On Wed, Jul 20, 2011 at 05:16:41PM -0400, Alan Stern wrote: > On Wed, 20 Jul 2011, Sarah Sharp wrote: > > > Here's the updated RFC for the software bandwidth checking quirk for the > > Intel Panther Point xHCI host. > > > > It's stable now, bug free IFAIK, and seems to track bandwidth properly. > > If I don't use a webcam that always requests a 3060 byte payload > > (requiring uvcvideo to always request the most bandwidth-hungry > > alternate interface setting), I can create a setup with an almost > > maxed-out setup where I can view some lower video resolutions, but not > > all of them. (This is with several LS/FS devices attached to two > > multi-TT hubs, one of them being a true 7-port hub.) All the while, > > previously added devices like mice, keyboards, ethernet adapters, and > > USB-to-serial adapters work correctly. > > > > This same setup under EHCI causes issues, with lots of Xact errors. The > > As has been pointed out repeatedly, the EHCI periodic scheduling code > is full of bugs. No doubt you are running across some of them. Yes, that was my intent. :) I was trying to make sure that it was an over-subscribed bus by verifying that the EHCI driver had issues with scheduling. Of course, that's not a true test, because the scheduler could have bugs with nearly over-subscribed schedules, but it was a close enough approximation for me. > > At that point, the xHCI driver > > should be removing the endpoints that were added during the > > re-enumeration. > > At this point in the re-enumeration, the only endpoint to have been > added is ep0. This occurs before the first call to > usb_hcd_alloc_bandwidth(). > > > I'm not certain exactly how to fix that issue, and given that the USB > > core does something like overwrite the descriptor values here, I'd > > rather treat descriptors as private to the USB core, and not use the > > endpoint descriptor pointers. > > That's up to you. But this is not a genuine problem. Ah, ok, I hadn't noticed that. Still, I think I want to make xHCI keep a copy of the information, at least at for a first pass. Any objections before I send this off to Greg? Sarah Sharp -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
