Backporting 3.6 USB/EHCI fixes for XO-1

Daniel Drake dsd at
Mon Sep 10 18:54:36 EDT 2012


Kevin Gordon showed me a case of a USB microscope (UVC) not working on
XO-1. It confuses the host controller and even knocks the wifi off the
bus. On XO-1.5 and XO-1.75 it is fine. These microscopes are used in
the field (e.g.

We're not alone with this class of problem; in other cases, hardware
oddities (in the host controller) have been shown to be the cause.
Anyway, a load of work landed in Linux 3.6 to address this, to make
the EHCI driver more resilient to odd hardware. Backporting to 3.3 is
very simple, and fixes the problem. The commits in question are:

USB: EHCI: resolve some unlikely races
USB: EHCI: fix up locking
USB: EHCI: initialize data before resetting hardware
USB: EHCI: simplify isochronous scanning
USB: EHCI: use hrtimer for the I/O watchdog
USB: EHCI: always scan each interrupt QH
USB: ehci-sched.c: remove dbg() usage
USB: EHCI: don't lose events during a scan
USB: EHCI: use hrtimer for unlinking empty async QHs
USB: EHCI: unlink multiple async QHs together
USB: EHCI: use hrtimer for the IAA watchdog
USB: EHCI: don't refcount iso_stream structures
USB: EHCI: use hrtimer for (s)iTD deallocation
USB: EHCI: use hrtimer for controller death
USB: EHCI: use hrtimer for interrupt QH unlink
USB: EHCI: use hrtimer for async schedule
USB: EHCI: remove PS3 status polling
USB: EHCI: return void instead of 0
USB: EHCI: use hrtimer for the periodic schedule
USB: EHCI: introduce high-res timer
USB: EHCI: add new root-hub state: STOPPING
USB: EHCI: add pointer to end of async-unlink list
USB: EHCI: rename "reclaim"
USB: EHCI: add symbolic constants for QHs
USB: EHCI: don't refcount QHs
USB: EHCI: remove unneeded suspend/resume code
USB: EHCI: initialize data before resetting hardware
USB: fix PS3 EHCI systems
USB: EHCI: fix command register configuration lost problem
USB: EHCI: improve full-speed isochronous scheduling routine
EHCI: maintain the ehci->command value properly
EHCI: keep track of ports being resumed and indicate in hub_status_data
EHCI: don't try to clear the IAAD bit

 10 files changed, 1123 insertions(+), 967 deletions(-)

I'm proposing backporting this to 3.3 for the 13.1.0 release
(affecting XO-1 and XO-1.5). I feel comfortable about it; this all
landed in 3.6-rc1 and hasn't needed followup in the form of later
fixes, we generally get a good amount of external device testing (I'm
sure Kevin will test a gazillion devices as well), and if it presents
problems which don't have obvious fixes we can revert.


More information about the Devel mailing list