> > Also for usbfs be careful not to use USBDEVFS_URB_TYPE_INTERRUPT if
> > your host controller is OHCI, otherwise you will likely kill your
> > system.  Use BULK instead, directed to the INT endpoint.
> 
> I thought this was fixed in the latest round of usb-ohci patches?

No, that 0120 patch made unlinking interrupt URBs in
completion handlers work right.  I think all the HCDs
can now do that safely (except maybe usb-uhci).  It
means that for both INTR-IN and INTR-OUT, protocol
steps that transfer N interrupt packets and then finish
(by unlinking) can be portable.  That's the primary model
for INTR-OUT, AFAICT; the "period = 0" (UHCI) magic
is a special case, just for N=1.

For 2.5 I think all the "automagic resubmit" mechanism
for periodic (interrupt, isochronous) URBs should vanish.
Apart from ripping out urb->next, I think that'd just mean
simplifying the interrupt transfer code in HCDs (and likely
also in a few other drivers too like usbfs).

Some drivers would need to change, of course.  They'd be
making completion handlers re-queue transfers when that's
needed.  It'd seem like it's creating a new a failure mode
(the submit fails), but that's not so:  HCDs currently just drop
such errors on the floor, since there's no error channel letting
them be reported to drivers.

- Dave





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

Reply via email to