Hi, On Mon, Aug 29, 2011 at 03:19:06PM +0200, Sebastian Andrzej Siewior wrote: > >> >composite has a 1k buffer. The buffer has to be big enough for a big > >> >load of descriptors. > >> > >> If so, why not unconditionally queue maxpacket size then? The host does > >> not see a difference. > > > >What if we have controllers which need request to be passed with correct > >size ? I mean, what if there's some DMA engine where if you don't start > >a transfer with correct size, it will never finish ? > > It is a short transfer in OUT direction. Don't these have to finish > anyway? the same way dwc3 has a quirk where it won't complete transfers which aren't aligned on wMaxPacketSize (on OUT direction), there could be another controller with a different quirk. > >Until we know this won't cause problems for anyone else, I rather keep > >this workaround on dwc3 driver, since this is purely a dwc3 limitation. > > Yes, I'm okay with this workaround beeing dwc3 specific. In the long > term I would prefer not having udc have to work around gadget's bugs. it's not a bug on the gadget at all. The gadget is doing exactly what the specification mandates. The bug, lies in dwc_usb3 IP. -- balbi
Attachment:
signature.asc
Description: Digital signature