I just read the notes from the last two weeks of USB latency
debugging. Very interesting results! Could somebody who was there
please document whether any of the tests were performed with a hub in
the system, and what the topology was in that case?

Sarah says:

- Re-try the experiment of putting more load on the bus, but rather
than using a kernel driver to create the load, use some existing
libusb-based application. Perhaps a gphoto2-based camera? Hypothesis
under test: There's something wrong with usbfs, and creating load with
a kernel driver is bypassing it somehow. If creating load using only
usbfs doesn't improve the latency measurements, then this hypothesis
looks likely.

- Ask Alan Stern for advice, CC'ing the linux-usb and libusb mailing
lists. If there's a problem in the UHCI driver, or a documented
limitation in the specification, writing a kernel driver isn't going
to help. If there's an issue with usbfs, that should get fixed.

- Try turning on CONFIG_USB_EHCI_TT_NEWSCHED and seeing if that helps
when the device is behind a high-speed hub. (It won't have any effect
when the GFE is plugged into a host port directly, since in that case
the UHCI driver will claim the device, rather than the EHCI driver.)

- Try turning off CONFIG_USB_DEBUG and seeing if that makes the delay go away.


psas-avionics mailing list

Reply via email to