On Sat, 3 Mar 2007, Gordon Messmer wrote:

> > The patch could shorten the time window from 2 seconds to 0.5 seconds.  
> > That would reduce the number of spurious resets even more, but it wouldn't 
> > eliminate them entirely.
> 
> And it wouldn't eliminate the IO errors the way that booting up
> and old kernel does.

Not for you, no.  It might help other people, though.  The cables used for 
USB mice are by design less robust than other cables.  Partly it's because 
mice run at low speed and so are relatively more immune to noise, and 
partly it's because a mouse with a thick, difficult-to-bend cable would be 
harder to use.

> > I don't see any regular pattern in those time intervals, do you?
> 
> Well, no, not now.  However, when the problem originated, it was
> very reliably 10 seconds of idle before it froze.  Or 90.  It
> depended on which kernel I was running.
> 
> I didn't know about usbmon a year ago or I would have had more
> information.  Maybe I can boot up one of those old FC5 kernels and
> see what happens.
> 
> Even now, there are periods where the timing is fairly stable, and
> I could be convinced that it was regular if I didn't watch it long
> enough.  I did a quick check this morning and saw a short sequence
> of IO error, 25 second delay, reset, 8 second delay, repeat (example
> is attached).  But after a few cycles, the timing became more erratic.

The actual specings between errors are 24.65, 7.70, 25.26, 7.61, 24.29.  I 
wouldn't bet on that being meaningful, but it could be.

> > Which 
> > leads me to think the cause is something more or less random, like 
> > electromagnetic interference. 
> 
> Do you think the interference could be one motherboard component
> interfering with another?  Perhaps one that wasn't supported, or was
> handled in a different manner under earlier kernels?

Yes, that's what I had in mind.

> It's a hell of a mystery.  I've tried building the new kernel
> using the old kernel's .config to eliminate configuration changes.
> I noticed that the new kernel uses CONFIG_HZ_1000, where the old
> one used CONFIG_HZ_250.  There's also CONFIG_USB_EHCI_TT_NEWSCHED
> set in the new kernel.  When the kernel was built with the old
> configuration, the resets and IO errors were still present.

Best to keep the configs the same as much as possible.

> I wonder if I could install FC4 on this host and use an FC6 kernel.

You're better off using an older kernel with a newer distribution.  Doing
it the other way might not work, because of changes to essential packages
like udev needed for booting.  The changes tend to be backward-compatible 
but not forward-compatible.

> I continue to be grateful for your assistance in looking at this.
> It may be just a hardware flaw exposed by changes in the kernel,
> but it's a confounding mystery to me, and I hate those.

Well, my advice still stands.  Try various intermediate kernels to narrow 
down the range where the problem was introduced.

Also, you're better off using vanilla kernels from kernel.org (or mirrors)  
rather than customized ones from Fedora, because we have a complete record
of all the changes made in the course of their development.

Alan Stern


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Linux-usb-users@lists.sourceforge.net
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-users

Reply via email to