On Wed, 20 Jun 2018, Sebastian Andrzej Siewior wrote: > On 2018-06-20 12:21:33 [-0400], Alan Stern wrote: > > > > There are two problems with using usb_fill_int_urb() to initialize an > > > > isochronous URB: > > > > > > > > The calculation of the interval value is wrong for full-speed > > > > devices. > > > > > > Why wrong? > > > > Because the interval value in the FS endpoint descriptor uses a > > logarithmic encoding, whereas the value stored in the URB uses a linear > > encoding. Simply copying one value to the other will store an > > incorrect number, except in the lucky cases where the interval is > > either 1 or 2. > > Hmmm. Now that I looked into USB 2.0 specification it really says > logarithmic encoding for ISOC endpoints on every speed and not just HS. > And INTR endpoints have logarithmic encoding only for HS. I remembered > it differently… > > But based on this, we should really introduce usb_fill_iso_urb() which > handles this correctly. Also the documentation should be updated because > the suggestion for ubs_fill_intr_urb() is misleading (especially if you > have wrong memory about the encoding on FS). That sounds like a good thing to do. Alan Stern
