Ok, I've been going back through the mainline builds, and this regression shows up, for my machines, in 2.6.39-rc1.
The last 2.6.38 was fine, and anything from that first .39 release candidate, inclusive, has shown the bug.
I'll try to get time to look through changelogs and patches for that version. (Though that's a big pile. And I'm no C programmer, nor do I have a git-bisect setup, or any experience with such things, alas... :-)
Ok, I've been going back through the mainline builds, and this regression shows up, for my machines, in 2.6.39-rc1.
The last 2.6.38 was fine, and anything from that first .39 release candidate, inclusive, has shown the bug.
I'll try to get time to look through changelogs and patches for that version. (Though that's a big pile. And I'm no C programmer, nor do I have a git-bisect setup, or any experience with such things, alas... :-)