On Wed, 24 Apr 2013, David Henningsson wrote:
Ok, Joseph Salisbury has build some bisection kernels, and Tyson Tan relentlessly tested all of them, and it turns out that
3e619d0415 ("USB: EHCI: fix bug in scheduling periodic split transfers")
Is the first bad commit. Also, reverting this commit from the current mainline head makes the problem disappear.
Alan, any idea?
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1136110
No ideas as to the cause.
For debugging, it would help to see a usbmon trace from a kernel where the problem occurs, together with a trace from another kernel where the problem does not occur.
Alan Stern
First, thanks for looking at this bug.
While a long-term solution is being discussed, this patch went to stable too, where it is causing regressions. Would it be okay just to revert this patch in the next stable series? (Even if this was a bug fix, few people seem to have noticed?) Or do you envision something else happening but the original -ENOSPC error showing up, due to other stuff that went to stable at the same time?
I think reverting it from stable won't cause any new problems to show up (although the original problem will return, obviously). Go ahead.
At the same time, it would be nice to know that some people are really carrying out tests to find the reason for the new problem. Otherwise we end up with this commit reverted and nobody trying to figure out how to fix the original bug correctly.
Alan Stern