Comment 59 for bug 349767

Like Albert Weichselbraun in comment #51, I had a much better experience with kernel 2.6.31. I could not *really* reproduce the error, and if I did, then it was a much more benign version of the bug (see further down for details). For testing I used version 2.6.31-22.67 (amd64) from the Karmic repos, along with the accompanying header files to let dkms compile the current nvidia drivers (as installed in Maverick).

Using the Karmic kernel, my backup solution (backintime) managed to backup the complete set (full backup, not incremental) of several hundred GBs from 3 external USB disks to a 4th one (and the laptop's internal disk, too, but that one never posed a problem even with the buggy kernels).

I wrote that I could not *really* reproduce it because the backup dd not run in one go without a problem: at one point in the middle of rsync'ing, all external disks did disconnect, which certainly should not happen. However, they remounted automatically and the whole episode did not lead to the nasty logfile entries about usb issues, which have been posted repeatedly above. The backup continued, and the files that were missed during the disconnect were picked up on a subsequent run.

This is remarkably different to the behavior of the other reported kernels, where the disconnect (1) leads to the nasty logfile entries, and (2) after that has happened once, the usb system seems to be in an incoherent state and produces errors all the time even with disks that have not be disconnected before.