On Mon, Jul 23, 2012 at 12:56:02PM +0300, Felipe Balbi wrote: > Hi guys, > > So, this is why I have been mostly silent during last week. We were debugging > this failure with the TD7.06 of the Link Layer Compliance Suite - a mandatory > test for USB3 certification - and it turned out to be a silicon issue. > > The workaround is somwhat simple - don't use on-demand transfers for > data phase - but the outcome of that is a rather big rework of ep0 handling, > because we now need to be careful about starting control_data transfers > while still maintaining XferNotReady(DATA) handling for the error condition. > > It took a while to make sure everything was still as stable as before, but > after these patches TI's OMAP5 is passing all USB3 Certification tests (that > includes the entire USB2 test-suite which is mandatory for USB3 devices). > > Please give it a good round of test on your platforms, I will queue this > for v3.7 merge window. Note that I reserve the benefit of still rebasing > these patches until we get -rc1, but I might need to rebase dwc3 again > after that in order to avoid conflicts with Greg's usb-next. This will > happen only during this cycle because I want to make sure we don't get auto- > merge mistakes and end up breaking ep0 handling. > > Now that we have a certifiable solution, we plan to mantain it that way, so > before sending any patches for drivers/usb/dwc3, or drivers/usb/gadget/composite.c > or drivers/usb/gadget/udc-core.c, make sure to run at leat USB30CV and USB20CV > on your platform. Ideally, you should also pass all Lecroy's test-suite, including > the extra assertions provided by Lecroy itself as they provide good confidence > that your patch works properly. BTW, these patches are merged on top of my master branch if you want an easy setup to test. -- balbi
Attachment:
signature.asc
Description: Digital signature
