Hello,

On Saturday 06 July 2002 04:58, David Brownell wrote:

> Hmm, that's only a semi-polite failure mode.

I call it a bloody nuisance since I have to reboot each time :)

> The OHCI driver has had some longstanding issues with unlinking
> URBs and endpoints, it's likely that some of the fixes in that
> area made some other failures more likely.  They made OHCI work
> on some hardware that's failed rudely before ... but you know
> the old "bugs hiding other bugs" game!  Or maybe there was a bug
> in one of those fixes, and you happened to see it.
>
> I've got one fix in that area still queued, which I'll send
> separately on the chance it makes a difference in this case.

That fix made only things worse: not a single device to accepts an address 
anymore. Actually, this looks a lot like some of the problems I had today 
with 2.5.20. I'm under the impression that the OHCI controller is kinda 
picky whether or not any devices are plugged in when activating the chip.

See pwc-24-ohci+patch file

>  > The same code works flawlessly on 2.5.20, so something must have been
>  > changed between those versions. I've added some debugging statements
>  > to my driver (see attached source file) ...
>
> Can you resend similar output but with the OHCI debugging turned on?

See attached pwc-24-ohci-debug file.

> Not the verbose stuff, at least not at first.  It doesn't help much
> to see what layers _above_ the driver are doing.  It would be good to
> enable the commented-out dbg() towards the top of ohci_irq() in the
> ohci-hcd.c file, too, but that might spit out rather a lot of data
> if you're streaming video (one irq per microsecond).

One per millisecond... microsecond interrupts would kill the processor :)) 
I've added a tiny bit of code to count the number of URBs processed between 
open() en close(). The CamStream probe leads to 0 URBs (I've started 
CamStream to see if I could get 1 or more URBs processed; only got 0).

> That's happened to me with CPiA ever since the V4L2 conversion, and
> regardless of which HCD was used.  I recall that down_interruptible
> call wasn't even in the "cpia_usb" module.  Or did you see that work
> in 2.4.20 somehow?

Nope. Must be the V4L2 problem. Kinda irritating, though.

 - Nemosoft

Attachment: pwc-24-ohci-debug.gz
Description: GNU Zip compressed data

Attachment: pwc-24-ohci+patch.gz
Description: GNU Zip compressed data

Reply via email to