the BIOS with power management stuff
> > that it is always dropping packets.
> 
> Intel makes the "best" UHCIs with the least flaws... We have discovered that
> by the hard way. For our DAB monitoring system via USB (384KB/s ISO
> streaming) we only use mainboards with Intel chipsets. We have tried VIA and
> SiS in the beginning, but they weren't as reliable for USB as Intel
> mainboards and caused much trouble for us and our customers... Especially
> VIA seems to have great problems with PCI-arbitration, there is
> significantly more lost data in the streamed data, and the "famous"
> IDE+Soundblaster-bug speaks for itself...

Thank you for that advice - it has been noted well!  We will need to
order 20 motherboards soon so it will be worth our while searching for
one with an Intel USB chipset I think.

> > Does anyone have any more ideas as to what is going on or ideas as to
> > what to try next?
> 
> If there's a BIOS that's worth mentioning, have you tried to play a bit with
> the PCI settings (posted write, buffering, latency, etc)?

There is a quite normal BIOS.  I'm afraid I don't normally play with
these things in the BIOS - do you have any specific ideas (eg increase
latency, disable buffering or something)?

Thanks

Nick
-- 
Nick Craig-Wood
[EMAIL PROTECTED]

_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to