On Monday 19 January 2009, Sergei Shtylyov wrote: > > Sounds better. Of course, the older "RNDIS" mode wasn't > > actually fully conformant with RNDIS specs, or it could > > I haven't even seen this spec. :-)
Grab them from the MSFT website somewhere. Ancient, incomplete, badly designed ... but downloadable. The issue is packet termination. If the DMA descriptor says "N bytes", that should terminate after either (a) a short packet, (b) RX N bytes, even if that didn't involve a short packet termination, or (c) on error, including RX of N+ bytes, with the last packet too big to fit into what's left of the buffer. The older mode got at least (b) wrong. > > have been used for all transfers. All RNDIS did was > > follow the DMA policy that (non-UHCI) PCI based host > > controllers use to avoid extra IRQs. > > Oh, UHCI can't coalesce packets? Its DMA descriptors are for single packets. _______________________________________________ Davinci-linux-open-source mailing list [email protected] http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
