https://bugzilla.kernel.org/show_bug.cgi?id=212955 --- Comment #6 from Alan Stern (stern@xxxxxxxxxxxxxxxxxxx) --- Unfortunately the logs and traces didn't show anything definitive. The working and non-working versions are essentially the same up to the point where in the non-working version, the devices fails to respond to a Set-Interface request. There's no indication of why the failure occurred. The same request succeeded earlier in the trace, and it succeeds in the working trace. Incidentally, you can restrict the usbmon trace to include only devices on a particular bus by copying the appropriate debugfs file. For example, to include only communications with devices on bus 4, you would copy the 4u file. Can you collect a similar usbmon trace for the USB-2 port under an earlier kernel where it was working? Although there have been a few changes to the ehci-hcd driver in the last couple of years, I don't see any changes that would have this effect. The only commit which might be relevant is 64cc3f12d1c7 ("USB: EHCI: Do not return -EPIPE when hub is disconnected"), and that would affect only what happens after the error has occurred -- and it is already present in 5.9. You can try reverting the commit anyway, just to see if it really does make any difference. In the end, the only was to solve the puzzle may be bisection. You can cut down the size of a git clone operation by using the --shallow-exclude= option. In your case, you would exclude everything before v5.9 (or whatever kernel version you identify as most recently working). -- You may reply to this email to add a comment. You are receiving this mail because: You are watching the assignee of the bug.
